{ "schema_version": 1, "status": "FINAL_EVIDENCE_VERIFIED", "study": { "title": "Strata v0.1.43 on RTX 5070 Ti 16GB / 64GB RAM", "brand": "Sonny's Labs", "run_date": "2026-10-12", "hardware": "NVIDIA RTX 5070 Ti 16GB", "installed_ram_gb": 64, "usable_host_ram_gib": 58.8, "lxc_ram_gib": 52, "cpu": "AMD Ryzen 5 5600G", "cpu_threads": 12, "pcie": "Gen3 x16", "storage": "WDC WDS100T2B0C-00PXH0 NVMe", "runtime": { "name": "Strata", "version": "0.1.43", "commit": "7695859aa13afb248bdb4d36f9b2e95707db079d", "llama_cpp_commit": "3cf03257f219afbe7334045ff7c6a06ac68c627d", "image": "sha256:f759e6167315dac77292d394a470240493dad78de6d23dd270eaa5c7ffc3e143", "driver": "615.71.09", "cuda_build": "13.0.0", "architecture": "sm_120" }, "cost": { "benchmark_cloud_rental_usd": 0, "quant_creation_usd": null, "idle_failed_cloud_rental_usd": 0, "electricity_usd": null, "total_project_cost_usd": null, "scope": "Local owned hardware. No cloud rentals or purchases; electricity, amortization, bandwidth and historic quant creation not metered. Zero rental is not zero total project cost." }, "context": { "sonny_allocated_tokens": 65536, "maximum_attempted_allocation_tokens": 131072, "exercised_prompt_tokens_by_model": { "strata-flash-next-iq2_xs": [ 57135, 57135, 57135 ], "strata-flash-next-iq3_xxs": [ 57135, 57135, 57135 ] } } }, "weighting": { "applied_percent": 90, "humaneval_plus_percent": 10, "category_weights": { "business": 20, "research": 20, "knowledge": 20, "agent_tools": 15, "practical_sysadmin": 5, "humaneval_plus": 10, "executive_content": 10 }, "role_weights": { "CFO / Analyst": { "business": 50, "research": 20, "knowledge": 15, "agent_tools": 10, "executive_content": 5 }, "Researcher / Scout": { "research": 40, "knowledge": 25, "agent_tools": 15, "executive_content": 10, "business": 10 }, "Hermes Coordinator": { "agent_tools": 30, "knowledge": 25, "research": 20, "executive_content": 15, "practical_sysadmin": 5, "humaneval_plus": 5 }, "Librarian / Knowledge": { "knowledge": 35, "research": 25, "agent_tools": 20, "executive_content": 15, "business": 5 }, "Forge / Coder": { "humaneval_plus": 50, "practical_sysadmin": 20, "agent_tools": 15, "knowledge": 10, "research": 5 } }, "formula": "sum(category_score * category_weight) / 100; category means are equally weighted tasks within each category" }, "metric_definitions": { "balanced_score": { "unit": "percent", "precision": 2, "direction": "higher" }, "humaneval_plus_percent": { "unit": "percent", "precision": 2, "direction": "higher" }, "decode_tps": { "unit": "tokens/s", "precision": 1, "direction": "higher" }, "prefill_tps": { "unit": "fresh tokens/s", "precision": 1, "direction": "higher" }, "ttft_s": { "unit": "seconds", "precision": 3, "direction": "lower" }, "latency_s": { "unit": "seconds", "precision": 2, "direction": "lower" } }, "models": [ { "id": "strata-flash-next-iq2_xs", "label": "ISTA-DASLab / GSQ-RCO \u00b7 IQ2_XS", "provider": "ISTA-DASLab / GSQ-RCO", "model": "Qwen3.8-Flash-Next", "quant": "IQ2_XS", "repository": "ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF", "revision": "ed59f92082b1e93c0e96d60a8b11aab089b52f09", "files": [ { "quant": "IQ2_XS", "repo": "ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF", "revision": "ed59f92082b1e93c0e96d60a8b11aab089b52f09", "filename": "IQ2_XS/Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00001-of-00002.gguf", "size_bytes": 39225954592, "sha256": "92cee27ae5bbadcd732416a0f7a7f0acc092399dbbe8f5a5efa707c2ec0a49d7" }, { "quant": "IQ2_XS", "repo": "ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF", "revision": "ed59f92082b1e93c0e96d60a8b11aab089b52f09", "filename": "IQ2_XS/Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00002-of-00002.gguf", "size_bytes": 28800138432, "sha256": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113" } ], "file_gib": 63.35423609614372, "mtp": true, "mtp_requested": 4, "mtp_min_p": 0.5, "reasoning_budget_tokens": 512, "balanced_score": 96.5, "humaneval_plus_percent": 93.29268292682927, "humaneval_base_percent": 96.95121951219512, "humaneval_evaluated": 164, "humaneval_plus_passed": 153, "role_scores": { "CFO / Analyst": 98.12, "Researcher / Scout": 96.72, "Hermes Coordinator": 95.11, "Librarian / Knowledge": 96.4, "Forge / Coder": 94.42 }, "category_scores": { "business": 100.0, "research": 96.8942, "knowledge": 100.0, "agent_tools": 90.625, "practical_sysadmin": 96.6667, "executive_content": 93.65 }, "practical_responses": 72, "practical_empty_finals": 0, "practical_length_finishes": 2, "dossier_evaluated_responses": 3, "dossier_valid_nonempty_responses": 3, "dossier_all_rows": [ { "repeat": 1, "status": "ok", "usage": { "prompt_tokens": 57135, "completion_tokens": 2534, "total_tokens": 59669, "prompt_tokens_details": { "cached_tokens": 0 } }, "timings": { "cache_n": 0, "prompt_n": 57135, "prompt_ms": 25690.5, "prompt_per_token_ms": 0.45, "prompt_per_second": 2224.0, "predicted_n": 2534, "predicted_ms": 26412.3, "predicted_per_token_ms": 10.498, "predicted_per_second": 95.3, "draft_n": 2008, "draft_n_accepted": 1564 }, "wall_time_s": 52.1555, "quality": { "overall_hermes": 99.4, "retrieval_facts": "14/14", "word_count": 985 }, "validated_quality": { "overall_hermes": 99.4, "retrieval_facts": "14/14", "word_count": 985 }, "empty_final": false, "finish_reason": "stop", "transport_failed": false }, { "repeat": 2, "status": "ok", "usage": { "prompt_tokens": 57135, "completion_tokens": 2093, "total_tokens": 59228, "prompt_tokens_details": { "cached_tokens": 57130 } }, "timings": { "cache_n": 57130, "prompt_n": 5, "prompt_ms": 219.4, "prompt_per_token_ms": 43.88, "prompt_per_second": 22.8, "predicted_n": 2093, "predicted_ms": 21400.1, "predicted_per_token_ms": 10.313, "predicted_per_second": 97.0, "draft_n": 1685, "draft_n_accepted": 1277 }, "wall_time_s": 21.674, "quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 782 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 782 }, "empty_final": false, "finish_reason": "stop", "transport_failed": false }, { "repeat": 3, "status": "ok", "usage": { "prompt_tokens": 57135, "completion_tokens": 2056, "total_tokens": 59191, "prompt_tokens_details": { "cached_tokens": 57130 } }, "timings": { "cache_n": 57130, "prompt_n": 5, "prompt_ms": 214.0, "prompt_per_token_ms": 42.8, "prompt_per_second": 23.4, "predicted_n": 2056, "predicted_ms": 21679.8, "predicted_per_token_ms": 10.633, "predicted_per_second": 94.1, "draft_n": 1649, "draft_n_accepted": 1238 }, "wall_time_s": 21.9465, "quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 765 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 765 }, "empty_final": false, "finish_reason": "stop", "transport_failed": false } ], "dossier_empty_finals": 0, "dossier_full_marker_recall_responses": 3, "dossier_successful_speed_eligible": true, "dossier_first_s": 52.1555, "dossier_warm_median_s": 21.81025, "fixed_median_s_original": 21.9465, "adaptive_seconds": null, "context": { "configured_sonny_tokens": 65536, "configured_max_attempted_tokens": 131072, "dossier_actual_prompt_tokens": [ 57135, 57135, 57135 ], "max_actual_recall_success_tokens": 124963, "recall_tests": [ { "test": "needles-128k-ctx131072", "source": "results/iq2_xs/needles-128k-ctx131072/results.json", "rows": [ { "length": "128k", "depth": 10, "prompt_tokens": 124962, "seconds": 59.2, "found": true, "word": "meadow-copper-504", "answer": "meadow-copper-504" }, { "length": "128k", "depth": 50, "prompt_tokens": 124962, "seconds": 59.4, "found": true, "word": "falcon-quartz-940", "answer": "falcon-quartz-940" }, { "length": "128k", "depth": 90, "prompt_tokens": 124963, "seconds": 52.2, "found": true, "word": "willow-cobalt-696", "answer": "willow-cobalt-696" } ], "cases": 3, "completed": 3, "found": 3, "empty_finals": 0, "transport_errors": 0, "actual_prompt_tokens": [ 124962, 124962, 124963 ] }, { "test": "needles-16k-ctx32768", "source": "results/iq2_xs/needles-16k-ctx32768/results.json", "rows": [ { "length": "16k", "depth": 10, "prompt_tokens": 15146, "seconds": 7.2, "found": true, "word": "meadow-copper-504", "answer": "meadow-copper-504" }, { "length": "16k", "depth": 50, "prompt_tokens": 15146, "seconds": 7.2, "found": true, "word": "falcon-quartz-940", "answer": "falcon-quartz-940" }, { "length": "16k", "depth": 90, "prompt_tokens": 15147, "seconds": 7.2, "found": true, "word": "willow-cobalt-696", "answer": "willow-cobalt-696" } ], "cases": 3, "completed": 3, "found": 3, "empty_finals": 0, "transport_errors": 0, "actual_prompt_tokens": [ 15146, 15146, 15147 ] }, { "test": "needles-32k-ctx32768", "source": "results/iq2_xs/needles-32k-ctx32768/results.json", "rows": [ { "length": "32k", "depth": 10, "error": "HTTP Error 400: Bad Request" }, { "length": "32k", "depth": 50, "error": "HTTP Error 400: Bad Request" }, { "length": "32k", "depth": 90, "error": "HTTP Error 400: Bad Request" } ], "cases": 3, "completed": 0, "found": 0, "empty_finals": 0, "transport_errors": 3, "actual_prompt_tokens": [ null, null, null ] }, { "test": "needles-32k-ctx65536", "source": "results/iq2_xs/needles-32k-ctx65536/results.json", "rows": [ { "length": "32k", "depth": 10, "prompt_tokens": 32804, "seconds": 15.9, "found": true, "word": "meadow-copper-504", "answer": "meadow-copper-504" }, { "length": "32k", "depth": 50, "prompt_tokens": 32804, "seconds": 15.2, "found": true, "word": "falcon-quartz-940", "answer": "falcon-quartz-940" }, { "length": "32k", "depth": 90, "prompt_tokens": 32805, "seconds": 15.2, "found": true, "word": "willow-cobalt-696", "answer": "willow-cobalt-696" } ], "cases": 3, "completed": 3, "found": 3, "empty_finals": 0, "transport_errors": 0, "actual_prompt_tokens": [ 32804, 32804, 32805 ] }, { "test": "needles-64k-ctx131072", "source": "results/iq2_xs/needles-64k-ctx131072/results.json", "rows": [ { "length": "64k", "depth": 10, "prompt_tokens": 64925, "seconds": 30.3, "found": true, "word": "meadow-copper-504", "answer": "meadow-copper-504" }, { "length": "64k", "depth": 50, "prompt_tokens": 64925, "seconds": 29.5, "found": true, "word": "falcon-quartz-940", "answer": "falcon-quartz-940" }, { "length": "64k", "depth": 90, "prompt_tokens": 64926, "seconds": 22.4, "found": true, "word": "willow-cobalt-696", "answer": "willow-cobalt-696" } ], "cases": 3, "completed": 3, "found": 3, "empty_finals": 0, "transport_errors": 0, "actual_prompt_tokens": [ 64925, 64925, 64926 ] }, { "test": "needles-8k-ctx32768", "source": "results/iq2_xs/needles-8k-ctx32768/results.json", "rows": [ { "length": "8k", "depth": 10, "prompt_tokens": 6920, "seconds": 4.4, "found": true, "word": "meadow-copper-504", "answer": "meadow-copper-504" }, { "length": "8k", "depth": 50, "prompt_tokens": 6920, "seconds": 3.7, "found": true, "word": "falcon-quartz-940", "answer": "falcon-quartz-940" }, { "length": "8k", "depth": 90, "prompt_tokens": 6921, "seconds": 3.7, "found": true, "word": "willow-cobalt-696", "answer": "willow-cobalt-696" } ], "cases": 3, "completed": 3, "found": 3, "empty_finals": 0, "transport_errors": 0, "actual_prompt_tokens": [ 6920, 6920, 6921 ] } ], "note": "Successful recall is limited to the exact tested inputs/depths. Context configuration is not proof of answer quality." }, "official_standard": { "mode": "standard-32768", "source": "official-benchmarks/iq2_xs/standard-32768/2026-10-12-community-rtx-5070-ti-qwen3-8-flash-next-iq2-xs/results.json", "schema": 1, "params": { "runs": 3, "think": false, "clients": 0, "long": false, "started_server": false, "command": "python tools/strata_bench.py" }, "elapsed_s": 100.38631105422974, "machine": { "gpu": [ "NVIDIA GeForce RTX 5070 Ti" ], "gpu_vram_mib": 16303, "gpu_driver": "615.71.09", "cuda": null, "cpu": "AMD Ryzen 5 5600G with Radeon Graphics", "cpu_cores": 6, "cpu_threads": 12, "ram_gib": 58.8, "os": "Linux 7.0.14-17-pve", "python": "3.12.3", "strata_version": "0.1.43", "strata_commit": "7695859a", "engine_version": "0.1.43", "model": "qwen3.8-flash-next-iq2_xs", "quant": "IQ2_XS", "context": 32768, "parallel_slots": 1, "config": { "exe": "strata", "args": [ "--pack", "iq2_xs", "--native", "Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00001-of-00002.gguf", "--ple-gguf", "Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00002-of-00002.gguf", "--expert-profile", "expert-profile.bin", "--expert-cache", "auto", "--prefill", "auto", "--spec", "4", "--spec-min-p", "0.5", "--mtp", "rt", "--max-context", "32768", "--kv", "int8", "--resident-experts", "--vram-reserve-mib", "1024" ], "model_name": "qwen3.8-flash-next-iq2_xs", "lib_dirs": [ "/usr/local/cuda/lib64" ], "port": 8080, "gpu": 0, "gpus_asked": true, "host": "127.0.0.1", "open_browser": false, "parallel": 1 }, "launch": "strata --pack iq2_xs --native Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00001-of-00002.gguf --ple-gguf Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00002-of-00002.gguf --expert-profile expert-profile.bin --expert-cache auto --prefill auto --spec 4 --spec-min-p 0.5 --mtp rt --max-context 32768 --kv int8 --resident-experts --vram-reserve-mib 1024", "engine_info": { "model": "qwen3.8-flash-next-iq2_xs", "max_context": 32768, "images": false, "version": "0.1.43", "logit_bias": 1, "context": 32768, "kv": "int8", "kv_resident": 0, "expert_slots": 7073, "expert_cache_mib": 9700, "expert_slots_primary": 7073, "expert_cache_primary_mib": 9700, "spec": 6, "mtp_max": 4, "lookup": 3, "vram_free_mib": 527, "cvec": 0, "arena_mib": 27975, "pool_workers": 5, "pcie_frac": "0.37", "spec_min_p": "0.50", "conversation_cache_mib": 0, "conversation_cache_slots": 4, "conversation_cache_min_free_mib": 2560, "conversation_cache_min_tokens": 0, "tail_role_token": -1, "vram_elastic": 0, "engine": "0.1.43" }, "vram_after": { "elastic": false } }, "metrics": { "chat": { "requests": [ { "ok": true, "err": null, "ttft_s": 0.415, "total_s": 1.454, "prompt_tokens": 51, "reused_tokens": 0, "read_tokens": 51, "generated_tokens": 100, "prompt_tps": 126.4, "prompt_tps_source": "engine", "decode_tps": 95.4, "decode_tps_source": "engine", "prompt_ms": 403.6, "decode_ms": 1048.7, "drafts_offered": 88, "drafts_accepted": 70, "run": 0, "prompt": 0 }, { "ok": true, "err": null, "ttft_s": 0.391, "total_s": 1.791, "prompt_tokens": 43, "reused_tokens": 0, "read_tokens": 43, "generated_tokens": 100, "prompt_tps": 113.8, "prompt_tps_source": "engine", "decode_tps": 70.9, "decode_tps_source": "engine", "prompt_ms": 377.9, "decode_ms": 1410.8, "drafts_offered": 88, "drafts_accepted": 52, "run": 0, "prompt": 1 }, { "ok": true, "err": null, "ttft_s": 0.473, "total_s": 2.041, "prompt_tokens": 46, "reused_tokens": 0, "read_tokens": 46, "generated_tokens": 100, "prompt_tps": 100.0, "prompt_tps_source": "engine", "decode_tps": 63.4, "decode_tps_source": "engine", "prompt_ms": 459.8, "decode_ms": 1578.5, "drafts_offered": 96, "drafts_accepted": 54, "run": 0, "prompt": 2 }, { "ok": true, "err": null, "ttft_s": 0.391, "total_s": 1.488, "prompt_tokens": 51, "reused_tokens": 0, "read_tokens": 51, "generated_tokens": 100, "prompt_tps": 134.6, "prompt_tps_source": "engine", "decode_tps": 90.3, "decode_tps_source": "engine", "prompt_ms": 378.8, "decode_ms": 1107.1, "drafts_offered": 87, "drafts_accepted": 69, "run": 1, "prompt": 0 }, { "ok": true, "err": null, "ttft_s": 0.345, "total_s": 1.542, "prompt_tokens": 43, "reused_tokens": 0, "read_tokens": 43, "generated_tokens": 100, "prompt_tps": 129.4, "prompt_tps_source": "engine", "decode_tps": 82.8, "decode_tps_source": "engine", "prompt_ms": 332.2, "decode_ms": 1207.7, "drafts_offered": 94, "drafts_accepted": 53, "run": 1, "prompt": 1 }, { "ok": true, "err": null, "ttft_s": 0.368, "total_s": 1.582, "prompt_tokens": 46, "reused_tokens": 0, "read_tokens": 46, "generated_tokens": 100, "prompt_tps": 129.5, "prompt_tps_source": "engine", "decode_tps": 81.7, "decode_tps_source": "engine", "prompt_ms": 355.2, "decode_ms": 1224.3, "drafts_offered": 94, "drafts_accepted": 56, "run": 1, "prompt": 2 }, { "ok": true, "err": null, "ttft_s": 0.389, "total_s": 1.524, "prompt_tokens": 51, "reused_tokens": 0, "read_tokens": 51, "generated_tokens": 100, "prompt_tps": 135.4, "prompt_tps_source": "engine", "decode_tps": 87.3, "decode_tps_source": "engine", "prompt_ms": 376.8, "decode_ms": 1145.1, "drafts_offered": 85, "drafts_accepted": 69, "run": 2, "prompt": 0 }, { "ok": true, "err": null, "ttft_s": 0.334, "total_s": 1.544, "prompt_tokens": 43, "reused_tokens": 0, "read_tokens": 43, "generated_tokens": 100, "prompt_tps": 133.8, "prompt_tps_source": "engine", "decode_tps": 81.9, "decode_tps_source": "engine", "prompt_ms": 321.4, "decode_ms": 1220.9, "drafts_offered": 87, "drafts_accepted": 51, "run": 2, "prompt": 1 }, { "ok": true, "err": null, "ttft_s": 0.355, "total_s": 1.583, "prompt_tokens": 46, "reused_tokens": 0, "read_tokens": 46, "generated_tokens": 100, "prompt_tps": 134.5, "prompt_tps_source": "engine", "decode_tps": 80.7, "decode_tps_source": "engine", "prompt_ms": 341.9, "decode_ms": 1238.6, "drafts_offered": 89, "drafts_accepted": 53, "run": 2, "prompt": 2 } ], "decode_tps": { "median": 81.9, "min": 63.4, "max": 95.4, "n": 9 }, "ttft_s": { "median": 0.389, "min": 0.334, "max": 0.473, "n": 9 }, "generated_tokens": { "median": 100, "min": 100, "max": 100, "n": 9 }, "prompt_tokens": { "median": 46, "min": 43, "max": 51, "n": 9 } }, "prompts": { "4K": { "target_tokens": 4096, "requests": [ { "ok": true, "err": null, "ttft_s": 2.868, "total_s": 3.215, "prompt_tokens": 4019, "reused_tokens": 0, "read_tokens": 4019, "generated_tokens": 24, "prompt_tps": 1408.8, "prompt_tps_source": "engine", "decode_tps": 67.1, "decode_tps_source": "engine", "prompt_ms": 2852.8, "decode_ms": 357.8, "drafts_offered": 22, "drafts_accepted": 10, "run": 0 }, { "ok": true, "err": null, "ttft_s": 2.769, "total_s": 3.145, "prompt_tokens": 4019, "reused_tokens": 0, "read_tokens": 4019, "generated_tokens": 24, "prompt_tps": 1459.5, "prompt_tps_source": "engine", "decode_tps": 62.1, "decode_tps_source": "engine", "prompt_ms": 2753.6, "decode_ms": 386.7, "drafts_offered": 25, "drafts_accepted": 8, "run": 1 }, { "ok": true, "err": null, "ttft_s": 2.762, "total_s": 3.116, "prompt_tokens": 4019, "reused_tokens": 0, "read_tokens": 4019, "generated_tokens": 24, "prompt_tps": 1462.7, "prompt_tps_source": "engine", "decode_tps": 65.9, "decode_tps_source": "engine", "prompt_ms": 2747.6, "decode_ms": 364.3, "drafts_offered": 24, "drafts_accepted": 9, "run": 2 } ], "prompt_tps": { "median": 1459.5, "min": 1408.8, "max": 1462.7, "n": 3 }, "ttft_s": { "median": 2.769, "min": 2.762, "max": 2.868, "n": 3 }, "prompt_tokens": { "median": 4019, "min": 4019, "max": 4019, "n": 3 }, "reused_tokens": { "median": 0, "min": 0, "max": 0, "n": 3 } }, "16K": { "target_tokens": 16384, "requests": [ { "ok": true, "err": null, "ttft_s": 7.491, "total_s": 7.741, "prompt_tokens": 16058, "reused_tokens": 0, "read_tokens": 16058, "generated_tokens": 24, "prompt_tps": 2149.9, "prompt_tps_source": "engine", "decode_tps": 92.1, "decode_tps_source": "engine", "prompt_ms": 7469.3, "decode_ms": 260.7, "drafts_offered": 23, "drafts_accepted": 10, "run": 0 }, { "ok": true, "err": null, "ttft_s": 7.398, "total_s": 7.638, "prompt_tokens": 16058, "reused_tokens": 0, "read_tokens": 16058, "generated_tokens": 24, "prompt_tps": 2176.8, "prompt_tps_source": "engine", "decode_tps": 95.9, "decode_tps_source": "engine", "prompt_ms": 7376.8, "decode_ms": 250.3, "drafts_offered": 22, "drafts_accepted": 8, "run": 1 }, { "ok": true, "err": null, "ttft_s": 7.395, "total_s": 7.621, "prompt_tokens": 16058, "reused_tokens": 0, "read_tokens": 16058, "generated_tokens": 24, "prompt_tps": 2177.5, "prompt_tps_source": "engine", "decode_tps": 102.0, "decode_tps_source": "engine", "prompt_ms": 7374.5, "decode_ms": 235.3, "drafts_offered": 22, "drafts_accepted": 10, "run": 2 } ], "prompt_tps": { "median": 2176.8, "min": 2149.9, "max": 2177.5, "n": 3 }, "ttft_s": { "median": 7.398, "min": 7.395, "max": 7.491, "n": 3 }, "prompt_tokens": { "median": 16058, "min": 16058, "max": 16058, "n": 3 }, "reused_tokens": { "median": 0, "min": 0, "max": 0, "n": 3 } } }, "agent": { "sessions": [ [ { "ok": true, "err": null, "ttft_s": 1.201, "total_s": 1.614, "prompt_tokens": 609, "reused_tokens": 0, "read_tokens": 609, "generated_tokens": 28, "prompt_tps": 514.2, "prompt_tps_source": "engine", "decode_tps": 65.7, "decode_tps_source": "engine", "prompt_ms": 1184.4, "decode_ms": 426.4, "drafts_offered": 35, "drafts_accepted": 17, "run": 0, "turn": 0 }, { "ok": true, "err": null, "ttft_s": 2.767, "total_s": 3.65, "prompt_tokens": 3719, "reused_tokens": 638, "read_tokens": 3081, "generated_tokens": 64, "prompt_tps": 1121.4, "prompt_tps_source": "engine", "decode_tps": 71.6, "decode_tps_source": "engine", "prompt_ms": 2747.4, "decode_ms": 894.3, "drafts_offered": 63, "drafts_accepted": 39, "run": 0, "turn": 1 }, { "ok": true, "err": null, "ttft_s": 2.306, "total_s": 3.047, "prompt_tokens": 5252, "reused_tokens": 3782, "read_tokens": 1470, "generated_tokens": 64, "prompt_tps": 642.6, "prompt_tps_source": "engine", "decode_tps": 85.1, "decode_tps_source": "engine", "prompt_ms": 2287.7, "decode_ms": 751.7, "drafts_offered": 66, "drafts_accepted": 39, "run": 0, "turn": 2 }, { "ok": true, "err": null, "ttft_s": 2.808, "total_s": 3.239, "prompt_tokens": 8696, "reused_tokens": 5315, "read_tokens": 3381, "generated_tokens": 40, "prompt_tps": 1214.6, "prompt_tps_source": "engine", "decode_tps": 90.5, "decode_tps_source": "engine", "prompt_ms": 2783.7, "decode_ms": 442.2, "drafts_offered": 37, "drafts_accepted": 23, "run": 0, "turn": 3 }, { "ok": true, "err": null, "ttft_s": 2.65, "total_s": 2.857, "prompt_tokens": 11595, "reused_tokens": 8735, "read_tokens": 2860, "generated_tokens": 18, "prompt_tps": 1088.2, "prompt_tps_source": "engine", "decode_tps": 82.9, "decode_tps_source": "engine", "prompt_ms": 2628.2, "decode_ms": 217.2, "drafts_offered": 21, "drafts_accepted": 10, "run": 0, "turn": 4 }, { "ok": true, "err": null, "ttft_s": 2.586, "total_s": 3.038, "prompt_tokens": 14458, "reused_tokens": 11612, "read_tokens": 2846, "generated_tokens": 39, "prompt_tps": 1110.0, "prompt_tps_source": "engine", "decode_tps": 84.3, "decode_tps_source": "engine", "prompt_ms": 2564.0, "decode_ms": 462.4, "drafts_offered": 44, "drafts_accepted": 21, "run": 0, "turn": 5 } ], [ { "ok": true, "err": null, "ttft_s": 1.242, "total_s": 1.36, "prompt_tokens": 609, "reused_tokens": 0, "read_tokens": 609, "generated_tokens": 10, "prompt_tps": 495.7, "prompt_tps_source": "engine", "decode_tps": 77.6, "decode_tps_source": "engine", "prompt_ms": 1228.6, "decode_ms": 128.8, "drafts_offered": 12, "drafts_accepted": 7, "run": 1, "turn": 0 }, { "ok": true, "err": null, "ttft_s": 2.513, "total_s": 3.216, "prompt_tokens": 3701, "reused_tokens": 620, "read_tokens": 3081, "generated_tokens": 64, "prompt_tps": 1233.0, "prompt_tps_source": "engine", "decode_tps": 89.8, "decode_tps_source": "engine", "prompt_ms": 2498.8, "decode_ms": 712.8, "drafts_offered": 67, "drafts_accepted": 38, "run": 1, "turn": 1 }, { "ok": true, "err": null, "ttft_s": 2.211, "total_s": 2.862, "prompt_tokens": 5234, "reused_tokens": 3694, "read_tokens": 1540, "generated_tokens": 64, "prompt_tps": 701.4, "prompt_tps_source": "engine", "decode_tps": 96.9, "decode_tps_source": "engine", "prompt_ms": 2195.6, "decode_ms": 660.6, "drafts_offered": 67, "drafts_accepted": 41, "run": 1, "turn": 2 }, { "ok": true, "err": null, "ttft_s": 2.641, "total_s": 3.012, "prompt_tokens": 8678, "reused_tokens": 5227, "read_tokens": 3451, "generated_tokens": 42, "prompt_tps": 1315.5, "prompt_tps_source": "engine", "decode_tps": 110.3, "decode_tps_source": "engine", "prompt_ms": 2623.3, "decode_ms": 380.7, "drafts_offered": 39, "drafts_accepted": 27, "run": 1, "turn": 3 }, { "ok": true, "err": null, "ttft_s": 2.493, "total_s": 2.859, "prompt_tokens": 11579, "reused_tokens": 8719, "read_tokens": 2860, "generated_tokens": 39, "prompt_tps": 1156.1, "prompt_tps_source": "engine", "decode_tps": 103.6, "decode_tps_source": "engine", "prompt_ms": 2473.9, "decode_ms": 376.3, "drafts_offered": 37, "drafts_accepted": 24, "run": 1, "turn": 4 }, { "ok": true, "err": null, "ttft_s": 2.491, "total_s": 2.972, "prompt_tokens": 14463, "reused_tokens": 11617, "read_tokens": 2846, "generated_tokens": 45, "prompt_tps": 1151.9, "prompt_tps_source": "engine", "decode_tps": 91.6, "decode_tps_source": "engine", "prompt_ms": 2470.7, "decode_ms": 491.1, "drafts_offered": 48, "drafts_accepted": 25, "run": 1, "turn": 5 } ], [ { "ok": true, "err": null, "ttft_s": 1.232, "total_s": 1.476, "prompt_tokens": 609, "reused_tokens": 0, "read_tokens": 609, "generated_tokens": 24, "prompt_tps": 500.1, "prompt_tps_source": "engine", "decode_tps": 94.0, "decode_tps_source": "engine", "prompt_ms": 1217.8, "decode_ms": 255.4, "drafts_offered": 21, "drafts_accepted": 15, "run": 2, "turn": 0 }, { "ok": true, "err": null, "ttft_s": 2.519, "total_s": 3.156, "prompt_tokens": 3715, "reused_tokens": 632, "read_tokens": 3083, "generated_tokens": 64, "prompt_tps": 1230.9, "prompt_tps_source": "engine", "decode_tps": 99.0, "decode_tps_source": "engine", "prompt_ms": 2504.7, "decode_ms": 646.6, "drafts_offered": 64, "drafts_accepted": 39, "run": 2, "turn": 1 }, { "ok": true, "err": null, "ttft_s": 2.198, "total_s": 2.846, "prompt_tokens": 5248, "reused_tokens": 3778, "read_tokens": 1470, "generated_tokens": 64, "prompt_tps": 673.6, "prompt_tps_source": "engine", "decode_tps": 97.3, "decode_tps_source": "engine", "prompt_ms": 2182.3, "decode_ms": 657.7, "drafts_offered": 66, "drafts_accepted": 41, "run": 2, "turn": 2 }, { "ok": true, "err": null, "ttft_s": 2.638, "total_s": 3.133, "prompt_tokens": 8692, "reused_tokens": 5241, "read_tokens": 3451, "generated_tokens": 39, "prompt_tps": 1316.8, "prompt_tps_source": "engine", "decode_tps": 77.2, "decode_tps_source": "engine", "prompt_ms": 2620.7, "decode_ms": 505.0, "drafts_offered": 44, "drafts_accepted": 17, "run": 2, "turn": 3 }, { "ok": true, "err": null, "ttft_s": 2.484, "total_s": 2.715, "prompt_tokens": 11590, "reused_tokens": 8730, "read_tokens": 2860, "generated_tokens": 28, "prompt_tps": 1160.3, "prompt_tps_source": "engine", "decode_tps": 116.3, "decode_tps_source": "engine", "prompt_ms": 2464.9, "decode_ms": 240.8, "drafts_offered": 26, "drafts_accepted": 18, "run": 2, "turn": 4 }, { "ok": true, "err": null, "ttft_s": 2.481, "total_s": 2.969, "prompt_tokens": 14463, "reused_tokens": 11617, "read_tokens": 2846, "generated_tokens": 39, "prompt_tps": 1157.4, "prompt_tps_source": "engine", "decode_tps": 78.6, "decode_tps_source": "engine", "prompt_ms": 2458.9, "decode_ms": 496.2, "drafts_offered": 45, "drafts_accepted": 19, "run": 2, "turn": 5 } ] ], "hit_pct": { "median": 68.633, "min": 68.439, "max": 68.806, "n": 3 }, "reread_tokens": { "median": 75, "min": 4, "max": 144, "n": 3 }, "new_tokens": { "median": 13634, "min": 13634, "max": 13635, "n": 3 }, "turn_ttft_s": { "median": 2.513, "min": 2.198, "max": 2.808, "n": 15 }, "last_turn_prompt_tokens": { "median": 14463, "min": 14458, "max": 14463, "n": 3 }, "turn_decode_tps": { "median": 90.5, "min": 71.6, "max": 116.3, "n": 15 }, "first_turn_ttft_s": { "median": 1.232, "min": 1.201, "max": 1.242, "n": 3 } }, "clients": {}, "skipped": {}, "failures": [] } }, "official_optional": { "mode": "optional-65536", "source": "official-benchmarks/iq2_xs/optional-65536/2026-10-12-community-rtx-5070-ti-qwen3-8-flash-next-iq2-xs/results.json", "schema": 1, "params": { "runs": 3, "think": false, "clients": 2, "long": true, "started_server": false, "command": "python tools/strata_bench.py --clients 2 --long" }, "elapsed_s": 161.40699243545532, "machine": { "gpu": [ "NVIDIA GeForce RTX 5070 Ti" ], "gpu_vram_mib": 16303, "gpu_driver": "615.71.09", "cuda": null, "cpu": "AMD Ryzen 5 5600G with Radeon Graphics", "cpu_cores": 6, "cpu_threads": 12, "ram_gib": 58.8, "os": "Linux 7.0.14-17-pve", "python": "3.12.3", "strata_version": "0.1.43", "strata_commit": "7695859a", "engine_version": "0.1.43", "model": "qwen3.8-flash-next-iq2_xs", "quant": "IQ2_XS", "context": 65536, "parallel_slots": 1, "config": { "exe": "strata", "args": [ "--pack", "iq2_xs", "--native", "Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00001-of-00002.gguf", "--ple-gguf", "Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00002-of-00002.gguf", "--expert-profile", "expert-profile.bin", "--expert-cache", "auto", "--prefill", "auto", "--spec", "4", "--spec-min-p", "0.5", "--mtp", "rt", "--max-context", "65536", "--kv", "int8", "--resident-experts", "--vram-reserve-mib", "1024", "--kv-resident", "32768" ], "model_name": "qwen3.8-flash-next-iq2_xs", "lib_dirs": [ "/usr/local/cuda/lib64" ], "port": 8080, "gpu": 0, "gpus_asked": true, "host": "127.0.0.1", "open_browser": false, "parallel": 1 }, "launch": "strata --pack iq2_xs --native Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00001-of-00002.gguf --ple-gguf Qwen3.8-Flash-Next-GSQ-RCO-IQ2_XS-00002-of-00002.gguf --expert-profile expert-profile.bin --expert-cache auto --prefill auto --spec 4 --spec-min-p 0.5 --mtp rt --max-context 65536 --kv int8 --resident-experts --vram-reserve-mib 1024 --kv-resident 32768", "engine_info": { "model": "qwen3.8-flash-next-iq2_xs", "max_context": 65536, "images": false, "version": "0.1.43", "logit_bias": 1, "context": 65536, "kv": "int8", "kv_resident": 32768, "expert_slots": 7031, "expert_cache_mib": 9644, "expert_slots_primary": 7031, "expert_cache_primary_mib": 9644, "spec": 6, "mtp_max": 4, "lookup": 3, "vram_free_mib": 529, "cvec": 0, "arena_mib": 28097, "pool_workers": 5, "pcie_frac": "0.37", "spec_min_p": "0.50", "conversation_cache_mib": 0, "conversation_cache_slots": 4, "conversation_cache_min_free_mib": 2560, "conversation_cache_min_tokens": 0, "tail_role_token": -1, "vram_elastic": 0, "engine": "0.1.43" }, "vram_after": { "elastic": false } }, "metrics": { "chat": { "requests": [ { "ok": true, "err": null, "ttft_s": 0.439, "total_s": 1.622, "prompt_tokens": 51, "reused_tokens": 0, "read_tokens": 51, "generated_tokens": 100, "prompt_tps": 119.6, "prompt_tps_source": "engine", "decode_tps": 83.8, "decode_tps_source": "engine", "prompt_ms": 426.4, "decode_ms": 1193.5, "drafts_offered": 95, "drafts_accepted": 68, "run": 0, "prompt": 0 }, { "ok": true, "err": null, "ttft_s": 0.432, "total_s": 1.8, "prompt_tokens": 43, "reused_tokens": 0, "read_tokens": 43, "generated_tokens": 100, "prompt_tps": 103.1, "prompt_tps_source": "engine", "decode_tps": 72.4, "decode_tps_source": "engine", "prompt_ms": 417.2, "decode_ms": 1380.7, "drafts_offered": 91, "drafts_accepted": 52, "run": 0, "prompt": 1 }, { "ok": true, "err": null, "ttft_s": 0.438, "total_s": 1.929, "prompt_tokens": 46, "reused_tokens": 0, "read_tokens": 46, "generated_tokens": 100, "prompt_tps": 108.6, "prompt_tps_source": "engine", "decode_tps": 66.5, "decode_tps_source": "engine", "prompt_ms": 423.6, "decode_ms": 1502.9, "drafts_offered": 97, "drafts_accepted": 55, "run": 0, "prompt": 2 }, { "ok": true, "err": null, "ttft_s": 0.417, "total_s": 1.521, "prompt_tokens": 51, "reused_tokens": 0, "read_tokens": 51, "generated_tokens": 100, "prompt_tps": 126.1, "prompt_tps_source": "engine", "decode_tps": 89.8, "decode_tps_source": "engine", "prompt_ms": 404.6, "decode_ms": 1113.7, "drafts_offered": 87, "drafts_accepted": 69, "run": 1, "prompt": 0 }, { "ok": true, "err": null, "ttft_s": 0.359, "total_s": 1.605, "prompt_tokens": 43, "reused_tokens": 0, "read_tokens": 43, "generated_tokens": 100, "prompt_tps": 124.3, "prompt_tps_source": "engine", "decode_tps": 79.5, "decode_tps_source": "engine", "prompt_ms": 345.8, "decode_ms": 1257.2, "drafts_offered": 94, "drafts_accepted": 48, "run": 1, "prompt": 1 }, { "ok": true, "err": null, "ttft_s": 0.368, "total_s": 1.589, "prompt_tokens": 46, "reused_tokens": 0, "read_tokens": 46, "generated_tokens": 100, "prompt_tps": 129.8, "prompt_tps_source": "engine", "decode_tps": 81.2, "decode_tps_source": "engine", "prompt_ms": 354.5, "decode_ms": 1232.1, "drafts_offered": 90, "drafts_accepted": 54, "run": 1, "prompt": 2 }, { "ok": true, "err": null, "ttft_s": 0.425, "total_s": 1.573, "prompt_tokens": 51, "reused_tokens": 0, "read_tokens": 51, "generated_tokens": 100, "prompt_tps": 124.0, "prompt_tps_source": "engine", "decode_tps": 86.3, "decode_tps_source": "engine", "prompt_ms": 411.2, "decode_ms": 1159.4, "drafts_offered": 85, "drafts_accepted": 70, "run": 2, "prompt": 0 }, { "ok": true, "err": null, "ttft_s": 0.344, "total_s": 1.568, "prompt_tokens": 43, "reused_tokens": 0, "read_tokens": 43, "generated_tokens": 100, "prompt_tps": 130.1, "prompt_tps_source": "engine", "decode_tps": 81.0, "decode_tps_source": "engine", "prompt_ms": 330.4, "decode_ms": 1235.3, "drafts_offered": 99, "drafts_accepted": 51, "run": 2, "prompt": 1 }, { "ok": true, "err": null, "ttft_s": 0.361, "total_s": 1.587, "prompt_tokens": 46, "reused_tokens": 0, "read_tokens": 46, "generated_tokens": 100, "prompt_tps": 132.2, "prompt_tps_source": "engine", "decode_tps": 80.8, "decode_tps_source": "engine", "prompt_ms": 347.9, "decode_ms": 1236.9, "drafts_offered": 93, "drafts_accepted": 55, "run": 2, "prompt": 2 } ], "decode_tps": { "median": 81.0, "min": 66.5, "max": 89.8, "n": 9 }, "ttft_s": { "median": 0.417, "min": 0.344, "max": 0.439, "n": 9 }, "generated_tokens": { "median": 100, "min": 100, "max": 100, "n": 9 }, "prompt_tokens": { "median": 46, "min": 43, "max": 51, "n": 9 } }, "prompts": { "4K": { "target_tokens": 4096, "requests": [ { "ok": true, "err": null, "ttft_s": 2.889, "total_s": 3.23, "prompt_tokens": 4019, "reused_tokens": 0, "read_tokens": 4019, "generated_tokens": 24, "prompt_tps": 1399.0, "prompt_tps_source": "engine", "decode_tps": 68.0, "decode_tps_source": "engine", "prompt_ms": 2872.7, "decode_ms": 352.8, "drafts_offered": 19, "drafts_accepted": 10, "run": 0 }, { "ok": true, "err": null, "ttft_s": 2.774, "total_s": 3.134, "prompt_tokens": 4019, "reused_tokens": 0, "read_tokens": 4019, "generated_tokens": 24, "prompt_tps": 1456.8, "prompt_tps_source": "engine", "decode_tps": 64.8, "decode_tps_source": "engine", "prompt_ms": 2758.8, "decode_ms": 370.6, "drafts_offered": 24, "drafts_accepted": 9, "run": 1 }, { "ok": true, "err": null, "ttft_s": 2.783, "total_s": 3.131, "prompt_tokens": 4019, "reused_tokens": 0, "read_tokens": 4019, "generated_tokens": 24, "prompt_tps": 1451.8, "prompt_tps_source": "engine", "decode_tps": 67.0, "decode_tps_source": "engine", "prompt_ms": 2768.2, "decode_ms": 358.4, "drafts_offered": 25, "drafts_accepted": 9, "run": 2 } ], "prompt_tps": { "median": 1451.8, "min": 1399.0, "max": 1456.8, "n": 3 }, "ttft_s": { "median": 2.783, "min": 2.774, "max": 2.889, "n": 3 }, "prompt_tokens": { "median": 4019, "min": 4019, "max": 4019, "n": 3 }, "reused_tokens": { "median": 0, "min": 0, "max": 0, "n": 3 } }, "16K": { "target_tokens": 16384, "requests": [ { "ok": true, "err": null, "ttft_s": 7.531, "total_s": 7.813, "prompt_tokens": 16058, "reused_tokens": 0, "read_tokens": 16058, "generated_tokens": 24, "prompt_tps": 2138.5, "prompt_tps_source": "engine", "decode_tps": 82.1, "decode_tps_source": "engine", "prompt_ms": 7509.0, "decode_ms": 292.5, "drafts_offered": 20, "drafts_accepted": 12, "run": 0 }, { "ok": true, "err": null, "ttft_s": 7.479, "total_s": 7.719, "prompt_tokens": 16058, "reused_tokens": 0, "read_tokens": 16058, "generated_tokens": 24, "prompt_tps": 2153.8, "prompt_tps_source": "engine", "decode_tps": 95.8, "decode_tps_source": "engine", "prompt_ms": 7455.6, "decode_ms": 250.5, "drafts_offered": 21, "drafts_accepted": 11, "run": 1 }, { "ok": true, "err": null, "ttft_s": 7.465, "total_s": 7.687, "prompt_tokens": 16058, "reused_tokens": 0, "read_tokens": 16058, "generated_tokens": 24, "prompt_tps": 2157.2, "prompt_tps_source": "engine", "decode_tps": 103.1, "decode_tps_source": "engine", "prompt_ms": 7443.8, "decode_ms": 232.8, "drafts_offered": 24, "drafts_accepted": 10, "run": 2 } ], "prompt_tps": { "median": 2153.8, "min": 2138.5, "max": 2157.2, "n": 3 }, "ttft_s": { "median": 7.479, "min": 7.465, "max": 7.531, "n": 3 }, "prompt_tokens": { "median": 16058, "min": 16058, "max": 16058, "n": 3 }, "reused_tokens": { "median": 0, "min": 0, "max": 0, "n": 3 } }, "32K": { "target_tokens": 32768, "requests": [ { "ok": true, "err": null, "ttft_s": 14.713, "total_s": 15.131, "prompt_tokens": 32049, "reused_tokens": 0, "read_tokens": 32049, "generated_tokens": 24, "prompt_tps": 2183.4, "prompt_tps_source": "engine", "decode_tps": 55.6, "decode_tps_source": "engine", "prompt_ms": 14678.2, "decode_ms": 431.5, "drafts_offered": 20, "drafts_accepted": 8, "run": 0 }, { "ok": true, "err": null, "ttft_s": 14.706, "total_s": 15.043, "prompt_tokens": 32049, "reused_tokens": 0, "read_tokens": 32049, "generated_tokens": 24, "prompt_tps": 2184.3, "prompt_tps_source": "engine", "decode_tps": 68.4, "decode_tps_source": "engine", "prompt_ms": 14672.3, "decode_ms": 350.7, "drafts_offered": 18, "drafts_accepted": 8, "run": 1 }, { "ok": true, "err": null, "ttft_s": 14.711, "total_s": 15.023, "prompt_tokens": 32049, "reused_tokens": 0, "read_tokens": 32049, "generated_tokens": 24, "prompt_tps": 2183.3, "prompt_tps_source": "engine", "decode_tps": 74.1, "decode_tps_source": "engine", "prompt_ms": 14679.1, "decode_ms": 323.8, "drafts_offered": 19, "drafts_accepted": 12, "run": 2 } ], "prompt_tps": { "median": 2183.4, "min": 2183.3, "max": 2184.3, "n": 3 }, "ttft_s": { "median": 14.711, "min": 14.706, "max": 14.713, "n": 3 }, "prompt_tokens": { "median": 32049, "min": 32049, "max": 32049, "n": 3 }, "reused_tokens": { "median": 0, "min": 0, "max": 0, "n": 3 } } }, "agent": { "sessions": [ [ { "ok": true, "err": null, "ttft_s": 1.237, "total_s": 1.428, "prompt_tokens": 609, "reused_tokens": 0, "read_tokens": 609, "generated_tokens": 10, "prompt_tps": 499.6, "prompt_tps_source": "engine", "decode_tps": 49.1, "decode_tps_source": "engine", "prompt_ms": 1218.9, "decode_ms": 203.8, "drafts_offered": 14, "drafts_accepted": 7, "run": 0, "turn": 0 }, { "ok": true, "err": null, "ttft_s": 2.841, "total_s": 3.724, "prompt_tokens": 3701, "reused_tokens": 621, "read_tokens": 3080, "generated_tokens": 64, "prompt_tps": 1091.9, "prompt_tps_source": "engine", "decode_tps": 71.6, "decode_tps_source": "engine", "prompt_ms": 2820.7, "decode_ms": 894.0, "drafts_offered": 62, "drafts_accepted": 37, "run": 0, "turn": 1 }, { "ok": true, "err": null, "ttft_s": 2.409, "total_s": 3.193, "prompt_tokens": 5234, "reused_tokens": 3764, "read_tokens": 1470, "generated_tokens": 64, "prompt_tps": 614.9, "prompt_tps_source": "engine", "decode_tps": 80.5, "decode_tps_source": "engine", "prompt_ms": 2390.5, "decode_ms": 795.2, "drafts_offered": 71, "drafts_accepted": 39, "run": 0, "turn": 2 }, { "ok": true, "err": null, "ttft_s": 2.913, "total_s": 3.494, "prompt_tokens": 8678, "reused_tokens": 5227, "read_tokens": 3451, "generated_tokens": 48, "prompt_tps": 1193.3, "prompt_tps_source": "engine", "decode_tps": 81.2, "decode_tps_source": "engine", "prompt_ms": 2892.0, "decode_ms": 591.4, "drafts_offered": 55, "drafts_accepted": 25, "run": 0, "turn": 3 }, { "ok": true, "err": null, "ttft_s": 2.75, "total_s": 3.208, "prompt_tokens": 11585, "reused_tokens": 8725, "read_tokens": 2860, "generated_tokens": 48, "prompt_tps": 1048.7, "prompt_tps_source": "engine", "decode_tps": 102.4, "decode_tps_source": "engine", "prompt_ms": 2727.3, "decode_ms": 468.7, "drafts_offered": 42, "drafts_accepted": 31, "run": 0, "turn": 4 }, { "ok": true, "err": null, "ttft_s": 2.682, "total_s": 3.25, "prompt_tokens": 14478, "reused_tokens": 11632, "read_tokens": 2846, "generated_tokens": 50, "prompt_tps": 1070.4, "prompt_tps_source": "engine", "decode_tps": 86.5, "decode_tps_source": "engine", "prompt_ms": 2658.7, "decode_ms": 578.3, "drafts_offered": 57, "drafts_accepted": 28, "run": 0, "turn": 5 } ], [ { "ok": true, "err": null, "ttft_s": 1.249, "total_s": 1.407, "prompt_tokens": 609, "reused_tokens": 0, "read_tokens": 609, "generated_tokens": 10, "prompt_tps": 493.2, "prompt_tps_source": "engine", "decode_tps": 59.1, "decode_tps_source": "engine", "prompt_ms": 1234.7, "decode_ms": 169.3, "drafts_offered": 14, "drafts_accepted": 7, "run": 1, "turn": 0 }, { "ok": true, "err": null, "ttft_s": 2.602, "total_s": 3.223, "prompt_tokens": 3701, "reused_tokens": 621, "read_tokens": 3080, "generated_tokens": 64, "prompt_tps": 1190.6, "prompt_tps_source": "engine", "decode_tps": 101.4, "decode_tps_source": "engine", "prompt_ms": 2587.0, "decode_ms": 631.2, "drafts_offered": 57, "drafts_accepted": 39, "run": 1, "turn": 1 }, { "ok": true, "err": null, "ttft_s": 2.297, "total_s": 2.91, "prompt_tokens": 5234, "reused_tokens": 3764, "read_tokens": 1470, "generated_tokens": 64, "prompt_tps": 644.5, "prompt_tps_source": "engine", "decode_tps": 102.7, "decode_tps_source": "engine", "prompt_ms": 2281.0, "decode_ms": 623.1, "drafts_offered": 59, "drafts_accepted": 42, "run": 1, "turn": 2 }, { "ok": true, "err": null, "ttft_s": 2.72, "total_s": 3.27, "prompt_tokens": 8678, "reused_tokens": 5227, "read_tokens": 3451, "generated_tokens": 58, "prompt_tps": 1277.4, "prompt_tps_source": "engine", "decode_tps": 103.5, "decode_tps_source": "engine", "prompt_ms": 2701.5, "decode_ms": 560.4, "drafts_offered": 58, "drafts_accepted": 35, "run": 1, "turn": 3 }, { "ok": true, "err": null, "ttft_s": 2.577, "total_s": 2.898, "prompt_tokens": 11595, "reused_tokens": 8735, "read_tokens": 2860, "generated_tokens": 40, "prompt_tps": 1118.5, "prompt_tps_source": "engine", "decode_tps": 120.9, "decode_tps_source": "engine", "prompt_ms": 2557.0, "decode_ms": 330.8, "drafts_offered": 33, "drafts_accepted": 27, "run": 1, "turn": 4 }, { "ok": true, "err": null, "ttft_s": 2.566, "total_s": 3.125, "prompt_tokens": 14480, "reused_tokens": 11634, "read_tokens": 2846, "generated_tokens": 47, "prompt_tps": 1118.6, "prompt_tps_source": "engine", "decode_tps": 82.5, "decode_tps_source": "engine", "prompt_ms": 2544.3, "decode_ms": 569.7, "drafts_offered": 54, "drafts_accepted": 24, "run": 1, "turn": 5 } ], [ { "ok": true, "err": null, "ttft_s": 1.242, "total_s": 1.649, "prompt_tokens": 609, "reused_tokens": 0, "read_tokens": 609, "generated_tokens": 28, "prompt_tps": 496.0, "prompt_tps_source": "engine", "decode_tps": 67.0, "decode_tps_source": "engine", "prompt_ms": 1227.8, "decode_ms": 418.2, "drafts_offered": 33, "drafts_accepted": 16, "run": 2, "turn": 0 }, { "ok": true, "err": null, "ttft_s": 2.583, "total_s": 3.343, "prompt_tokens": 3719, "reused_tokens": 639, "read_tokens": 3080, "generated_tokens": 64, "prompt_tps": 1199.4, "prompt_tps_source": "engine", "decode_tps": 83.1, "decode_tps_source": "engine", "prompt_ms": 2567.9, "decode_ms": 770.0, "drafts_offered": 59, "drafts_accepted": 38, "run": 2, "turn": 1 }, { "ok": true, "err": null, "ttft_s": 2.312, "total_s": 3.044, "prompt_tokens": 5252, "reused_tokens": 3712, "read_tokens": 1540, "generated_tokens": 64, "prompt_tps": 670.6, "prompt_tps_source": "engine", "decode_tps": 86.3, "decode_tps_source": "engine", "prompt_ms": 2296.3, "decode_ms": 741.7, "drafts_offered": 70, "drafts_accepted": 37, "run": 2, "turn": 2 }, { "ok": true, "err": null, "ttft_s": 2.693, "total_s": 3.382, "prompt_tokens": 8696, "reused_tokens": 5316, "read_tokens": 3380, "generated_tokens": 50, "prompt_tps": 1263.6, "prompt_tps_source": "engine", "decode_tps": 71.5, "decode_tps_source": "engine", "prompt_ms": 2674.9, "decode_ms": 699.1, "drafts_offered": 64, "drafts_accepted": 21, "run": 2, "turn": 3 }, { "ok": true, "err": null, "ttft_s": 2.574, "total_s": 3.026, "prompt_tokens": 11605, "reused_tokens": 8745, "read_tokens": 2860, "generated_tokens": 51, "prompt_tps": 1121.1, "prompt_tps_source": "engine", "decode_tps": 110.4, "decode_tps_source": "engine", "prompt_ms": 2551.1, "decode_ms": 462.1, "drafts_offered": 48, "drafts_accepted": 33, "run": 2, "turn": 4 }, { "ok": true, "err": null, "ttft_s": 2.577, "total_s": 3.052, "prompt_tokens": 14501, "reused_tokens": 11655, "read_tokens": 2846, "generated_tokens": 44, "prompt_tps": 1114.7, "prompt_tps_source": "engine", "decode_tps": 90.6, "decode_tps_source": "engine", "prompt_ms": 2553.2, "decode_ms": 485.4, "drafts_offered": 45, "drafts_accepted": 24, "run": 2, "turn": 5 } ] ], "hit_pct": { "median": 68.625, "min": 68.617, "max": 68.688, "n": 3 }, "reread_tokens": { "median": 74, "min": 73, "max": 74, "n": 3 }, "new_tokens": { "median": 13633, "min": 13633, "max": 13633, "n": 3 }, "turn_ttft_s": { "median": 2.583, "min": 2.297, "max": 2.913, "n": 15 }, "last_turn_prompt_tokens": { "median": 14480, "min": 14478, "max": 14501, "n": 3 }, "turn_decode_tps": { "median": 86.5, "min": 71.5, "max": 120.9, "n": 15 }, "first_turn_ttft_s": { "median": 1.242, "min": 1.237, "max": 1.249, "n": 3 } }, "clients": { "clients": 2, "rounds": [ { "round": 0, "wall_s": 4.7, "tokens": 256, "total_tps": 54.46, "first_s": [ 0.7499102100009623, 3.2104223019996425 ], "latency_s": [ 2.666773988999921, 4.6997582349995355 ], "errors": [], "error_count": 0 }, { "round": 1, "wall_s": 3.69, "tokens": 256, "total_tps": 69.41, "first_s": [ 0.509523980999802, 2.21098651399916 ], "latency_s": [ 1.719431622999764, 3.6877483099997335 ], "errors": [], "error_count": 0 }, { "round": 2, "wall_s": 3.88, "tokens": 256, "total_tps": 65.93, "first_s": [ 0.4533425449990318, 2.322404643999107 ], "latency_s": [ 1.8312690089987882, 3.8826534029994946 ], "errors": [], "error_count": 0 } ], "total_tps": { "median": 65.93, "min": 54.46, "max": 69.41, "n": 3 }, "ttft_p50_s": 1.48, "ttft_p95_s": 2.988, "latency_p50_s": 3.177, "latency_p95_s": 4.495, "errors": 0 }, "skipped": {}, "failures": [] } }, "offline": { "status": "passed", "started_utc": "2026-10-12T02:08:07.531047+00:00", "finished_utc": "2026-10-12T02:08:07.749265+00:00", "network_namespace_interfaces": [ "lo" ], "external_tcp_connection_blocked": true, "expected_marker_present": true, "inference_wall_s": 0.19691806700029701 }, "vram_gib": 15.140625, "container_memory_peak_gib": 31.245433807373047, "resource_scope": "one-second observed peak during official standard suite; Docker charged memory includes cache, not host total RAM", "historical_eligibility_note": "Old daily_driver_eligible needs llama.cpp adaptive probes, which do not apply to Strata. It is not used as a new suitability verdict.", "workload_metrics": { "practical": { "responses": 72, "end_to_end_s": { "n": 72, "median": 5.5783000000000005, "min": 2.4485, "max": 31.5262 }, "engine_decode_tps": { "n": 72, "median": 95.1, "min": 66.2, "max": 118.2 }, "engine_uncached_prefill_tps": { "n": 72, "median": 227.2, "min": 100.3, "max": 2175.7 }, "drafts_offered": 27330, "drafts_accepted": 22233, "draft_acceptance_fraction": 0.8135016465422612, "total_prompt_tokens": 412752, "reused_prompt_tokens": 0, "prefix_reuse_fraction": 0.0, "client_ttft_s": null, "note": "Original clients are nonstreaming; engine decode/prefill and client end-to-end latency are distinct. Draft counters include speculation, not a separately isolated MTP-only rate." }, "humaneval": { "responses": 164, "end_to_end_s": { "n": 164, "median": 5.0706, "min": 1.4327, "max": 10.9235 }, "engine_decode_tps": { "n": 164, "median": 93.1, "min": 74.0, "max": 114.0 }, "engine_uncached_prefill_tps": { "n": 164, "median": 258.4, "min": 117.2, "max": 427.9 }, "drafts_offered": 51706, "drafts_accepted": 40520, "draft_acceptance_fraction": 0.7836614706223649, "total_prompt_tokens": 34439, "reused_prompt_tokens": 0, "prefix_reuse_fraction": 0.0, "client_ttft_s": null, "note": "Original clients are nonstreaming; engine decode/prefill and client end-to-end latency are distinct. Draft counters include speculation, not a separately isolated MTP-only rate." } }, "coding_failures": [ { "task_id": "HumanEval/10", "base_status": "fail", "plus_status": "fail" }, { "task_id": "HumanEval/32", "base_status": "fail", "plus_status": "fail" }, { "task_id": "HumanEval/76", "base_status": "pass", "plus_status": "fail" }, { "task_id": "HumanEval/91", "base_status": "pass", "plus_status": "fail" }, { "task_id": "HumanEval/99", "base_status": "pass", "plus_status": "fail" }, { "task_id": "HumanEval/132", "base_status": "fail", "plus_status": "fail" }, { "task_id": "HumanEval/134", "base_status": "pass", "plus_status": "fail" }, { "task_id": "HumanEval/141", "base_status": "pass", "plus_status": "fail" }, { "task_id": "HumanEval/145", "base_status": "fail", "plus_status": "fail" }, { "task_id": "HumanEval/151", "base_status": "pass", "plus_status": "fail" }, { "task_id": "HumanEval/160", "base_status": "fail", "plus_status": "fail" } ], "practical_missed_checks": [ { "task_id": "research-vendor-filter", "repetition": 0, "category": "research", "score": 90.91, "finish_reason": "stop", "failed_checks": [ "$625" ] }, { "task_id": "agent-tool-sequence", "repetition": 0, "category": "agent_tools", "score": 62.5, "finish_reason": "stop", "failed_checks": [ "search query supplied", "read URL is search placeholder", "save includes title, body and source" ] }, { "task_id": "executive-decision-memo", "repetition": 0, "category": "executive_content", "score": 85.71, "finish_reason": "stop", "failed_checks": [ "$100" ] }, { "task_id": "executive-waiting-on", "repetition": 0, "category": "executive_content", "score": 88.89, "finish_reason": "stop", "failed_checks": [ "approve" ] }, { "task_id": "research-vendor-filter", "repetition": 1, "category": "research", "score": 90.91, "finish_reason": "stop", "failed_checks": [ "$625" ] }, { "task_id": "agent-tool-sequence", "repetition": 1, "category": "agent_tools", "score": 62.5, "finish_reason": "stop", "failed_checks": [ "search query supplied", "read URL is search placeholder", "save includes title, body and source" ] }, { "task_id": "sysadmin-recovery", "repetition": 1, "category": "practical_sysadmin", "score": 80.0, "finish_reason": "length", "failed_checks": [ "API response completed" ] }, { "task_id": "executive-decision-memo", "repetition": 1, "category": "executive_content", "score": 85.71, "finish_reason": "stop", "failed_checks": [ "$100" ] }, { "task_id": "executive-waiting-on", "repetition": 1, "category": "executive_content", "score": 88.89, "finish_reason": "stop", "failed_checks": [ "approve" ] }, { "task_id": "research-source-conflict", "repetition": 2, "category": "research", "score": 90.0, "finish_reason": "stop", "failed_checks": [ "superseded" ] }, { "task_id": "research-vendor-filter", "repetition": 2, "category": "research", "score": 90.91, "finish_reason": "stop", "failed_checks": [ "$625" ] }, { "task_id": "agent-tool-sequence", "repetition": 2, "category": "agent_tools", "score": 62.5, "finish_reason": "stop", "failed_checks": [ "search query supplied", "read URL is search placeholder", "save includes title, body and source" ] }, { "task_id": "sysadmin-recovery", "repetition": 2, "category": "practical_sysadmin", "score": 80.0, "finish_reason": "length", "failed_checks": [ "no write/truncate file mode", "API response completed" ] }, { "task_id": "executive-decision-memo", "repetition": 2, "category": "executive_content", "score": 85.71, "finish_reason": "stop", "failed_checks": [ "$100" ] }, { "task_id": "executive-waiting-on", "repetition": 2, "category": "executive_content", "score": 88.89, "finish_reason": "stop", "failed_checks": [ "approve" ] } ], "api_diagnostics": { "status": "completed_diagnostics_not_a_quality_benchmark", "model": "qwen3.8-flash-next-iq2_xs", "tool_call_semantically_valid": true, "structured_output_semantically_valid": true, "streaming_final_content_seen": true }, "container_stop_states": [ { "context": 131072, "image": "sha256:f759e6167315dac77292d394a470240493dad78de6d23dd270eaa5c7ffc3e143", "status": "exited", "oom_killed": false, "exit_code": 0 }, { "context": 32768, "image": "sha256:f759e6167315dac77292d394a470240493dad78de6d23dd270eaa5c7ffc3e143", "status": "exited", "oom_killed": false, "exit_code": 0 }, { "context": 65536, "image": "sha256:f759e6167315dac77292d394a470240493dad78de6d23dd270eaa5c7ffc3e143", "status": "exited", "oom_killed": false, "exit_code": 0 } ] }, { "id": "strata-flash-next-iq3_xxs", "label": "ISTA-DASLab / GSQ-RCO \u00b7 IQ3_XXS", "provider": "ISTA-DASLab / GSQ-RCO", "model": "Qwen3.8-Flash-Next", "quant": "IQ3_XXS", "repository": "ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF", "revision": "ed59f92082b1e93c0e96d60a8b11aab089b52f09", "files": [ { "quant": "IQ3_XXS", "repo": "ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF", "revision": "ed59f92082b1e93c0e96d60a8b11aab089b52f09", "filename": "IQ3_XXS/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-00001-of-00002.gguf", "size_bytes": 47039860096, "sha256": "219ea929900dfa9ef091f3aa473fdba6874b65fcb36526d7d851ac9e95856d15" }, { "quant": "IQ3_XXS", "repo": "ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF", "revision": "ed59f92082b1e93c0e96d60a8b11aab089b52f09", "filename": "IQ3_XXS/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-00002-of-00002.gguf", "size_bytes": 28800138432, "sha256": "316b46f3a2dbd68c900f43136ab9449f9dcc3725dfd8c794847c204bc161e113" } ], "file_gib": 70.63150268793106, "mtp": true, "mtp_requested": 4, "mtp_min_p": 0.5, "reasoning_budget_tokens": 512, "balanced_score": 96.21, "humaneval_plus_percent": 93.90243902439023, "humaneval_base_percent": 97.5609756097561, "humaneval_evaluated": 164, "humaneval_plus_passed": 154, "role_scores": { "CFO / Analyst": 96.94, "Researcher / Scout": 97.04, "Hermes Coordinator": 95.77, "Librarian / Knowledge": 96.89, "Forge / Coder": 94.21 }, "category_scores": { "business": 96.875, "research": 97.5758, "knowledge": 100.0, "agent_tools": 92.9167, "practical_sysadmin": 92.2225, "executive_content": 93.8183 }, "practical_responses": 72, "practical_empty_finals": 0, "practical_length_finishes": 3, "dossier_evaluated_responses": 3, "dossier_valid_nonempty_responses": 3, "dossier_all_rows": [ { "repeat": 1, "status": "ok", "usage": { "prompt_tokens": 57135, "completion_tokens": 1997, "total_tokens": 59132, "prompt_tokens_details": { "cached_tokens": 0 } }, "timings": { "cache_n": 0, "prompt_n": 57135, "prompt_ms": 28997.6, "prompt_per_token_ms": 0.508, "prompt_per_second": 1970.3, "predicted_n": 1997, "predicted_ms": 25586.1, "predicted_per_token_ms": 12.929, "predicted_per_second": 77.3, "draft_n": 1637, "draft_n_accepted": 1292 }, "wall_time_s": 54.6304, "quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 771 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 771 }, "empty_final": false, "finish_reason": "stop", "transport_failed": false }, { "repeat": 2, "status": "ok", "usage": { "prompt_tokens": 57135, "completion_tokens": 2693, "total_tokens": 59828, "prompt_tokens_details": { "cached_tokens": 57130 } }, "timings": { "cache_n": 57130, "prompt_n": 5, "prompt_ms": 292.0, "prompt_per_token_ms": 58.4, "prompt_per_second": 17.1, "predicted_n": 2693, "predicted_ms": 34227.5, "predicted_per_token_ms": 12.795, "predicted_per_second": 78.2, "draft_n": 2238, "draft_n_accepted": 1740 }, "wall_time_s": 34.5712, "quality": { "overall_hermes": 97.5, "retrieval_facts": "14/14", "word_count": 1048 }, "validated_quality": { "overall_hermes": 97.5, "retrieval_facts": "14/14", "word_count": 1048 }, "empty_final": false, "finish_reason": "stop", "transport_failed": false }, { "repeat": 3, "status": "ok", "usage": { "prompt_tokens": 57135, "completion_tokens": 2028, "total_tokens": 59163, "prompt_tokens_details": { "cached_tokens": 57130 } }, "timings": { "cache_n": 57130, "prompt_n": 5, "prompt_ms": 1113.5, "prompt_per_token_ms": 222.7, "prompt_per_second": 4.5, "predicted_n": 2028, "predicted_ms": 26064.2, "predicted_per_token_ms": 12.948, "predicted_per_second": 77.2, "draft_n": 1700, "draft_n_accepted": 1295 }, "wall_time_s": 27.2296, "quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 770 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 770 }, "empty_final": false, "finish_reason": "stop", "transport_failed": false } ], "dossier_empty_finals": 0, "dossier_full_marker_recall_responses": 3, "dossier_successful_speed_eligible": true, "dossier_first_s": 54.6304, "dossier_warm_median_s": 30.900399999999998, "fixed_median_s_original": 34.5712, "adaptive_seconds": null, "context": { "configured_sonny_tokens": 65536, "configured_max_attempted_tokens": 131072, "dossier_actual_prompt_tokens": [ 57135, 57135, 57135 ], "max_actual_recall_success_tokens": 124963, "recall_tests": [ { "test": "needles-128k-ctx131072", "source": "results/iq3_xxs/needles-128k-ctx131072/results.json", "rows": [ { "length": "128k", "depth": 10, "prompt_tokens": 124962, "seconds": 66.6, "found": true, "word": "meadow-copper-504", "answer": "meadow-copper-504" }, { "length": "128k", "depth": 50, "prompt_tokens": 124962, "seconds": 66.8, "found": true, "word": "falcon-quartz-940", "answer": "falcon-quartz-940" }, { "length": "128k", "depth": 90, "prompt_tokens": 124963, "seconds": 58.7, "found": true, "word": "willow-cobalt-696", "answer": "willow-cobalt-696" } ], "cases": 3, "completed": 3, "found": 3, "empty_finals": 0, "transport_errors": 0, "actual_prompt_tokens": [ 124962, 124962, 124963 ] }, { "test": "needles-16k-ctx32768", "source": "results/iq3_xxs/needles-16k-ctx32768/results.json", "rows": [ { "length": "16k", "depth": 10, "prompt_tokens": 15146, "seconds": 8.2, "found": true, "word": "meadow-copper-504", "answer": "meadow-copper-504" }, { "length": "16k", "depth": 50, "prompt_tokens": 15146, "seconds": 8.1, "found": true, "word": "falcon-quartz-940", "answer": "falcon-quartz-940" }, { "length": "16k", "depth": 90, "prompt_tokens": 15147, "seconds": 8.1, "found": true, "word": "willow-cobalt-696", "answer": "willow-cobalt-696" } ], "cases": 3, "completed": 3, "found": 3, "empty_finals": 0, "transport_errors": 0, "actual_prompt_tokens": [ 15146, 15146, 15147 ] }, { "test": "needles-32k-ctx32768", "source": "results/iq3_xxs/needles-32k-ctx32768/results.json", "rows": [ { "length": "32k", "depth": 10, "error": "HTTP Error 400: Bad Request" }, { "length": "32k", "depth": 50, "error": "HTTP Error 400: Bad Request" }, { "length": "32k", "depth": 90, "error": "HTTP Error 400: Bad Request" } ], "cases": 3, "completed": 0, "found": 0, "empty_finals": 0, "transport_errors": 3, "actual_prompt_tokens": [ null, null, null ] }, { "test": "needles-32k-ctx65536", "source": "results/iq3_xxs/needles-32k-ctx65536/results.json", "rows": [ { "length": "32k", "depth": 10, "prompt_tokens": 32804, "seconds": 17.8, "found": true, "word": "meadow-copper-504", "answer": "meadow-copper-504" }, { "length": "32k", "depth": 50, "prompt_tokens": 32804, "seconds": 17.1, "found": true, "word": "falcon-quartz-940", "answer": "falcon-quartz-940" }, { "length": "32k", "depth": 90, "prompt_tokens": 32805, "seconds": 17.1, "found": true, "word": "willow-cobalt-696", "answer": "willow-cobalt-696" } ], "cases": 3, "completed": 3, "found": 3, "empty_finals": 0, "transport_errors": 0, "actual_prompt_tokens": [ 32804, 32804, 32805 ] }, { "test": "needles-64k-ctx131072", "source": "results/iq3_xxs/needles-64k-ctx131072/results.json", "rows": [ { "length": "64k", "depth": 10, "prompt_tokens": 64925, "seconds": 34.0, "found": true, "word": "meadow-copper-504", "answer": "meadow-copper-504" }, { "length": "64k", "depth": 50, "prompt_tokens": 64925, "seconds": 33.0, "found": true, "word": "falcon-quartz-940", "answer": "falcon-quartz-940" }, { "length": "64k", "depth": 90, "prompt_tokens": 64926, "seconds": 25.1, "found": true, "word": "willow-cobalt-696", "answer": "willow-cobalt-696" } ], "cases": 3, "completed": 3, "found": 3, "empty_finals": 0, "transport_errors": 0, "actual_prompt_tokens": [ 64925, 64925, 64926 ] }, { "test": "needles-8k-ctx32768", "source": "results/iq3_xxs/needles-8k-ctx32768/results.json", "rows": [ { "length": "8k", "depth": 10, "prompt_tokens": 6920, "seconds": 4.8, "found": true, "word": "meadow-copper-504", "answer": "meadow-copper-504" }, { "length": "8k", "depth": 50, "prompt_tokens": 6920, "seconds": 4.2, "found": true, "word": "falcon-quartz-940", "answer": "falcon-quartz-940" }, { "length": "8k", "depth": 90, "prompt_tokens": 6921, "seconds": 4.1, "found": true, "word": "willow-cobalt-696", "answer": "willow-cobalt-696" } ], "cases": 3, "completed": 3, "found": 3, "empty_finals": 0, "transport_errors": 0, "actual_prompt_tokens": [ 6920, 6920, 6921 ] } ], "note": "Successful recall is limited to the exact tested inputs/depths. Context configuration is not proof of answer quality." }, "official_standard": { "mode": "standard-32768", "source": "official-benchmarks/iq3_xxs/standard-32768/2026-10-12-community-rtx-5070-ti-qwen3-8-flash-next-iq3-xxs/results.json", "schema": 1, "params": { "runs": 3, "think": false, "clients": 0, "long": false, "started_server": false, "command": "python tools/strata_bench.py" }, "elapsed_s": 119.55528283119202, "machine": { "gpu": [ "NVIDIA GeForce RTX 5070 Ti" ], "gpu_vram_mib": 16303, "gpu_driver": "615.71.09", "cuda": null, "cpu": "AMD Ryzen 5 5600G with Radeon Graphics", "cpu_cores": 6, "cpu_threads": 12, "ram_gib": 58.8, "os": "Linux 7.0.14-17-pve", "python": "3.12.3", "strata_version": "0.1.43", "strata_commit": "7695859a", "engine_version": "0.1.43", "model": "qwen3.8-flash-next-iq3_xxs", "quant": "IQ3_XXS", "context": 32768, "parallel_slots": 1, "config": { "exe": "strata", "args": [ "--pack", "iq3_xxs", "--native", "Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-00001-of-00002.gguf", "--ple-gguf", "Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-00002-of-00002.gguf", "--expert-profile", "expert-profile.bin", "--expert-cache", "auto", "--prefill", "auto", "--spec", "4", "--spec-min-p", "0.5", "--mtp", "rt", "--max-context", "32768", "--kv", "int8", "--resident-experts", "--vram-reserve-mib", "1024" ], "model_name": "qwen3.8-flash-next-iq3_xxs", "lib_dirs": [ "/usr/local/cuda/lib64" ], "port": 8080, "gpu": 0, "gpus_asked": true, "host": "127.0.0.1", "open_browser": false, "parallel": 1 }, "launch": "strata --pack iq3_xxs --native Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-00001-of-00002.gguf --ple-gguf Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-00002-of-00002.gguf --expert-profile expert-profile.bin --expert-cache auto --prefill auto --spec 4 --spec-min-p 0.5 --mtp rt --max-context 32768 --kv int8 --resident-experts --vram-reserve-mib 1024", "engine_info": { "model": "qwen3.8-flash-next-iq3_xxs", "max_context": 32768, "images": false, "version": "0.1.43", "logit_bias": 1, "context": 32768, "kv": "int8", "kv_resident": 0, "expert_slots": 5518, "expert_cache_mib": 9154, "expert_slots_primary": 5518, "expert_cache_primary_mib": 9154, "spec": 6, "mtp_max": 4, "lookup": 3, "vram_free_mib": 505, "cvec": 0, "arena_mib": 35220, "pool_workers": 5, "pcie_frac": "0.37", "spec_min_p": "0.50", "conversation_cache_mib": 0, "conversation_cache_slots": 4, "conversation_cache_min_free_mib": 2560, "conversation_cache_min_tokens": 0, "tail_role_token": -1, "vram_elastic": 0, "engine": "0.1.43" }, "vram_after": { "elastic": false } }, "metrics": { "chat": { "requests": [ { "ok": true, "err": null, "ttft_s": 0.501, "total_s": 1.991, "prompt_tokens": 51, "reused_tokens": 0, "read_tokens": 51, "generated_tokens": 100, "prompt_tps": 104.6, "prompt_tps_source": "engine", "decode_tps": 66.6, "decode_tps_source": "engine", "prompt_ms": 487.8, "decode_ms": 1500.7, "drafts_offered": 86, "drafts_accepted": 70, "run": 0, "prompt": 0 }, { "ok": true, "err": null, "ttft_s": 0.493, "total_s": 2.361, "prompt_tokens": 43, "reused_tokens": 0, "read_tokens": 43, "generated_tokens": 100, "prompt_tps": 89.9, "prompt_tps_source": "engine", "decode_tps": 53.2, "decode_tps_source": "engine", "prompt_ms": 478.4, "decode_ms": 1880.7, "drafts_offered": 97, "drafts_accepted": 53, "run": 0, "prompt": 1 }, { "ok": true, "err": null, "ttft_s": 0.506, "total_s": 2.294, "prompt_tokens": 46, "reused_tokens": 0, "read_tokens": 46, "generated_tokens": 100, "prompt_tps": 93.6, "prompt_tps_source": "engine", "decode_tps": 55.6, "decode_tps_source": "engine", "prompt_ms": 491.3, "decode_ms": 1800.1, "drafts_offered": 100, "drafts_accepted": 56, "run": 0, "prompt": 2 }, { "ok": true, "err": null, "ttft_s": 0.488, "total_s": 1.856, "prompt_tokens": 51, "reused_tokens": 0, "read_tokens": 51, "generated_tokens": 100, "prompt_tps": 107.5, "prompt_tps_source": "engine", "decode_tps": 72.5, "decode_tps_source": "engine", "prompt_ms": 474.4, "decode_ms": 1379.6, "drafts_offered": 85, "drafts_accepted": 71, "run": 1, "prompt": 0 }, { "ok": true, "err": null, "ttft_s": 0.415, "total_s": 1.898, "prompt_tokens": 43, "reused_tokens": 0, "read_tokens": 43, "generated_tokens": 100, "prompt_tps": 107.3, "prompt_tps_source": "engine", "decode_tps": 66.9, "decode_tps_source": "engine", "prompt_ms": 400.8, "decode_ms": 1495.2, "drafts_offered": 95, "drafts_accepted": 56, "run": 1, "prompt": 1 }, { "ok": true, "err": null, "ttft_s": 0.465, "total_s": 2.088, "prompt_tokens": 46, "reused_tokens": 0, "read_tokens": 46, "generated_tokens": 100, "prompt_tps": 102.1, "prompt_tps_source": "engine", "decode_tps": 61.2, "decode_tps_source": "engine", "prompt_ms": 450.4, "decode_ms": 1635.2, "drafts_offered": 107, "drafts_accepted": 57, "run": 1, "prompt": 2 }, { "ok": true, "err": null, "ttft_s": 0.502, "total_s": 1.979, "prompt_tokens": 51, "reused_tokens": 0, "read_tokens": 51, "generated_tokens": 100, "prompt_tps": 104.6, "prompt_tps_source": "engine", "decode_tps": 67.2, "decode_tps_source": "engine", "prompt_ms": 487.7, "decode_ms": 1488.8, "drafts_offered": 89, "drafts_accepted": 69, "run": 2, "prompt": 0 }, { "ok": true, "err": null, "ttft_s": 0.424, "total_s": 1.918, "prompt_tokens": 43, "reused_tokens": 0, "read_tokens": 43, "generated_tokens": 100, "prompt_tps": 105.2, "prompt_tps_source": "engine", "decode_tps": 66.4, "decode_tps_source": "engine", "prompt_ms": 408.8, "decode_ms": 1506.7, "drafts_offered": 102, "drafts_accepted": 56, "run": 2, "prompt": 1 }, { "ok": true, "err": null, "ttft_s": 0.463, "total_s": 2.071, "prompt_tokens": 46, "reused_tokens": 0, "read_tokens": 46, "generated_tokens": 100, "prompt_tps": 102.7, "prompt_tps_source": "engine", "decode_tps": 61.7, "decode_tps_source": "engine", "prompt_ms": 447.7, "decode_ms": 1621.1, "drafts_offered": 105, "drafts_accepted": 55, "run": 2, "prompt": 2 } ], "decode_tps": { "median": 66.4, "min": 53.2, "max": 72.5, "n": 9 }, "ttft_s": { "median": 0.488, "min": 0.415, "max": 0.506, "n": 9 }, "generated_tokens": { "median": 100, "min": 100, "max": 100, "n": 9 }, "prompt_tokens": { "median": 46, "min": 43, "max": 51, "n": 9 } }, "prompts": { "4K": { "target_tokens": 4096, "requests": [ { "ok": true, "err": null, "ttft_s": 3.34, "total_s": 3.754, "prompt_tokens": 4019, "reused_tokens": 0, "read_tokens": 4019, "generated_tokens": 24, "prompt_tps": 1210.0, "prompt_tps_source": "engine", "decode_tps": 56.1, "decode_tps_source": "engine", "prompt_ms": 3321.6, "decode_ms": 427.6, "drafts_offered": 19, "drafts_accepted": 12, "run": 0 }, { "ok": true, "err": null, "ttft_s": 3.591, "total_s": 3.907, "prompt_tokens": 4019, "reused_tokens": 0, "read_tokens": 4019, "generated_tokens": 24, "prompt_tps": 1124.7, "prompt_tps_source": "engine", "decode_tps": 72.9, "decode_tps_source": "engine", "prompt_ms": 3573.5, "decode_ms": 329.1, "drafts_offered": 20, "drafts_accepted": 13, "run": 1 }, { "ok": true, "err": null, "ttft_s": 3.237, "total_s": 3.569, "prompt_tokens": 4019, "reused_tokens": 0, "read_tokens": 4019, "generated_tokens": 24, "prompt_tps": 1248.1, "prompt_tps_source": "engine", "decode_tps": 69.7, "decode_tps_source": "engine", "prompt_ms": 3220.0, "decode_ms": 344.3, "drafts_offered": 20, "drafts_accepted": 12, "run": 2 } ], "prompt_tps": { "median": 1210.0, "min": 1124.7, "max": 1248.1, "n": 3 }, "ttft_s": { "median": 3.34, "min": 3.237, "max": 3.591, "n": 3 }, "prompt_tokens": { "median": 4019, "min": 4019, "max": 4019, "n": 3 }, "reused_tokens": { "median": 0, "min": 0, "max": 0, "n": 3 } }, "16K": { "target_tokens": 16384, "requests": [ { "ok": true, "err": null, "ttft_s": 8.34, "total_s": 8.715, "prompt_tokens": 16058, "reused_tokens": 0, "read_tokens": 16058, "generated_tokens": 24, "prompt_tps": 1932.1, "prompt_tps_source": "engine", "decode_tps": 62.0, "decode_tps_source": "engine", "prompt_ms": 8311.2, "decode_ms": 387.4, "drafts_offered": 18, "drafts_accepted": 11, "run": 0 }, { "ok": true, "err": null, "ttft_s": 8.278, "total_s": 8.59, "prompt_tokens": 16058, "reused_tokens": 0, "read_tokens": 16058, "generated_tokens": 24, "prompt_tps": 1945.4, "prompt_tps_source": "engine", "decode_tps": 74.2, "decode_tps_source": "engine", "prompt_ms": 8254.5, "decode_ms": 323.6, "drafts_offered": 23, "drafts_accepted": 13, "run": 1 }, { "ok": true, "err": null, "ttft_s": 8.263, "total_s": 8.51, "prompt_tokens": 16058, "reused_tokens": 0, "read_tokens": 16058, "generated_tokens": 24, "prompt_tps": 1948.7, "prompt_tps_source": "engine", "decode_tps": 92.8, "decode_tps_source": "engine", "prompt_ms": 8240.2, "decode_ms": 258.6, "drafts_offered": 20, "drafts_accepted": 11, "run": 2 } ], "prompt_tps": { "median": 1945.4, "min": 1932.1, "max": 1948.7, "n": 3 }, "ttft_s": { "median": 8.278, "min": 8.263, "max": 8.34, "n": 3 }, "prompt_tokens": { "median": 16058, "min": 16058, "max": 16058, "n": 3 }, "reused_tokens": { "median": 0, "min": 0, "max": 0, "n": 3 } } }, "agent": { "sessions": [ [ { "ok": true, "err": null, "ttft_s": 1.432, "total_s": 1.768, "prompt_tokens": 609, "reused_tokens": 0, "read_tokens": 609, "generated_tokens": 15, "prompt_tps": 430.5, "prompt_tps_source": "engine", "decode_tps": 42.8, "decode_tps_source": "engine", "prompt_ms": 1414.5, "decode_ms": 350.1, "drafts_offered": 19, "drafts_accepted": 8, "run": 0, "turn": 0 }, { "ok": true, "err": null, "ttft_s": 3.285, "total_s": 4.172, "prompt_tokens": 3706, "reused_tokens": 625, "read_tokens": 3081, "generated_tokens": 64, "prompt_tps": 944.5, "prompt_tps_source": "engine", "decode_tps": 71.0, "decode_tps_source": "engine", "prompt_ms": 3262.0, "decode_ms": 901.3, "drafts_offered": 47, "drafts_accepted": 43, "run": 0, "turn": 1 }, { "ok": true, "err": null, "ttft_s": 2.879, "total_s": 3.875, "prompt_tokens": 5239, "reused_tokens": 3769, "read_tokens": 1470, "generated_tokens": 64, "prompt_tps": 514.3, "prompt_tps_source": "engine", "decode_tps": 63.5, "decode_tps_source": "engine", "prompt_ms": 2858.4, "decode_ms": 1008.5, "drafts_offered": 69, "drafts_accepted": 39, "run": 0, "turn": 2 }, { "ok": true, "err": null, "ttft_s": 3.289, "total_s": 4.217, "prompt_tokens": 8683, "reused_tokens": 5303, "read_tokens": 3380, "generated_tokens": 64, "prompt_tps": 1034.6, "prompt_tps_source": "engine", "decode_tps": 68.0, "decode_tps_source": "engine", "prompt_ms": 3266.9, "decode_ms": 940.5, "drafts_offered": 70, "drafts_accepted": 35, "run": 0, "turn": 3 }, { "ok": true, "err": null, "ttft_s": 3.186, "total_s": 3.789, "prompt_tokens": 11607, "reused_tokens": 8746, "read_tokens": 2861, "generated_tokens": 54, "prompt_tps": 904.9, "prompt_tps_source": "engine", "decode_tps": 87.8, "decode_tps_source": "engine", "prompt_ms": 3161.6, "decode_ms": 614.8, "drafts_offered": 50, "drafts_accepted": 34, "run": 0, "turn": 4 }, { "ok": true, "err": null, "ttft_s": 3.103, "total_s": 3.73, "prompt_tokens": 14506, "reused_tokens": 11660, "read_tokens": 2846, "generated_tokens": 48, "prompt_tps": 924.2, "prompt_tps_source": "engine", "decode_tps": 75.2, "decode_tps_source": "engine", "prompt_ms": 3079.5, "decode_ms": 638.7, "drafts_offered": 54, "drafts_accepted": 28, "run": 0, "turn": 5 } ], [ { "ok": true, "err": null, "ttft_s": 1.456, "total_s": 1.677, "prompt_tokens": 609, "reused_tokens": 0, "read_tokens": 609, "generated_tokens": 13, "prompt_tps": 422.9, "prompt_tps_source": "engine", "decode_tps": 55.6, "decode_tps_source": "engine", "prompt_ms": 1440.2, "decode_ms": 233.7, "drafts_offered": 15, "drafts_accepted": 10, "run": 1, "turn": 0 }, { "ok": true, "err": null, "ttft_s": 3.026, "total_s": 3.709, "prompt_tokens": 3704, "reused_tokens": 624, "read_tokens": 3080, "generated_tokens": 64, "prompt_tps": 1023.4, "prompt_tps_source": "engine", "decode_tps": 92.2, "decode_tps_source": "engine", "prompt_ms": 3009.7, "decode_ms": 694.5, "drafts_offered": 48, "drafts_accepted": 42, "run": 1, "turn": 1 }, { "ok": true, "err": null, "ttft_s": 2.814, "total_s": 3.665, "prompt_tokens": 5237, "reused_tokens": 3767, "read_tokens": 1470, "generated_tokens": 64, "prompt_tps": 525.6, "prompt_tps_source": "engine", "decode_tps": 74.3, "decode_tps_source": "engine", "prompt_ms": 2797.0, "decode_ms": 861.9, "drafts_offered": 63, "drafts_accepted": 41, "run": 1, "turn": 2 }, { "ok": true, "err": null, "ttft_s": 3.135, "total_s": 3.916, "prompt_tokens": 8681, "reused_tokens": 5230, "read_tokens": 3451, "generated_tokens": 54, "prompt_tps": 1107.8, "prompt_tps_source": "engine", "decode_tps": 68.1, "decode_tps_source": "engine", "prompt_ms": 3115.3, "decode_ms": 793.2, "drafts_offered": 58, "drafts_accepted": 27, "run": 1, "turn": 3 }, { "ok": true, "err": null, "ttft_s": 3.027, "total_s": 3.294, "prompt_tokens": 11594, "reused_tokens": 8734, "read_tokens": 2860, "generated_tokens": 27, "prompt_tps": 951.5, "prompt_tps_source": "engine", "decode_tps": 96.9, "decode_tps_source": "engine", "prompt_ms": 3005.7, "decode_ms": 278.7, "drafts_offered": 23, "drafts_accepted": 18, "run": 1, "turn": 4 }, { "ok": true, "err": null, "ttft_s": 3.017, "total_s": 3.252, "prompt_tokens": 14466, "reused_tokens": 11620, "read_tokens": 2846, "generated_tokens": 12, "prompt_tps": 950.3, "prompt_tps_source": "engine", "decode_tps": 48.7, "decode_tps_source": "engine", "prompt_ms": 2994.8, "decode_ms": 246.4, "drafts_offered": 20, "drafts_accepted": 4, "run": 1, "turn": 5 } ], [ { "ok": true, "err": null, "ttft_s": 1.444, "total_s": 1.637, "prompt_tokens": 609, "reused_tokens": 0, "read_tokens": 609, "generated_tokens": 13, "prompt_tps": 426.5, "prompt_tps_source": "engine", "decode_tps": 63.0, "decode_tps_source": "engine", "prompt_ms": 1427.8, "decode_ms": 206.5, "drafts_offered": 12, "drafts_accepted": 7, "run": 2, "turn": 0 }, { "ok": true, "err": null, "ttft_s": 3.023, "total_s": 3.721, "prompt_tokens": 3704, "reused_tokens": 621, "read_tokens": 3083, "generated_tokens": 64, "prompt_tps": 1025.5, "prompt_tps_source": "engine", "decode_tps": 90.1, "decode_tps_source": "engine", "prompt_ms": 3006.4, "decode_ms": 710.2, "drafts_offered": 54, "drafts_accepted": 45, "run": 2, "turn": 1 }, { "ok": true, "err": null, "ttft_s": 2.822, "total_s": 3.595, "prompt_tokens": 5237, "reused_tokens": 3697, "read_tokens": 1540, "generated_tokens": 64, "prompt_tps": 549.0, "prompt_tps_source": "engine", "decode_tps": 81.6, "decode_tps_source": "engine", "prompt_ms": 2804.9, "decode_ms": 784.6, "drafts_offered": 62, "drafts_accepted": 40, "run": 2, "turn": 2 }, { "ok": true, "err": null, "ttft_s": 3.125, "total_s": 3.633, "prompt_tokens": 8681, "reused_tokens": 5300, "read_tokens": 3381, "generated_tokens": 42, "prompt_tps": 1088.6, "prompt_tps_source": "engine", "decode_tps": 80.8, "decode_tps_source": "engine", "prompt_ms": 3105.7, "decode_ms": 519.5, "drafts_offered": 44, "drafts_accepted": 24, "run": 2, "turn": 3 }, { "ok": true, "err": null, "ttft_s": 3.008, "total_s": 3.301, "prompt_tokens": 11582, "reused_tokens": 8722, "read_tokens": 2860, "generated_tokens": 25, "prompt_tps": 957.4, "prompt_tps_source": "engine", "decode_tps": 82.0, "decode_tps_source": "engine", "prompt_ms": 2987.2, "decode_ms": 304.9, "drafts_offered": 26, "drafts_accepted": 15, "run": 2, "turn": 4 }, { "ok": true, "err": null, "ttft_s": 3.022, "total_s": 3.35, "prompt_tokens": 14452, "reused_tokens": 11606, "read_tokens": 2846, "generated_tokens": 21, "prompt_tps": 948.9, "prompt_tps_source": "engine", "decode_tps": 61.8, "decode_tps_source": "engine", "prompt_ms": 2999.4, "decode_ms": 339.7, "drafts_offered": 27, "drafts_accepted": 10, "run": 2, "turn": 5 } ] ], "hit_pct": { "median": 68.621, "min": 68.595, "max": 68.821, "n": 3 }, "reread_tokens": { "median": 74, "min": 3, "max": 75, "n": 3 }, "new_tokens": { "median": 13635, "min": 13633, "max": 13635, "n": 3 }, "turn_ttft_s": { "median": 3.026, "min": 2.814, "max": 3.289, "n": 15 }, "last_turn_prompt_tokens": { "median": 14466, "min": 14452, "max": 14506, "n": 3 }, "turn_decode_tps": { "median": 75.2, "min": 48.7, "max": 96.9, "n": 15 }, "first_turn_ttft_s": { "median": 1.444, "min": 1.432, "max": 1.456, "n": 3 } }, "clients": {}, "skipped": {}, "failures": [] } }, "official_optional": { "mode": "optional-65536", "source": "official-benchmarks/iq3_xxs/optional-65536/2026-10-12-community-rtx-5070-ti-qwen3-8-flash-next-iq3-xxs/results.json", "schema": 1, "params": { "runs": 3, "think": false, "clients": 2, "long": true, "started_server": false, "command": "python tools/strata_bench.py --clients 2 --long" }, "elapsed_s": 189.6228790283203, "machine": { "gpu": [ "NVIDIA GeForce RTX 5070 Ti" ], "gpu_vram_mib": 16303, "gpu_driver": "615.71.09", "cuda": null, "cpu": "AMD Ryzen 5 5600G with Radeon Graphics", "cpu_cores": 6, "cpu_threads": 12, "ram_gib": 58.8, "os": "Linux 7.0.14-17-pve", "python": "3.12.3", "strata_version": "0.1.43", "strata_commit": "7695859a", "engine_version": "0.1.43", "model": "qwen3.8-flash-next-iq3_xxs", "quant": "IQ3_XXS", "context": 65536, "parallel_slots": 1, "config": { "exe": "strata", "args": [ "--pack", "iq3_xxs", "--native", "Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-00001-of-00002.gguf", "--ple-gguf", "Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-00002-of-00002.gguf", "--expert-profile", "expert-profile.bin", "--expert-cache", "auto", "--prefill", "auto", "--spec", "4", "--spec-min-p", "0.5", "--mtp", "rt", "--max-context", "65536", "--kv", "int8", "--resident-experts", "--vram-reserve-mib", "1024", "--kv-resident", "32768" ], "model_name": "qwen3.8-flash-next-iq3_xxs", "lib_dirs": [ "/usr/local/cuda/lib64" ], "port": 8080, "gpu": 0, "gpus_asked": true, "host": "127.0.0.1", "open_browser": false, "parallel": 1 }, "launch": "strata --pack iq3_xxs --native Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-00001-of-00002.gguf --ple-gguf Qwen3.8-Flash-Next-GSQ-RCO-IQ3_XXS-00002-of-00002.gguf --expert-profile expert-profile.bin --expert-cache auto --prefill auto --spec 4 --spec-min-p 0.5 --mtp rt --max-context 65536 --kv int8 --resident-experts --vram-reserve-mib 1024 --kv-resident 32768", "engine_info": { "model": "qwen3.8-flash-next-iq3_xxs", "max_context": 65536, "images": false, "version": "0.1.43", "logit_bias": 1, "context": 65536, "kv": "int8", "kv_resident": 32768, "expert_slots": 5480, "expert_cache_mib": 9098, "expert_slots_primary": 5480, "expert_cache_primary_mib": 9098, "spec": 6, "mtp_max": 4, "lookup": 3, "vram_free_mib": 505, "cvec": 0, "arena_mib": 35342, "pool_workers": 5, "pcie_frac": "0.37", "spec_min_p": "0.50", "conversation_cache_mib": 0, "conversation_cache_slots": 4, "conversation_cache_min_free_mib": 2560, "conversation_cache_min_tokens": 0, "tail_role_token": -1, "vram_elastic": 0, "engine": "0.1.43" }, "vram_after": { "elastic": false } }, "metrics": { "chat": { "requests": [ { "ok": true, "err": null, "ttft_s": 0.555, "total_s": 1.873, "prompt_tokens": 51, "reused_tokens": 0, "read_tokens": 51, "generated_tokens": 100, "prompt_tps": 94.6, "prompt_tps_source": "engine", "decode_tps": 75.1, "decode_tps_source": "engine", "prompt_ms": 539.0, "decode_ms": 1331.5, "drafts_offered": 84, "drafts_accepted": 71, "run": 0, "prompt": 0 }, { "ok": true, "err": null, "ttft_s": 0.492, "total_s": 2.197, "prompt_tokens": 43, "reused_tokens": 0, "read_tokens": 43, "generated_tokens": 100, "prompt_tps": 90.4, "prompt_tps_source": "engine", "decode_tps": 58.2, "decode_tps_source": "engine", "prompt_ms": 475.7, "decode_ms": 1718.5, "drafts_offered": 100, "drafts_accepted": 55, "run": 0, "prompt": 1 }, { "ok": true, "err": null, "ttft_s": 0.545, "total_s": 2.335, "prompt_tokens": 46, "reused_tokens": 0, "read_tokens": 46, "generated_tokens": 100, "prompt_tps": 87.0, "prompt_tps_source": "engine", "decode_tps": 55.4, "decode_tps_source": "engine", "prompt_ms": 528.7, "decode_ms": 1804.2, "drafts_offered": 95, "drafts_accepted": 54, "run": 0, "prompt": 2 }, { "ok": true, "err": null, "ttft_s": 0.537, "total_s": 1.956, "prompt_tokens": 51, "reused_tokens": 0, "read_tokens": 51, "generated_tokens": 100, "prompt_tps": 97.8, "prompt_tps_source": "engine", "decode_tps": 69.8, "decode_tps_source": "engine", "prompt_ms": 521.6, "decode_ms": 1432.3, "drafts_offered": 86, "drafts_accepted": 71, "run": 1, "prompt": 0 }, { "ok": true, "err": null, "ttft_s": 0.454, "total_s": 1.949, "prompt_tokens": 43, "reused_tokens": 0, "read_tokens": 43, "generated_tokens": 100, "prompt_tps": 98.3, "prompt_tps_source": "engine", "decode_tps": 66.3, "decode_tps_source": "engine", "prompt_ms": 437.6, "decode_ms": 1509.2, "drafts_offered": 99, "drafts_accepted": 56, "run": 1, "prompt": 1 }, { "ok": true, "err": null, "ttft_s": 0.479, "total_s": 2.077, "prompt_tokens": 46, "reused_tokens": 0, "read_tokens": 46, "generated_tokens": 100, "prompt_tps": 99.5, "prompt_tps_source": "engine", "decode_tps": 62.0, "decode_tps_source": "engine", "prompt_ms": 462.4, "decode_ms": 1611.8, "drafts_offered": 103, "drafts_accepted": 56, "run": 1, "prompt": 2 }, { "ok": true, "err": null, "ttft_s": 0.525, "total_s": 1.982, "prompt_tokens": 51, "reused_tokens": 0, "read_tokens": 51, "generated_tokens": 100, "prompt_tps": 100.2, "prompt_tps_source": "engine", "decode_tps": 68.0, "decode_tps_source": "engine", "prompt_ms": 509.1, "decode_ms": 1471.1, "drafts_offered": 87, "drafts_accepted": 70, "run": 2, "prompt": 0 }, { "ok": true, "err": null, "ttft_s": 0.432, "total_s": 1.897, "prompt_tokens": 43, "reused_tokens": 0, "read_tokens": 43, "generated_tokens": 100, "prompt_tps": 103.4, "prompt_tps_source": "engine", "decode_tps": 67.6, "decode_tps_source": "engine", "prompt_ms": 415.7, "decode_ms": 1478.6, "drafts_offered": 99, "drafts_accepted": 56, "run": 2, "prompt": 1 }, { "ok": true, "err": null, "ttft_s": 0.462, "total_s": 2.131, "prompt_tokens": 46, "reused_tokens": 0, "read_tokens": 46, "generated_tokens": 100, "prompt_tps": 103.3, "prompt_tps_source": "engine", "decode_tps": 59.4, "decode_tps_source": "engine", "prompt_ms": 445.2, "decode_ms": 1683.0, "drafts_offered": 102, "drafts_accepted": 53, "run": 2, "prompt": 2 } ], "decode_tps": { "median": 66.3, "min": 55.4, "max": 75.1, "n": 9 }, "ttft_s": { "median": 0.492, "min": 0.432, "max": 0.555, "n": 9 }, "generated_tokens": { "median": 100, "min": 100, "max": 100, "n": 9 }, "prompt_tokens": { "median": 46, "min": 43, "max": 51, "n": 9 } }, "prompts": { "4K": { "target_tokens": 4096, "requests": [ { "ok": true, "err": null, "ttft_s": 3.355, "total_s": 3.827, "prompt_tokens": 4019, "reused_tokens": 0, "read_tokens": 4019, "generated_tokens": 24, "prompt_tps": 1204.7, "prompt_tps_source": "engine", "decode_tps": 49.4, "decode_tps_source": "engine", "prompt_ms": 3336.0, "decode_ms": 486.2, "drafts_offered": 21, "drafts_accepted": 12, "run": 0 }, { "ok": true, "err": null, "ttft_s": 3.253, "total_s": 3.608, "prompt_tokens": 4019, "reused_tokens": 0, "read_tokens": 4019, "generated_tokens": 24, "prompt_tps": 1242.6, "prompt_tps_source": "engine", "decode_tps": 65.0, "decode_tps_source": "engine", "prompt_ms": 3234.4, "decode_ms": 369.2, "drafts_offered": 20, "drafts_accepted": 13, "run": 1 }, { "ok": true, "err": null, "ttft_s": 3.269, "total_s": 3.592, "prompt_tokens": 4019, "reused_tokens": 0, "read_tokens": 4019, "generated_tokens": 24, "prompt_tps": 1236.6, "prompt_tps_source": "engine", "decode_tps": 71.2, "decode_tps_source": "engine", "prompt_ms": 3250.0, "decode_ms": 337.3, "drafts_offered": 20, "drafts_accepted": 12, "run": 2 } ], "prompt_tps": { "median": 1236.6, "min": 1204.7, "max": 1242.6, "n": 3 }, "ttft_s": { "median": 3.269, "min": 3.253, "max": 3.355, "n": 3 }, "prompt_tokens": { "median": 4019, "min": 4019, "max": 4019, "n": 3 }, "reused_tokens": { "median": 0, "min": 0, "max": 0, "n": 3 } }, "16K": { "target_tokens": 16384, "requests": [ { "ok": true, "err": null, "ttft_s": 8.386, "total_s": 8.819, "prompt_tokens": 16058, "reused_tokens": 0, "read_tokens": 16058, "generated_tokens": 24, "prompt_tps": 1920.7, "prompt_tps_source": "engine", "decode_tps": 53.8, "decode_tps_source": "engine", "prompt_ms": 8360.5, "decode_ms": 446.3, "drafts_offered": 25, "drafts_accepted": 10, "run": 0 }, { "ok": true, "err": null, "ttft_s": 8.356, "total_s": 8.73, "prompt_tokens": 16058, "reused_tokens": 0, "read_tokens": 16058, "generated_tokens": 24, "prompt_tps": 1927.2, "prompt_tps_source": "engine", "decode_tps": 62.2, "decode_tps_source": "engine", "prompt_ms": 8332.5, "decode_ms": 385.9, "drafts_offered": 20, "drafts_accepted": 10, "run": 1 }, { "ok": true, "err": null, "ttft_s": 8.338, "total_s": 8.665, "prompt_tokens": 16058, "reused_tokens": 0, "read_tokens": 16058, "generated_tokens": 24, "prompt_tps": 1931.4, "prompt_tps_source": "engine", "decode_tps": 70.6, "decode_tps_source": "engine", "prompt_ms": 8314.1, "decode_ms": 339.8, "drafts_offered": 25, "drafts_accepted": 8, "run": 2 } ], "prompt_tps": { "median": 1927.2, "min": 1920.7, "max": 1931.4, "n": 3 }, "ttft_s": { "median": 8.356, "min": 8.338, "max": 8.386, "n": 3 }, "prompt_tokens": { "median": 16058, "min": 16058, "max": 16058, "n": 3 }, "reused_tokens": { "median": 0, "min": 0, "max": 0, "n": 3 } }, "32K": { "target_tokens": 32768, "requests": [ { "ok": true, "err": null, "ttft_s": 16.613, "total_s": 17.106, "prompt_tokens": 32049, "reused_tokens": 0, "read_tokens": 32049, "generated_tokens": 24, "prompt_tps": 1934.1, "prompt_tps_source": "engine", "decode_tps": 46.7, "decode_tps_source": "engine", "prompt_ms": 16570.7, "decode_ms": 514.4, "drafts_offered": 16, "drafts_accepted": 9, "run": 0 }, { "ok": true, "err": null, "ttft_s": 16.56, "total_s": 16.997, "prompt_tokens": 32049, "reused_tokens": 0, "read_tokens": 32049, "generated_tokens": 24, "prompt_tps": 1939.8, "prompt_tps_source": "engine", "decode_tps": 52.8, "decode_tps_source": "engine", "prompt_ms": 16521.8, "decode_ms": 454.9, "drafts_offered": 26, "drafts_accepted": 9, "run": 1 }, { "ok": true, "err": null, "ttft_s": 16.555, "total_s": 16.926, "prompt_tokens": 32049, "reused_tokens": 0, "read_tokens": 32049, "generated_tokens": 24, "prompt_tps": 1940.2, "prompt_tps_source": "engine", "decode_tps": 62.0, "decode_tps_source": "engine", "prompt_ms": 16518.2, "decode_ms": 387.3, "drafts_offered": 19, "drafts_accepted": 9, "run": 2 } ], "prompt_tps": { "median": 1939.8, "min": 1934.1, "max": 1940.2, "n": 3 }, "ttft_s": { "median": 16.56, "min": 16.555, "max": 16.613, "n": 3 }, "prompt_tokens": { "median": 32049, "min": 32049, "max": 32049, "n": 3 }, "reused_tokens": { "median": 0, "min": 0, "max": 0, "n": 3 } } }, "agent": { "sessions": [ [ { "ok": true, "err": null, "ttft_s": 1.481, "total_s": 1.773, "prompt_tokens": 609, "reused_tokens": 0, "read_tokens": 609, "generated_tokens": 13, "prompt_tps": 416.5, "prompt_tps_source": "engine", "decode_tps": 42.5, "decode_tps_source": "engine", "prompt_ms": 1462.3, "decode_ms": 306.2, "drafts_offered": 18, "drafts_accepted": 9, "run": 0, "turn": 0 }, { "ok": true, "err": null, "ttft_s": 3.361, "total_s": 4.263, "prompt_tokens": 3704, "reused_tokens": 624, "read_tokens": 3080, "generated_tokens": 64, "prompt_tps": 922.8, "prompt_tps_source": "engine", "decode_tps": 69.9, "decode_tps_source": "engine", "prompt_ms": 3337.7, "decode_ms": 915.6, "drafts_offered": 51, "drafts_accepted": 43, "run": 0, "turn": 1 }, { "ok": true, "err": null, "ttft_s": 2.982, "total_s": 3.864, "prompt_tokens": 5237, "reused_tokens": 3767, "read_tokens": 1470, "generated_tokens": 64, "prompt_tps": 496.7, "prompt_tps_source": "engine", "decode_tps": 71.6, "decode_tps_source": "engine", "prompt_ms": 2959.8, "decode_ms": 894.3, "drafts_offered": 58, "drafts_accepted": 42, "run": 0, "turn": 2 }, { "ok": true, "err": null, "ttft_s": 3.371, "total_s": 4.123, "prompt_tokens": 8681, "reused_tokens": 5301, "read_tokens": 3380, "generated_tokens": 55, "prompt_tps": 1009.3, "prompt_tps_source": "engine", "decode_tps": 72.0, "decode_tps_source": "engine", "prompt_ms": 3348.8, "decode_ms": 764.1, "drafts_offered": 52, "drafts_accepted": 31, "run": 0, "turn": 3 }, { "ok": true, "err": null, "ttft_s": 3.274, "total_s": 3.509, "prompt_tokens": 11595, "reused_tokens": 8735, "read_tokens": 2860, "generated_tokens": 18, "prompt_tps": 880.2, "prompt_tps_source": "engine", "decode_tps": 72.8, "decode_tps_source": "engine", "prompt_ms": 3249.4, "decode_ms": 247.4, "drafts_offered": 18, "drafts_accepted": 10, "run": 0, "turn": 4 }, { "ok": true, "err": null, "ttft_s": 3.195, "total_s": 3.47, "prompt_tokens": 14458, "reused_tokens": 11612, "read_tokens": 2846, "generated_tokens": 12, "prompt_tps": 897.7, "prompt_tps_source": "engine", "decode_tps": 41.9, "decode_tps_source": "engine", "prompt_ms": 3170.4, "decode_ms": 286.6, "drafts_offered": 23, "drafts_accepted": 3, "run": 0, "turn": 5 } ], [ { "ok": true, "err": null, "ttft_s": 1.467, "total_s": 1.717, "prompt_tokens": 609, "reused_tokens": 0, "read_tokens": 609, "generated_tokens": 13, "prompt_tps": 419.6, "prompt_tps_source": "engine", "decode_tps": 49.5, "decode_tps_source": "engine", "prompt_ms": 1451.4, "decode_ms": 262.4, "drafts_offered": 15, "drafts_accepted": 9, "run": 1, "turn": 0 }, { "ok": true, "err": null, "ttft_s": 3.116, "total_s": 3.851, "prompt_tokens": 3704, "reused_tokens": 624, "read_tokens": 3080, "generated_tokens": 64, "prompt_tps": 994.2, "prompt_tps_source": "engine", "decode_tps": 85.6, "decode_tps_source": "engine", "prompt_ms": 3097.9, "decode_ms": 747.9, "drafts_offered": 56, "drafts_accepted": 42, "run": 1, "turn": 1 }, { "ok": true, "err": null, "ttft_s": 2.898, "total_s": 3.915, "prompt_tokens": 5237, "reused_tokens": 3767, "read_tokens": 1470, "generated_tokens": 64, "prompt_tps": 510.3, "prompt_tps_source": "engine", "decode_tps": 62.2, "decode_tps_source": "engine", "prompt_ms": 2880.4, "decode_ms": 1029.1, "drafts_offered": 78, "drafts_accepted": 37, "run": 1, "turn": 2 }, { "ok": true, "err": null, "ttft_s": 3.22, "total_s": 3.987, "prompt_tokens": 8681, "reused_tokens": 5230, "read_tokens": 3451, "generated_tokens": 64, "prompt_tps": 1078.6, "prompt_tps_source": "engine", "decode_tps": 82.2, "decode_tps_source": "engine", "prompt_ms": 3199.5, "decode_ms": 779.0, "drafts_offered": 59, "drafts_accepted": 38, "run": 1, "turn": 3 }, { "ok": true, "err": null, "ttft_s": 3.114, "total_s": 3.739, "prompt_tokens": 11604, "reused_tokens": 8744, "read_tokens": 2860, "generated_tokens": 49, "prompt_tps": 925.0, "prompt_tps_source": "engine", "decode_tps": 76.9, "decode_tps_source": "engine", "prompt_ms": 3091.8, "decode_ms": 636.8, "drafts_offered": 48, "drafts_accepted": 29, "run": 1, "turn": 4 }, { "ok": true, "err": null, "ttft_s": 3.124, "total_s": 3.409, "prompt_tokens": 14498, "reused_tokens": 11652, "read_tokens": 2846, "generated_tokens": 12, "prompt_tps": 917.8, "prompt_tps_source": "engine", "decode_tps": 40.4, "decode_tps_source": "engine", "prompt_ms": 3100.9, "decode_ms": 296.9, "drafts_offered": 23, "drafts_accepted": 3, "run": 1, "turn": 5 } ], [ { "ok": true, "err": null, "ttft_s": 1.466, "total_s": 1.829, "prompt_tokens": 609, "reused_tokens": 0, "read_tokens": 609, "generated_tokens": 24, "prompt_tps": 419.9, "prompt_tps_source": "engine", "decode_tps": 63.8, "decode_tps_source": "engine", "prompt_ms": 1450.2, "decode_ms": 376.4, "drafts_offered": 20, "drafts_accepted": 14, "run": 2, "turn": 0 }, { "ok": true, "err": null, "ttft_s": 3.12, "total_s": 3.831, "prompt_tokens": 3715, "reused_tokens": 632, "read_tokens": 3083, "generated_tokens": 64, "prompt_tps": 993.6, "prompt_tps_source": "engine", "decode_tps": 88.5, "decode_tps_source": "engine", "prompt_ms": 3102.9, "decode_ms": 723.4, "drafts_offered": 56, "drafts_accepted": 44, "run": 2, "turn": 1 }, { "ok": true, "err": null, "ttft_s": 2.933, "total_s": 3.801, "prompt_tokens": 5248, "reused_tokens": 3708, "read_tokens": 1540, "generated_tokens": 64, "prompt_tps": 528.3, "prompt_tps_source": "engine", "decode_tps": 72.7, "decode_tps_source": "engine", "prompt_ms": 2915.2, "decode_ms": 879.9, "drafts_offered": 69, "drafts_accepted": 40, "run": 2, "turn": 2 }, { "ok": true, "err": null, "ttft_s": 3.222, "total_s": 4.036, "prompt_tokens": 8692, "reused_tokens": 5241, "read_tokens": 3451, "generated_tokens": 64, "prompt_tps": 1077.9, "prompt_tps_source": "engine", "decode_tps": 77.5, "decode_tps_source": "engine", "prompt_ms": 3201.7, "decode_ms": 825.8, "drafts_offered": 63, "drafts_accepted": 39, "run": 2, "turn": 3 }, { "ok": true, "err": null, "ttft_s": 3.141, "total_s": 3.716, "prompt_tokens": 11616, "reused_tokens": 8685, "read_tokens": 2931, "generated_tokens": 48, "prompt_tps": 939.8, "prompt_tps_source": "engine", "decode_tps": 81.8, "decode_tps_source": "engine", "prompt_ms": 3118.9, "decode_ms": 586.9, "drafts_offered": 49, "drafts_accepted": 29, "run": 2, "turn": 4 }, { "ok": true, "err": null, "ttft_s": 3.129, "total_s": 3.352, "prompt_tokens": 14509, "reused_tokens": 11663, "read_tokens": 2846, "generated_tokens": 14, "prompt_tps": 916.4, "prompt_tps_source": "engine", "decode_tps": 59.5, "decode_tps_source": "engine", "prompt_ms": 3105.7, "decode_ms": 235.4, "drafts_offered": 19, "drafts_accepted": 6, "run": 2, "turn": 5 } ] ], "hit_pct": { "median": 68.651, "min": 68.362, "max": 68.778, "n": 3 }, "reread_tokens": { "median": 74, "min": 3, "max": 215, "n": 3 }, "new_tokens": { "median": 13633, "min": 13633, "max": 13636, "n": 3 }, "turn_ttft_s": { "median": 3.129, "min": 2.898, "max": 3.371, "n": 15 }, "last_turn_prompt_tokens": { "median": 14498, "min": 14458, "max": 14509, "n": 3 }, "turn_decode_tps": { "median": 72.7, "min": 40.4, "max": 88.5, "n": 15 }, "first_turn_ttft_s": { "median": 1.467, "min": 1.466, "max": 1.481, "n": 3 } }, "clients": { "clients": 2, "rounds": [ { "round": 0, "wall_s": 5.95, "tokens": 256, "total_tps": 43.01, "first_s": [ 0.8996901249993243, 4.124765323998872 ], "latency_s": [ 3.424592978999499, 5.951380917000279 ], "errors": [], "error_count": 0 }, { "round": 1, "wall_s": 5.18, "tokens": 256, "total_tps": 49.4, "first_s": [ 0.6685062090000429, 3.0936049499996443 ], "latency_s": [ 2.3363291189998563, 5.181394266999632 ], "errors": [], "error_count": 0 }, { "round": 2, "wall_s": 5.26, "tokens": 256, "total_tps": 48.71, "first_s": [ 0.7017926609987626, 3.1312591320001957 ], "latency_s": [ 2.4025190839984134, 5.254824739000469 ], "errors": [], "error_count": 0 } ], "total_tps": { "median": 48.71, "min": 43.01, "max": 49.4, "n": 3 }, "ttft_p50_s": 1.997, "ttft_p95_s": 3.876, "latency_p50_s": 4.303, "latency_p95_s": 5.777, "errors": 0 }, "skipped": {}, "failures": [] } }, "offline": { "status": "passed", "started_utc": "2026-10-12T03:01:45.150401+00:00", "finished_utc": "2026-10-12T03:02:59.664752+00:00", "network_namespace_interfaces": [ "lo" ], "external_tcp_connection_blocked": true, "expected_marker_present": true, "inference_wall_s": 0.4813707830016938 }, "vram_gib": 15.1640625, "container_memory_peak_gib": 42.065467834472656, "resource_scope": "one-second observed peak during official standard suite; Docker charged memory includes cache, not host total RAM", "historical_eligibility_note": "Old daily_driver_eligible needs llama.cpp adaptive probes, which do not apply to Strata. It is not used as a new suitability verdict.", "workload_metrics": { "practical": { "responses": 72, "end_to_end_s": { "n": 72, "median": 8.1632, "min": 3.0513, "max": 37.9985 }, "engine_decode_tps": { "n": 72, "median": 74.45, "min": 54.5, "max": 93.7 }, "engine_uncached_prefill_tps": { "n": 72, "median": 183.15, "min": 72.3, "max": 1930.2 }, "drafts_offered": 30319, "drafts_accepted": 24127, "draft_acceptance_fraction": 0.7957716283518586, "total_prompt_tokens": 412752, "reused_prompt_tokens": 0, "prefix_reuse_fraction": 0.0, "client_ttft_s": null, "note": "Original clients are nonstreaming; engine decode/prefill and client end-to-end latency are distinct. Draft counters include speculation, not a separately isolated MTP-only rate." }, "humaneval": { "responses": 164, "end_to_end_s": { "n": 164, "median": 6.8298000000000005, "min": 1.5699, "max": 15.1303 }, "engine_decode_tps": { "n": 164, "median": 73.1, "min": 60.8, "max": 90.8 }, "engine_uncached_prefill_tps": { "n": 164, "median": 209.55, "min": 82.5, "max": 358.8 }, "drafts_offered": 56742, "drafts_accepted": 44682, "draft_acceptance_fraction": 0.7874590250608016, "total_prompt_tokens": 34439, "reused_prompt_tokens": 0, "prefix_reuse_fraction": 0.0, "client_ttft_s": null, "note": "Original clients are nonstreaming; engine decode/prefill and client end-to-end latency are distinct. Draft counters include speculation, not a separately isolated MTP-only rate." } }, "coding_failures": [ { "task_id": "HumanEval/10", "base_status": "fail", "plus_status": "fail" }, { "task_id": "HumanEval/22", "base_status": "pass", "plus_status": "fail" }, { "task_id": "HumanEval/32", "base_status": "fail", "plus_status": "fail" }, { "task_id": "HumanEval/76", "base_status": "pass", "plus_status": "fail" }, { "task_id": "HumanEval/91", "base_status": "pass", "plus_status": "fail" }, { "task_id": "HumanEval/99", "base_status": "pass", "plus_status": "fail" }, { "task_id": "HumanEval/132", "base_status": "fail", "plus_status": "fail" }, { "task_id": "HumanEval/141", "base_status": "pass", "plus_status": "fail" }, { "task_id": "HumanEval/145", "base_status": "fail", "plus_status": "fail" }, { "task_id": "HumanEval/151", "base_status": "pass", "plus_status": "fail" } ], "practical_missed_checks": [ { "task_id": "business-margin-forecast", "repetition": 0, "category": "business", "score": 62.5, "finish_reason": "stop", "failed_checks": [ "1,680,000", "70%", "20%" ] }, { "task_id": "research-evidence-table", "repetition": 0, "category": "research", "score": 90.0, "finish_reason": "stop", "failed_checks": [ "forum" ] }, { "task_id": "agent-tool-sequence", "repetition": 0, "category": "agent_tools", "score": 62.5, "finish_reason": "stop", "failed_checks": [ "search query supplied", "read URL is search placeholder", "save includes title, body and source" ] }, { "task_id": "sysadmin-recovery", "repetition": 0, "category": "practical_sysadmin", "score": 80.0, "finish_reason": "length", "failed_checks": [ "no write/truncate file mode", "API response completed" ] }, { "task_id": "executive-decision-memo", "repetition": 0, "category": "executive_content", "score": 85.71, "finish_reason": "stop", "failed_checks": [ "$100" ] }, { "task_id": "executive-content-brief", "repetition": 0, "category": "executive_content", "score": 90.91, "finish_reason": "stop", "failed_checks": [ "three evidence beats" ] }, { "task_id": "research-source-conflict", "repetition": 1, "category": "research", "score": 90.0, "finish_reason": "stop", "failed_checks": [ "superseded" ] }, { "task_id": "research-vendor-filter", "repetition": 1, "category": "research", "score": 90.91, "finish_reason": "stop", "failed_checks": [ "$625" ] }, { "task_id": "sysadmin-python-bug", "repetition": 1, "category": "practical_sysadmin", "score": 66.67, "finish_reason": "stop", "failed_checks": [ "complete average function with empty guard and ordinary calculation" ] }, { "task_id": "sysadmin-recovery", "repetition": 1, "category": "practical_sysadmin", "score": 80.0, "finish_reason": "length", "failed_checks": [ "API response completed" ] }, { "task_id": "executive-decision-memo", "repetition": 1, "category": "executive_content", "score": 85.71, "finish_reason": "stop", "failed_checks": [ "$100" ] }, { "task_id": "executive-waiting-on", "repetition": 1, "category": "executive_content", "score": 88.89, "finish_reason": "stop", "failed_checks": [ "approve" ] }, { "task_id": "agent-tool-sequence", "repetition": 2, "category": "agent_tools", "score": 62.5, "finish_reason": "stop", "failed_checks": [ "search query supplied", "read URL is search placeholder", "save includes title, body and source" ] }, { "task_id": "agent-parallel-plan", "repetition": 2, "category": "agent_tools", "score": 90.0, "finish_reason": "stop", "failed_checks": [ "parallel independent source branches" ] }, { "task_id": "sysadmin-recovery", "repetition": 2, "category": "practical_sysadmin", "score": 80.0, "finish_reason": "length", "failed_checks": [ "no write/truncate file mode", "API response completed" ] }, { "task_id": "executive-decision-memo", "repetition": 2, "category": "executive_content", "score": 85.71, "finish_reason": "stop", "failed_checks": [ "$100" ] }, { "task_id": "executive-waiting-on", "repetition": 2, "category": "executive_content", "score": 88.89, "finish_reason": "stop", "failed_checks": [ "approve" ] } ], "api_diagnostics": { "status": "completed_diagnostics_not_a_quality_benchmark", "model": "qwen3.8-flash-next-iq3_xxs", "tool_call_semantically_valid": true, "structured_output_semantically_valid": true, "streaming_final_content_seen": true }, "container_stop_states": [ { "context": 131072, "image": "sha256:f759e6167315dac77292d394a470240493dad78de6d23dd270eaa5c7ffc3e143", "status": "exited", "oom_killed": false, "exit_code": 0 }, { "context": 32768, "image": "sha256:f759e6167315dac77292d394a470240493dad78de6d23dd270eaa5c7ffc3e143", "status": "exited", "oom_killed": false, "exit_code": 0 }, { "context": 65536, "image": "sha256:f759e6167315dac77292d394a470240493dad78de6d23dd270eaa5c7ffc3e143", "status": "exited", "oom_killed": false, "exit_code": 0 } ] } ], "rankings": [ "strata-flash-next-iq2_xs", "strata-flash-next-iq3_xxs" ], "role_recommendations": { "CFO / Analyst": { "score": 98.12, "winners": [ "strata-flash-next-iq2_xs" ] }, "Researcher / Scout": { "score": 97.04, "winners": [ "strata-flash-next-iq3_xxs" ] }, "Hermes Coordinator": { "score": 95.77, "winners": [ "strata-flash-next-iq3_xxs" ] }, "Librarian / Knowledge": { "score": 96.89, "winners": [ "strata-flash-next-iq3_xxs" ] }, "Forge / Coder": { "score": 94.42, "winners": [ "strata-flash-next-iq2_xs" ] } }, "caveats": [ "IQ2_XS vs IQ3_XXS is a same-model, same-hardware comparison. Historical report4 uses Qwen3.8-27B and a different runtime; cross-model differences cannot be attributed solely to quantization.", "Both quantizations use mixed tensor formats and share the verified PLE shard; the labels do not mean every tensor uses two or three bits.", "Model preparation warmed filesystem cache. No storage-cold performance is claimed. Model load, initial uncached prompt processing and prefix-cached follow-ups are separate measurements.", "Dossier repeats retain the existing benchmark prompts and cache policy. First response and warmed repeats are reported separately; cached prefill rates must not be used as full-input throughput.", "The original v2.1 practical rubric and weighted score are retained. Hermes literal-marker heuristic is reported separately, with empty finals explicitly zeroed; it is not an independent broad-quality assessment.", "The practical rubric checks explicit content and structure, not arbitrary semantic correctness. These fixed synthetic workload scores are not a general intelligence or production-agent guarantee.", "The practical/coding index excludes long-context quality. Empty final answers count as dossier failures and receive zero validated dossier quality, even if reasoning exists or the API reports stop. Successful-speed rankings require valid completed answers. A separate full-marker reliability gate does not alter the original weighted index.", "HumanEval+ uses one generation per problem (pass@1); practical tasks have three repeats. Small score differences have not been established as statistically significant or universal quantization effects.", "Strict JSON-schema enforcement is unavailable in the tested image because an optional dependency is absent. Successful JSON examples and forced client tool calls do not prove general schema/tool reliability.", "Server-side MCP execution and vision are disabled. The optional two-client speed test uses one serial service slot, not GPU batching.", "Nonstreaming original Sonny clients measure end-to-end latency and retain engine timings; client TTFT is measured by the upstream streaming suite, not inferred for every Sonny request.", "One-second telemetry can miss subsecond peaks. Host physical-device disk traffic is not model-specific. Docker charged-memory, process RSS, page cache and host memory have distinct definitions.", "The engine retains upstream route-tail-skip default 7. Missing experts may be skipped, so speed/quality results include that approximation. Greedy output can vary with draft grouping, adaptive expert placement and prefix-cache state; fixed seed/temperature do not promise byte-identical responses.", "Apt repositories and some rebuilt scorer transitive dependencies are not snapshot-pinned. Source/image/model hashes and package inventories improve auditability but do not constitute a security certification." ], "telemetry": [ { "quant": "IQ2_XS", "phase": "practical", "scope": "docker_container", "samples": 593, "start_utc": "2026-10-12T02:29:44.813589+00:00", "end_utc": "2026-10-12T02:39:36.813621+00:00", "sample_interval_s": { "n": 592, "median": 0.9999998504990799, "min": 0.9981299020000733, "max": 1.0018555110000307 }, "collector_duration_s": { "n": 593, "median": 0.04263217699917732, "min": 0.03559588800089841, "max": 0.050985259000299266 }, "gpu_utilization_pct": { "n": 593, "median": 93.0, "min": 31.0, "max": 100.0 }, "vram_used_mib": { "n": 593, "median": 15494.0, "min": 15494.0, "max": 15494.0 }, "gpu_power_w": { "n": 593, "median": 114.56, "min": 12.35, "max": 155.87 }, "gpu_temperature_c": { "n": 593, "median": 45.0, "min": 40.0, "max": 55.0 }, "cpu_utilization_pct": { "n": 593, "median": 58.47542627883651, "min": 10.473313192346424, "max": 69.51710261569417 }, "proc_memory_available_gib": { "n": 593, "median": 20.22526264190674, "min": 19.892364501953125, "max": 20.344162940979004 }, "proc_memory_total_minus_available_gib": { "n": 593, "median": 31.77473735809326, "min": 31.655837059020996, "max": 32.107635498046875 }, "proc_page_cache_cached_gib": { "n": 593, "median": 45.95822525024414, "min": 45.95111083984375, "max": 45.96571350097656 }, "proc_shmem_gib": { "n": 593, "median": 29.22247314453125, "min": 29.22247314453125, "max": 29.222476959228516 }, "proc_cached_minus_shmem_estimate_gib": { "n": 593, "median": 16.73575210571289, "min": 16.7286376953125, "max": 16.743240356445312 }, "proc_unevictable_gib": { "n": 593, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 593, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 593, "median": 54.03515625, "min": 54.03515625, "max": 54.03515625 }, "pswpin_counter_delta": 55, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 61, "container_memory_current_gib": { "n": 593, "median": 36.221256256103516, "min": 36.103759765625, "max": 36.55194091796875 }, "container_memory_lifetime_peak_gib": { "n": 593, "median": 36.9876708984375, "min": 36.9876708984375, "max": 36.9876708984375 }, "container_swap_mib": { "n": 593, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 593, "median": 2.3174171447753906, "min": 2.200916290283203, "max": 2.6469764709472656 }, "container_memory_stat_file_gib": { "n": 593, "median": 33.762699127197266, "min": 33.76254653930664, "max": 33.76274108886719 }, "container_memory_stat_shmem_gib": { "n": 593, "median": 29.222198486328125, "min": 29.222198486328125, "max": 29.222198486328125 }, "container_memory_stat_file_mapped_gib": { "n": 593, "median": 29.173667907714844, "min": 29.173648834228516, "max": 29.173667907714844 }, "container_memory_stat_unevictable_gib": { "n": 593, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 928047104, "read_mib_per_s": { "n": 592, "median": 0.7109377978138487, "min": 0.03515613665666299, "max": 8.316465978869925 }, "write_bytes": 139264, "write_mib_per_s": { "n": 592, "median": 0.0, "min": 0.0, "max": 0.011718911111591469 } }, "252:3": { "read_bytes": 928047104, "read_mib_per_s": { "n": 592, "median": 0.7109377978138487, "min": 0.03515613665666299, "max": 8.316465978869925 }, "write_bytes": 139264, "write_mib_per_s": { "n": 592, "median": 0.0, "min": 0.0, "max": 0.011718911111591469 } }, "252:6": { "read_bytes": 928047104, "read_mib_per_s": { "n": 592, "median": 0.7109377978138487, "min": 0.03515613665666299, "max": 8.316465978869925 }, "write_bytes": 139264, "write_mib_per_s": { "n": 592, "median": 0.0, "min": 0.0, "max": 0.011718911111591469 } }, "259:0": { "read_bytes": 928047104, "read_mib_per_s": { "n": 592, "median": 0.7109377978138487, "min": 0.03515613665666299, "max": 8.316465978869925 }, "write_bytes": 139264, "write_mib_per_s": { "n": 592, "median": 0.0, "min": 0.0, "max": 0.011718911111591469 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 593, "median": 32975776, "min": 32857000, "max": 33320524 }, "VmHWM": { "n": 593, "median": 33514628, "min": 33514628, "max": 33514628 }, "VmSwap": { "n": 593, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "26431": { "read_bytes": 928047104, "write_bytes": 147456, "rchar": 930086855, "wchar": 323261 } } }, "python": { "VmRSS": { "n": 593, "median": 222652, "min": 219168, "max": 223412 }, "VmHWM": { "n": 593, "median": 223144, "min": 219168, "max": 223412 }, "VmSwap": { "n": 593, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "26259": { "read_bytes": 0, "write_bytes": 0, "rchar": 3917790, "wchar": 1908004 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "practical", "scope": "host", "samples": 593, "start_utc": "2026-10-12T02:29:44.697259+00:00", "end_utc": "2026-10-12T02:39:36.697308+00:00", "sample_interval_s": { "n": 592, "median": 1.0000002899996616, "min": 0.9997580399995059, "max": 1.000241759000346 }, "collector_duration_s": { "n": 593, "median": 0.05327611400025489, "min": 0.0422389720006322, "max": 0.06179366600008507 }, "gpu_utilization_pct": { "n": 593, "median": 93.0, "min": 31.0, "max": 100.0 }, "vram_used_mib": { "n": 593, "median": 15494.0, "min": 15494.0, "max": 15494.0 }, "gpu_power_w": { "n": 593, "median": 114.56, "min": 12.35, "max": 155.87 }, "gpu_temperature_c": { "n": 593, "median": 45.0, "min": 40.0, "max": 55.0 }, "cpu_utilization_pct": { "n": 593, "median": 48.9983305509182, "min": 9.174311926605505, "max": 58.50796311818944 }, "proc_memory_available_gib": { "n": 593, "median": 23.128860473632812, "min": 22.556529998779297, "max": 23.51280975341797 }, "proc_memory_total_minus_available_gib": { "n": 593, "median": 35.62387466430664, "min": 35.239925384521484, "max": 36.196205139160156 }, "proc_page_cache_cached_gib": { "n": 593, "median": 46.22208786010742, "min": 46.21182632446289, "max": 46.23944854736328 }, "proc_shmem_gib": { "n": 593, "median": 29.288585662841797, "min": 29.288585662841797, "max": 29.28866195678711 }, "proc_cached_minus_shmem_estimate_gib": { "n": 593, "median": 16.933502197265625, "min": 16.923240661621094, "max": 16.95085906982422 }, "proc_unevictable_gib": { "n": 593, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 593, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 593, "median": 54.94140625, "min": 54.94140625, "max": 54.94140625 }, "pswpin_counter_delta": 55, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 61, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 593, "median": 32975776, "min": 32857000, "max": 33320524 }, "VmHWM": { "n": 593, "median": 33514628, "min": 33514628, "max": 33514628 }, "VmSwap": { "n": 593, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65735": { "read_bytes": 927907840, "write_bytes": 147456, "rchar": 929947591, "wchar": 323206 } } }, "python": { "VmRSS": { "n": 593, "median": 222652, "min": 219168, "max": 223412 }, "VmHWM": { "n": 593, "median": 223144, "min": 219168, "max": 223412 }, "VmSwap": { "n": 593, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65561": { "read_bytes": 0, "write_bytes": 0, "rchar": 3918506, "wchar": 1908004 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 983830528, "read_mib_per_s": { "n": 592, "median": 0.8320249528656039, "min": 0.03515619069151001, "max": 7.898264204189585 }, "write_bytes": 69287936, "write_mib_per_s": { "n": 592, "median": 0.015625139516852606, "min": 0, "max": 2.05861745086389 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "dossier", "scope": "docker_container", "samples": 97, "start_utc": "2026-10-12T02:39:37.813821+00:00", "end_utc": "2026-10-12T02:41:13.813757+00:00", "sample_interval_s": { "n": 96, "median": 0.999999900000148, "min": 0.9996563509994303, "max": 1.0003204709992133 }, "collector_duration_s": { "n": 97, "median": 0.042779547000463936, "min": 0.037501438999242964, "max": 0.04852384599871584 }, "gpu_utilization_pct": { "n": 97, "median": 93.0, "min": 64.0, "max": 100.0 }, "vram_used_mib": { "n": 97, "median": 15494.0, "min": 15494.0, "max": 15494.0 }, "gpu_power_w": { "n": 97, "median": 117.99, "min": 62.01, "max": 151.03 }, "gpu_temperature_c": { "n": 97, "median": 45.0, "min": 42.0, "max": 49.0 }, "cpu_utilization_pct": { "n": 97, "median": 58.93574297188755, "min": 11.055276381909549, "max": 66.46646646646647 }, "proc_memory_available_gib": { "n": 97, "median": 19.757963180541992, "min": 19.752820014953613, "max": 20.19429111480713 }, "proc_memory_total_minus_available_gib": { "n": 97, "median": 32.24203681945801, "min": 31.80570888519287, "max": 32.24717998504639 }, "proc_page_cache_cached_gib": { "n": 97, "median": 45.96843719482422, "min": 45.96688461303711, "max": 45.969482421875 }, "proc_shmem_gib": { "n": 97, "median": 29.22247314453125, "min": 29.22247314453125, "max": 29.22247314453125 }, "proc_cached_minus_shmem_estimate_gib": { "n": 97, "median": 16.74596405029297, "min": 16.74441146850586, "max": 16.74700927734375 }, "proc_unevictable_gib": { "n": 97, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 97, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 97, "median": 54.03515625, "min": 54.03515625, "max": 54.03515625 }, "pswpin_counter_delta": 3, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 5, "container_memory_current_gib": { "n": 97, "median": 36.55195999145508, "min": 36.11286163330078, "max": 36.55254364013672 }, "container_memory_lifetime_peak_gib": { "n": 97, "median": 36.9876708984375, "min": 36.9876708984375, "max": 36.9876708984375 }, "container_swap_mib": { "n": 97, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 97, "median": 2.647663116455078, "min": 2.2090988159179688, "max": 2.6480445861816406 }, "container_memory_stat_file_gib": { "n": 97, "median": 33.76274108886719, "min": 33.76274108886719, "max": 33.76274490356445 }, "container_memory_stat_shmem_gib": { "n": 97, "median": 29.222198486328125, "min": 29.222198486328125, "max": 29.222198486328125 }, "container_memory_stat_file_mapped_gib": { "n": 97, "median": 29.173667907714844, "min": 29.173667907714844, "max": 29.173667907714844 }, "container_memory_stat_unevictable_gib": { "n": 97, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 216723456, "read_mib_per_s": { "n": 96, "median": 1.8261519324504212, "min": 0.03906205375505654, "max": 16.173526766308928 }, "write_bytes": 12288, "write_mib_per_s": { "n": 96, "median": 0.0, "min": 0.0, "max": 0.007812477867253339 } }, "252:3": { "read_bytes": 216723456, "read_mib_per_s": { "n": 96, "median": 1.8261519324504212, "min": 0.03906205375505654, "max": 16.173526766308928 }, "write_bytes": 12288, "write_mib_per_s": { "n": 96, "median": 0.0, "min": 0.0, "max": 0.007812477867253339 } }, "252:6": { "read_bytes": 216723456, "read_mib_per_s": { "n": 96, "median": 1.8261519324504212, "min": 0.03906205375505654, "max": 16.173526766308928 }, "write_bytes": 12288, "write_mib_per_s": { "n": 96, "median": 0.0, "min": 0.0, "max": 0.007812477867253339 } }, "259:0": { "read_bytes": 216723456, "read_mib_per_s": { "n": 96, "median": 1.8261519324504212, "min": 0.03906205375505654, "max": 16.173526766308928 }, "write_bytes": 12288, "write_mib_per_s": { "n": 96, "median": 0.0, "min": 0.0, "max": 0.007812477867253339 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 97, "median": 33320596, "min": 32860860, "max": 33320596 }, "VmHWM": { "n": 97, "median": 33514628, "min": 33514628, "max": 33514628 }, "VmSwap": { "n": 97, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "26431": { "read_bytes": 216723456, "write_bytes": 16384, "rchar": 218272854, "wchar": 48463 } } }, "python": { "VmRSS": { "n": 97, "median": 224060, "min": 222548, "max": 224460 }, "VmHWM": { "n": 97, "median": 224060, "min": 223144, "max": 224460 }, "VmSwap": { "n": 97, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "26259": { "read_bytes": 0, "write_bytes": 0, "rchar": 627695, "wchar": 1509605 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "dossier", "scope": "host", "samples": 97, "start_utc": "2026-10-12T02:39:37.697266+00:00", "end_utc": "2026-10-12T02:41:13.697308+00:00", "sample_interval_s": { "n": 96, "median": 1.0000012970003809, "min": 0.9999781090009492, "max": 1.0000186180004675 }, "collector_duration_s": { "n": 97, "median": 0.05417378799938888, "min": 0.0434506510009669, "max": 0.061839837000661646 }, "gpu_utilization_pct": { "n": 97, "median": 93.0, "min": 64.0, "max": 100.0 }, "vram_used_mib": { "n": 97, "median": 15494.0, "min": 15494.0, "max": 15494.0 }, "gpu_power_w": { "n": 97, "median": 117.99, "min": 62.01, "max": 151.03 }, "gpu_temperature_c": { "n": 97, "median": 45.0, "min": 42.0, "max": 49.0 }, "cpu_utilization_pct": { "n": 97, "median": 49.3734335839599, "min": 9.273182957393484, "max": 55.04587155963303 }, "proc_memory_available_gib": { "n": 97, "median": 22.35812759399414, "min": 22.059368133544922, "max": 23.015029907226562 }, "proc_memory_total_minus_available_gib": { "n": 97, "median": 36.39460754394531, "min": 35.73770523071289, "max": 36.69336700439453 }, "proc_page_cache_cached_gib": { "n": 97, "median": 46.242469787597656, "min": 46.24027633666992, "max": 46.243995666503906 }, "proc_shmem_gib": { "n": 97, "median": 29.288589477539062, "min": 29.28841781616211, "max": 29.288650512695312 }, "proc_cached_minus_shmem_estimate_gib": { "n": 97, "median": 16.953880310058594, "min": 16.95169448852539, "max": 16.955406188964844 }, "proc_unevictable_gib": { "n": 97, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 97, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 97, "median": 54.94140625, "min": 54.94140625, "max": 54.94140625 }, "pswpin_counter_delta": 3, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 5, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 97, "median": 33320596, "min": 32860932, "max": 33320596 }, "VmHWM": { "n": 97, "median": 33514628, "min": 33514628, "max": 33514628 }, "VmSwap": { "n": 97, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65735": { "read_bytes": 216723456, "write_bytes": 16384, "rchar": 218272854, "wchar": 48396 } } }, "python": { "VmRSS": { "n": 97, "median": 224060, "min": 222548, "max": 224460 }, "VmHWM": { "n": 97, "median": 224060, "min": 223144, "max": 224460 }, "VmSwap": { "n": 97, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65561": { "read_bytes": 0, "write_bytes": 0, "rchar": 633675, "wchar": 1509605 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 224346112, "read_mib_per_s": { "n": 96, "median": 1.8574289417747323, "min": 0.039062181955702914, "max": 16.382867923244618 }, "write_bytes": 9375744, "write_mib_per_s": { "n": 96, "median": 0.0039062996355471655, "min": 0, "max": 1.7617037913750806 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "humaneval", "scope": "docker_container", "samples": 862, "start_utc": "2026-10-12T02:41:14.813615+00:00", "end_utc": "2026-10-12T02:55:35.813507+00:00", "sample_interval_s": { "n": 861, "median": 1.0000003740005923, "min": 0.9999150429994188, "max": 1.000075817000834 }, "collector_duration_s": { "n": 862, "median": 0.0443701449994478, "min": 0.03519940299884183, "max": 0.05236029099978623 }, "gpu_utilization_pct": { "n": 862, "median": 92.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 862, "median": 15494.0, "min": 15494.0, "max": 15494.0 }, "gpu_power_w": { "n": 862, "median": 104.205, "min": 47.57, "max": 130.74 }, "gpu_temperature_c": { "n": 862, "median": 42.0, "min": 38.0, "max": 45.0 }, "cpu_utilization_pct": { "n": 862, "median": 59.136546184738954, "min": 9.58627648839556, "max": 69.61770623742454 }, "proc_memory_available_gib": { "n": 862, "median": 20.11665439605713, "min": 19.824585914611816, "max": 20.228290557861328 }, "proc_memory_total_minus_available_gib": { "n": 862, "median": 31.88334560394287, "min": 31.771709442138672, "max": 32.175414085388184 }, "proc_page_cache_cached_gib": { "n": 862, "median": 46.072587966918945, "min": 45.9705696105957, "max": 46.08281707763672 }, "proc_shmem_gib": { "n": 862, "median": 29.22247314453125, "min": 29.22247314453125, "max": 29.222476959228516 }, "proc_cached_minus_shmem_estimate_gib": { "n": 862, "median": 16.850114822387695, "min": 16.748096466064453, "max": 16.86034393310547 }, "proc_unevictable_gib": { "n": 862, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 862, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 862, "median": 53.859375, "min": 53.859375, "max": 54.03515625 }, "pswpin_counter_delta": 16, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 652, "container_memory_current_gib": { "n": 862, "median": 36.22322082519531, "min": 36.112953186035156, "max": 36.55246353149414 }, "container_memory_lifetime_peak_gib": { "n": 862, "median": 36.9876708984375, "min": 36.9876708984375, "max": 36.9876708984375 }, "container_swap_mib": { "n": 862, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 862, "median": 2.3192672729492188, "min": 2.209644317626953, "max": 2.6480026245117188 }, "container_memory_stat_file_gib": { "n": 862, "median": 33.76284408569336, "min": 33.76274871826172, "max": 33.76293182373047 }, "container_memory_stat_shmem_gib": { "n": 862, "median": 29.222198486328125, "min": 29.222198486328125, "max": 29.222198486328125 }, "container_memory_stat_file_mapped_gib": { "n": 862, "median": 29.173667907714844, "min": 29.173667907714844, "max": 29.173683166503906 }, "container_memory_stat_unevictable_gib": { "n": 862, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 2260180992, "read_mib_per_s": { "n": 861, "median": 2.3203526328167863, "min": 0.03906147869861418, "max": 13.38266822146436 }, "write_bytes": 294912, "write_mib_per_s": { "n": 861, "median": 0.0, "min": 0.0, "max": 0.015625336585375755 } }, "252:3": { "read_bytes": 2260180992, "read_mib_per_s": { "n": 861, "median": 2.3203526328167863, "min": 0.03906147869861418, "max": 13.38266822146436 }, "write_bytes": 294912, "write_mib_per_s": { "n": 861, "median": 0.0, "min": 0.0, "max": 0.015625336585375755 } }, "252:6": { "read_bytes": 2260180992, "read_mib_per_s": { "n": 861, "median": 2.3203526328167863, "min": 0.03906147869861418, "max": 13.38266822146436 }, "write_bytes": 294912, "write_mib_per_s": { "n": 861, "median": 0.0, "min": 0.0, "max": 0.015625336585375755 } }, "259:0": { "read_bytes": 2260180992, "read_mib_per_s": { "n": 861, "median": 2.3203526328167863, "min": 0.03906147869861418, "max": 13.38266822146436 }, "write_bytes": 294912, "write_mib_per_s": { "n": 861, "median": 0.0, "min": 0.0, "max": 0.015625336585375755 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 862, "median": 32975848.0, "min": 32860932, "max": 33320596 }, "VmHWM": { "n": 862, "median": 33514628.0, "min": 33514628, "max": 33514628 }, "VmSwap": { "n": 862, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "26431": { "read_bytes": 2260180992, "write_bytes": 303104, "rchar": 2260993878, "wchar": 650500 } } }, "python": { "VmRSS": { "n": 862, "median": 224460.0, "min": 224416, "max": 224524 }, "VmHWM": { "n": 862, "median": 224460.0, "min": 224416, "max": 224524 }, "VmSwap": { "n": 862, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "26259": { "read_bytes": 0, "write_bytes": 0, "rchar": 5923199, "wchar": 427938 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "humaneval", "scope": "host", "samples": 862, "start_utc": "2026-10-12T02:41:14.697206+00:00", "end_utc": "2026-10-12T02:55:35.697276+00:00", "sample_interval_s": { "n": 861, "median": 1.0000003739987733, "min": 0.9997662810001202, "max": 1.000243506001425 }, "collector_duration_s": { "n": 862, "median": 0.05546766550014581, "min": 0.0428051789986057, "max": 0.06369453899969812 }, "gpu_utilization_pct": { "n": 862, "median": 92.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 862, "median": 15494.0, "min": 15494.0, "max": 15494.0 }, "gpu_power_w": { "n": 862, "median": 104.205, "min": 47.57, "max": 130.74 }, "gpu_temperature_c": { "n": 862, "median": 42.0, "min": 38.0, "max": 45.0 }, "cpu_utilization_pct": { "n": 862, "median": 49.54051796157059, "min": 8.976510067114093, "max": 58.54271356783919 }, "proc_memory_available_gib": { "n": 862, "median": 22.8187198638916, "min": 22.282344818115234, "max": 23.105331420898438 }, "proc_memory_total_minus_available_gib": { "n": 862, "median": 35.93401527404785, "min": 35.647403717041016, "max": 36.47039031982422 }, "proc_page_cache_cached_gib": { "n": 862, "median": 46.347984313964844, "min": 46.24485778808594, "max": 46.36245346069336 }, "proc_shmem_gib": { "n": 862, "median": 29.288585662841797, "min": 29.288585662841797, "max": 29.288650512695312 }, "proc_cached_minus_shmem_estimate_gib": { "n": 862, "median": 17.059398651123047, "min": 16.956268310546875, "max": 17.073867797851562 }, "proc_unevictable_gib": { "n": 862, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 862, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 862, "median": 54.765625, "min": 54.765625, "max": 54.94140625 }, "pswpin_counter_delta": 19, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 654, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 862, "median": 32975848.0, "min": 32860932, "max": 33320596 }, "VmHWM": { "n": 862, "median": 33514628.0, "min": 33514628, "max": 33514628 }, "VmSwap": { "n": 862, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65735": { "read_bytes": 2259791872, "write_bytes": 303104, "rchar": 2260604758, "wchar": 650446 } } }, "python": { "VmRSS": { "n": 862, "median": 224460.0, "min": 224416, "max": 224524 }, "VmHWM": { "n": 862, "median": 224460.0, "min": 224416, "max": 224524 }, "VmSwap": { "n": 862, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65561": { "read_bytes": 0, "write_bytes": 0, "rchar": 5929193, "wchar": 427938 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 2427166720, "read_mib_per_s": { "n": 861, "median": 2.3633021534067997, "min": 0.03906111044000912, "max": 91.38009214915091 }, "write_bytes": 79249408, "write_mib_per_s": { "n": 861, "median": 0.011718705515795343, "min": 0, "max": 2.199257863804508 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "official-optional-65536", "scope": "docker_container", "samples": 162, "start_utc": "2026-10-12T02:20:45.813386+00:00", "end_utc": "2026-10-12T02:23:26.813773+00:00", "sample_interval_s": { "n": 161, "median": 1.0000002439992386, "min": 0.9997630639991257, "max": 1.00024273300005 }, "collector_duration_s": { "n": 162, "median": 0.04004882000026555, "min": 0.03580757499912579, "max": 0.04903186299998197 }, "gpu_utilization_pct": { "n": 162, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 162, "median": 15484.0, "min": 15404.0, "max": 15486.0 }, "gpu_power_w": { "n": 162, "median": 114.62, "min": 10.72, "max": 150.16 }, "gpu_temperature_c": { "n": 162, "median": 47.0, "min": 39.0, "max": 55.0 }, "cpu_utilization_pct": { "n": 162, "median": 12.268355065195587, "min": 10.080645161290322, "max": 59.919436052366564 }, "proc_memory_available_gib": { "n": 162, "median": 20.260717391967773, "min": 19.707592010498047, "max": 20.798110008239746 }, "proc_memory_total_minus_available_gib": { "n": 162, "median": 31.739282608032227, "min": 31.201889991760254, "max": 32.29240798950195 }, "proc_page_cache_cached_gib": { "n": 162, "median": 45.92232322692871, "min": 45.8623161315918, "max": 45.924110412597656 }, "proc_shmem_gib": { "n": 162, "median": 29.222488403320312, "min": 29.173583984375, "max": 29.222488403320312 }, "proc_cached_minus_shmem_estimate_gib": { "n": 162, "median": 16.6998348236084, "min": 16.688732147216797, "max": 16.701622009277344 }, "proc_unevictable_gib": { "n": 162, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 162, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 162, "median": 54.83203125, "min": 54.703125, "max": 54.83203125 }, "pswpin_counter_delta": 9, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 113, "container_memory_current_gib": { "n": 162, "median": 36.18068504333496, "min": 35.65583801269531, "max": 36.73159408569336 }, "container_memory_lifetime_peak_gib": { "n": 162, "median": 36.9876708984375, "min": 36.9876708984375, "max": 36.9876708984375 }, "container_swap_mib": { "n": 162, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 162, "median": 2.2773361206054688, "min": 1.8085174560546875, "max": 2.8267669677734375 }, "container_memory_stat_file_gib": { "n": 162, "median": 33.762264251708984, "min": 33.70745849609375, "max": 33.76227951049805 }, "container_memory_stat_shmem_gib": { "n": 162, "median": 29.222198486328125, "min": 29.173397064208984, "max": 29.222198486328125 }, "container_memory_stat_file_mapped_gib": { "n": 162, "median": 29.173507690429688, "min": 29.120079040527344, "max": 29.173507690429688 }, "container_memory_stat_unevictable_gib": { "n": 162, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 962867200, "read_mib_per_s": { "n": 161, "median": 0.039062967857138815, "min": 0.03515174867797426, "max": 115.09019339418285 }, "write_bytes": 57344, "write_mib_per_s": { "n": 161, "median": 0.0, "min": 0.0, "max": 0.01562502564065882 } }, "252:3": { "read_bytes": 962867200, "read_mib_per_s": { "n": 161, "median": 0.039062967857138815, "min": 0.03515174867797426, "max": 115.09019339418285 }, "write_bytes": 57344, "write_mib_per_s": { "n": 161, "median": 0.0, "min": 0.0, "max": 0.01562502564065882 } }, "252:6": { "read_bytes": 962867200, "read_mib_per_s": { "n": 161, "median": 0.039062967857138815, "min": 0.03515174867797426, "max": 115.09019339418285 }, "write_bytes": 57344, "write_mib_per_s": { "n": 161, "median": 0.0, "min": 0.0, "max": 0.01562502564065882 } }, "259:0": { "read_bytes": 962867200, "read_mib_per_s": { "n": 161, "median": 0.039062967857138815, "min": 0.03515174867797426, "max": 115.09019339418285 }, "write_bytes": 57344, "write_mib_per_s": { "n": 161, "median": 0.0, "min": 0.0, "max": 0.01562502564065882 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 162, "median": 32939980.0, "min": 32387652, "max": 33515844 }, "VmHWM": { "n": 162, "median": 33514352.0, "min": 32827852, "max": 33515844 }, "VmSwap": { "n": 162, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "26431": { "read_bytes": 962723840, "write_bytes": 61440, "rchar": 958076206, "wchar": 65634 } } }, "python": { "VmRSS": { "n": 162, "median": 216344.0, "min": 213796, "max": 218316 }, "VmHWM": { "n": 162, "median": 216344.0, "min": 213796, "max": 218316 }, "VmSwap": { "n": 162, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "26259": { "read_bytes": 143360, "write_bytes": 0, "rchar": 1064204, "wchar": 1502825 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "official-optional-65536", "scope": "host", "samples": 162, "start_utc": "2026-10-12T02:20:45.697252+00:00", "end_utc": "2026-10-12T02:23:26.697238+00:00", "sample_interval_s": { "n": 161, "median": 0.9999999659994501, "min": 0.9997608300000138, "max": 1.0002415449998807 }, "collector_duration_s": { "n": 162, "median": 0.045679346500037354, "min": 0.04254682999999204, "max": 0.06210068599830265 }, "gpu_utilization_pct": { "n": 162, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 162, "median": 15484.0, "min": 15404.0, "max": 15486.0 }, "gpu_power_w": { "n": 162, "median": 114.83500000000001, "min": 10.72, "max": 149.93 }, "gpu_temperature_c": { "n": 162, "median": 47.0, "min": 39.0, "max": 56.0 }, "cpu_utilization_pct": { "n": 162, "median": 10.20477775103414, "min": 8.473154362416107, "max": 50.96234309623431 }, "proc_memory_available_gib": { "n": 162, "median": 23.14569854736328, "min": 22.70122528076172, "max": 23.9146728515625 }, "proc_memory_total_minus_available_gib": { "n": 162, "median": 35.60703659057617, "min": 34.83806228637695, "max": 36.051509857177734 }, "proc_page_cache_cached_gib": { "n": 162, "median": 46.18009376525879, "min": 46.122703552246094, "max": 46.18324279785156 }, "proc_shmem_gib": { "n": 162, "median": 29.28860092163086, "min": 29.23979949951172, "max": 29.29343032836914 }, "proc_cached_minus_shmem_estimate_gib": { "n": 162, "median": 16.891450881958008, "min": 16.879104614257812, "max": 16.894641876220703 }, "proc_unevictable_gib": { "n": 162, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 162, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 162, "median": 55.73828125, "min": 55.609375, "max": 55.73828125 }, "pswpin_counter_delta": 14, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 120, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 162, "median": 32939980.0, "min": 32387652, "max": 33515844 }, "VmHWM": { "n": 162, "median": 33514352.0, "min": 32827852, "max": 33515844 }, "VmSwap": { "n": 162, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65735": { "read_bytes": 962715648, "write_bytes": 61440, "rchar": 958068014, "wchar": 65634 } } }, "python": { "VmRSS": { "n": 162, "median": 216344.0, "min": 213796, "max": 218316 }, "VmHWM": { "n": 162, "median": 216344.0, "min": 213796, "max": 218316 }, "VmSwap": { "n": 162, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65561": { "read_bytes": 143360, "write_bytes": 0, "rchar": 1064204, "wchar": 1502825 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 980189184, "read_mib_per_s": { "n": 161, "median": 0.09765705820979319, "min": 0.03515624138673201, "max": 115.84055165165778 }, "write_bytes": 20701184, "write_mib_per_s": { "n": 161, "median": 0.007812459890835873, "min": 0, "max": 5.44524880148458 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "official-standard-32768", "scope": "docker_container", "samples": 101, "start_utc": "2026-10-12T02:13:13.849137+00:00", "end_utc": "2026-10-12T02:14:53.849549+00:00", "sample_interval_s": { "n": 100, "median": 1.0000002734996087, "min": 0.999780064999868, "max": 1.0002225789994554 }, "collector_duration_s": { "n": 101, "median": 0.04072684799939452, "min": 0.037574425999991945, "max": 0.04839017399899603 }, "gpu_utilization_pct": { "n": 101, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 101, "median": 15502.0, "min": 15502.0, "max": 15504.0 }, "gpu_power_w": { "n": 101, "median": 111.11, "min": 10.59, "max": 152.65 }, "gpu_temperature_c": { "n": 101, "median": 46.0, "min": 38.0, "max": 54.0 }, "cpu_utilization_pct": { "n": 101, "median": 15.446339017051153, "min": 9.19080919080919, "max": 65.33066132264528 }, "proc_memory_available_gib": { "n": 101, "median": 21.250516891479492, "min": 20.672718048095703, "max": 21.37357807159424 }, "proc_memory_total_minus_available_gib": { "n": 101, "median": 30.749483108520508, "min": 30.62642192840576, "max": 31.327281951904297 }, "proc_page_cache_cached_gib": { "n": 101, "median": 17.39410400390625, "min": 17.362030029296875, "max": 17.405433654785156 }, "proc_shmem_gib": { "n": 101, "median": 0.9385452270507812, "min": 0.9190139770507812, "max": 0.9485054016113281 }, "proc_cached_minus_shmem_estimate_gib": { "n": 101, "median": 16.45555877685547, "min": 16.443016052246094, "max": 16.457199096679688 }, "proc_unevictable_gib": { "n": 101, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 101, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 101, "median": 62.28125, "min": 62.28125, "max": 62.28125 }, "pswpin_counter_delta": 181, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 309, "container_memory_current_gib": { "n": 101, "median": 30.670391082763672, "min": 30.517742156982422, "max": 31.245433807373047 }, "container_memory_lifetime_peak_gib": { "n": 101, "median": 32.25005340576172, "min": 32.25005340576172, "max": 32.25005340576172 }, "container_swap_mib": { "n": 101, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 101, "median": 29.5823974609375, "min": 29.455516815185547, "max": 30.156566619873047 }, "container_memory_stat_file_gib": { "n": 101, "median": 0.9597663879394531, "min": 0.9344940185546875, "max": 0.9600486755371094 }, "container_memory_stat_shmem_gib": { "n": 101, "median": 0.9383010864257812, "min": 0.9188728332519531, "max": 0.9383010864257812 }, "container_memory_stat_file_mapped_gib": { "n": 101, "median": 0.7604293823242188, "min": 0.7375755310058594, "max": 0.7641220092773438 }, "container_memory_stat_unevictable_gib": { "n": 101, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:6": { "read_bytes": 846475264, "read_mib_per_s": { "n": 100, "median": 0.12304749071933413, "min": 0.03906203371651421, "max": 137.71517137437235 }, "write_bytes": 40960, "write_mib_per_s": { "n": 100, "median": 0.0, "min": 0.0, "max": 0.015625037171961124 } }, "252:0": { "read_bytes": 0, "read_mib_per_s": { "n": 100, "median": 0.0, "min": 0.0, "max": 0.0 }, "write_bytes": 0, "write_mib_per_s": { "n": 100, "median": 0.0, "min": 0.0, "max": 0.0 } }, "252:4": { "read_bytes": 846475264, "read_mib_per_s": { "n": 100, "median": 0.12304749071933413, "min": 0.03906203371651421, "max": 137.71517137437235 }, "write_bytes": 40960, "write_mib_per_s": { "n": 100, "median": 0.0, "min": 0.0, "max": 0.015625037171961124 } }, "252:3": { "read_bytes": 846475264, "read_mib_per_s": { "n": 100, "median": 0.12304749071933413, "min": 0.03906203371651421, "max": 137.71517137437235 }, "write_bytes": 40960, "write_mib_per_s": { "n": 100, "median": 0.0, "min": 0.0, "max": 0.015625037171961124 } }, "259:0": { "read_bytes": 846475264, "read_mib_per_s": { "n": 100, "median": 0.12304749071933413, "min": 0.03906203371651421, "max": 137.71517137437235 }, "write_bytes": 40960, "write_mib_per_s": { "n": 100, "median": 0.0, "min": 0.0, "max": 0.015625037171961124 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 101, "median": 31711852, "min": 31554916, "max": 32312608 }, "VmHWM": { "n": 101, "median": 31954404, "min": 31954404, "max": 32312608 }, "VmSwap": { "n": 101, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "20554": { "read_bytes": 846467072, "write_bytes": 40960, "rchar": 840923980, "wchar": 45980 } } }, "python": { "VmRSS": { "n": 101, "median": 211400, "min": 209052, "max": 211564 }, "VmHWM": { "n": 101, "median": 214072, "min": 214072, "max": 214072 }, "VmSwap": { "n": 101, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "20458": { "read_bytes": 8192, "write_bytes": 0, "rchar": 680095, "wchar": 948643 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "official-standard-32768", "scope": "host", "samples": 101, "start_utc": "2026-10-12T02:13:13.697250+00:00", "end_utc": "2026-10-12T02:14:53.697211+00:00", "sample_interval_s": { "n": 100, "median": 0.9999998550001692, "min": 0.9997663059984916, "max": 1.0002412970006844 }, "collector_duration_s": { "n": 101, "median": 0.04706514700046682, "min": 0.043711442000130774, "max": 0.06666866399973514 }, "gpu_utilization_pct": { "n": 101, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 101, "median": 15502.0, "min": 15502.0, "max": 15504.0 }, "gpu_power_w": { "n": 101, "median": 113.14, "min": 10.59, "max": 151.5 }, "gpu_temperature_c": { "n": 101, "median": 46.0, "min": 38.0, "max": 54.0 }, "cpu_utilization_pct": { "n": 101, "median": 12.781954887218046, "min": 5.008347245409015, "max": 56.35451505016722 }, "proc_memory_available_gib": { "n": 101, "median": 24.319686889648438, "min": 23.891536712646484, "max": 24.497215270996094 }, "proc_memory_total_minus_available_gib": { "n": 101, "median": 34.433048248291016, "min": 34.25551986694336, "max": 34.86119842529297 }, "proc_page_cache_cached_gib": { "n": 101, "median": 17.64553451538086, "min": 17.612346649169922, "max": 17.64788818359375 }, "proc_shmem_gib": { "n": 101, "median": 1.0046577453613281, "min": 0.9851341247558594, "max": 1.004730224609375 }, "proc_cached_minus_shmem_estimate_gib": { "n": 101, "median": 16.640872955322266, "min": 16.627212524414062, "max": 16.643230438232422 }, "proc_unevictable_gib": { "n": 101, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 101, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 101, "median": 63.1875, "min": 63.1875, "max": 63.1875 }, "pswpin_counter_delta": 207, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 331, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 101, "median": 31709776, "min": 31582980, "max": 32312476 }, "VmHWM": { "n": 101, "median": 31954404, "min": 31954404, "max": 32312476 }, "VmSwap": { "n": 101, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "55503": { "read_bytes": 846462976, "write_bytes": 40960, "rchar": 840919884, "wchar": 45980 } } }, "python": { "VmRSS": { "n": 101, "median": 211400, "min": 209052, "max": 211564 }, "VmHWM": { "n": 101, "median": 214072, "min": 214072, "max": 214072 }, "VmSwap": { "n": 101, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "55404": { "read_bytes": 8192, "write_bytes": 0, "rchar": 680095, "wchar": 948643 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 873476096, "read_mib_per_s": { "n": 100, "median": 0.3887192621049406, "min": 0.03906181270737311, "max": 138.46367428570727 }, "write_bytes": 16044032, "write_mib_per_s": { "n": 100, "median": 0.015624932969067943, "min": 0, "max": 3.5312189429254346 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "needles-128k-ctx131072", "scope": "docker_container", "samples": 171, "start_utc": "2026-10-12T02:58:47.757943+00:00", "end_utc": "2026-10-12T03:01:37.757843+00:00", "sample_interval_s": { "n": 170, "median": 0.9999990824990164, "min": 0.999572946000626, "max": 1.0004211679988657 }, "collector_duration_s": { "n": 171, "median": 0.03920158199980506, "min": 0.03576497699941683, "max": 0.042204544000924216 }, "gpu_utilization_pct": { "n": 171, "median": 100.0, "min": 43.0, "max": 100.0 }, "vram_used_mib": { "n": 171, "median": 15480.0, "min": 15480.0, "max": 15480.0 }, "gpu_power_w": { "n": 171, "median": 150.13, "min": 66.52, "max": 154.68 }, "gpu_temperature_c": { "n": 171, "median": 49.0, "min": 42.0, "max": 52.0 }, "cpu_utilization_pct": { "n": 171, "median": 11.465603190428714, "min": 10.753768844221106, "max": 26.541274817136888 }, "proc_memory_available_gib": { "n": 171, "median": 18.911967277526855, "min": 18.688754081726074, "max": 19.366679191589355 }, "proc_memory_total_minus_available_gib": { "n": 171, "median": 33.088032722473145, "min": 32.633320808410645, "max": 33.311245918273926 }, "proc_page_cache_cached_gib": { "n": 171, "median": 46.290077209472656, "min": 46.28641891479492, "max": 46.29279327392578 }, "proc_shmem_gib": { "n": 171, "median": 30.187644958496094, "min": 30.187644958496094, "max": 30.187644958496094 }, "proc_cached_minus_shmem_estimate_gib": { "n": 171, "median": 16.102432250976562, "min": 16.098773956298828, "max": 16.105148315429688 }, "proc_unevictable_gib": { "n": 171, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 171, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 171, "median": 53.859375, "min": 53.7890625, "max": 53.859375 }, "pswpin_counter_delta": 2, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 39, "container_memory_current_gib": { "n": 171, "median": 32.9884147644043, "min": 32.530086517333984, "max": 33.21181106567383 }, "container_memory_lifetime_peak_gib": { "n": 171, "median": 33.967227935791016, "min": 33.967227935791016, "max": 33.967227935791016 }, "container_swap_mib": { "n": 171, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 171, "median": 2.6601181030273438, "min": 2.2032432556152344, "max": 2.8825950622558594 }, "container_memory_stat_file_gib": { "n": 171, "median": 30.18741226196289, "min": 30.18741226196289, "max": 30.18741226196289 }, "container_memory_stat_shmem_gib": { "n": 171, "median": 30.187366485595703, "min": 30.187366485595703, "max": 30.187366485595703 }, "container_memory_stat_file_mapped_gib": { "n": 171, "median": 29.942607879638672, "min": 29.942607879638672, "max": 29.942607879638672 }, "container_memory_stat_unevictable_gib": { "n": 171, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 4787453952, "read_mib_per_s": { "n": 170, "median": 0.03906267363361324, "min": 0.035156251582030576, "max": 276.73194981984045 }, "write_bytes": 8192, "write_mib_per_s": { "n": 170, "median": 0.0, "min": 0.0, "max": 0.003907143309707002 } }, "252:3": { "read_bytes": 4787453952, "read_mib_per_s": { "n": 170, "median": 0.03906267363361324, "min": 0.035156251582030576, "max": 276.73194981984045 }, "write_bytes": 8192, "write_mib_per_s": { "n": 170, "median": 0.0, "min": 0.0, "max": 0.003907143309707002 } }, "252:6": { "read_bytes": 4787453952, "read_mib_per_s": { "n": 170, "median": 0.03906267363361324, "min": 0.035156251582030576, "max": 276.73194981984045 }, "write_bytes": 8192, "write_mib_per_s": { "n": 170, "median": 0.0, "min": 0.0, "max": 0.003907143309707002 } }, "259:0": { "read_bytes": 4787453952, "read_mib_per_s": { "n": 170, "median": 0.03906267363361324, "min": 0.035156251582030576, "max": 276.73194981984045 }, "write_bytes": 8192, "write_mib_per_s": { "n": 170, "median": 0.0, "min": 0.0, "max": 0.003907143309707002 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 171, "median": 34340624, "min": 33861588, "max": 34573872 }, "VmHWM": { "n": 171, "median": 34685700, "min": 34319380, "max": 34687392 }, "VmSwap": { "n": 171, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "63181": { "read_bytes": 4787453952, "write_bytes": 8192, "rchar": 4788605117, "wchar": 3479 } } }, "python": { "VmRSS": { "n": 171, "median": 231220, "min": 231040, "max": 231236 }, "VmHWM": { "n": 171, "median": 233640, "min": 233556, "max": 233796 }, "VmSwap": { "n": 171, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "63086": { "read_bytes": 0, "write_bytes": 0, "rchar": 1058746, "wchar": 1081936 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "needles-128k-ctx131072", "scope": "host", "samples": 171, "start_utc": "2026-10-12T02:58:47.884173+00:00", "end_utc": "2026-10-12T03:01:37.884195+00:00", "sample_interval_s": { "n": 170, "median": 0.9999999555002432, "min": 0.9995746919994417, "max": 1.0004134850005357 }, "collector_duration_s": { "n": 171, "median": 0.045627460998730385, "min": 0.0422882139991998, "max": 0.055259609000131604 }, "gpu_utilization_pct": { "n": 171, "median": 100.0, "min": 43.0, "max": 100.0 }, "vram_used_mib": { "n": 171, "median": 15480.0, "min": 15480.0, "max": 15480.0 }, "gpu_power_w": { "n": 171, "median": 150.13, "min": 66.52, "max": 154.68 }, "gpu_temperature_c": { "n": 171, "median": 49.0, "min": 42.0, "max": 52.0 }, "cpu_utilization_pct": { "n": 171, "median": 9.690893901420218, "min": 9.265442404006677, "max": 21.26334519572954 }, "proc_memory_available_gib": { "n": 171, "median": 22.129779815673828, "min": 21.74036407470703, "max": 22.6702880859375 }, "proc_memory_total_minus_available_gib": { "n": 171, "median": 36.622955322265625, "min": 36.08244705200195, "max": 37.01237106323242 }, "proc_page_cache_cached_gib": { "n": 171, "median": 46.570499420166016, "min": 46.56599426269531, "max": 46.574058532714844 }, "proc_shmem_gib": { "n": 171, "median": 30.253761291503906, "min": 30.253761291503906, "max": 30.253822326660156 }, "proc_cached_minus_shmem_estimate_gib": { "n": 171, "median": 16.31673812866211, "min": 16.312232971191406, "max": 16.320297241210938 }, "proc_unevictable_gib": { "n": 171, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 171, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 171, "median": 54.765625, "min": 54.6953125, "max": 54.765625 }, "pswpin_counter_delta": 2, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 39, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 171, "median": 34340624, "min": 33861588, "max": 34573872 }, "VmHWM": { "n": 171, "median": 34685700, "min": 34319380, "max": 34687392 }, "VmSwap": { "n": 171, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "114103": { "read_bytes": 4787453952, "write_bytes": 8192, "rchar": 4788609213, "wchar": 3479 } } }, "python": { "VmRSS": { "n": 171, "median": 231220, "min": 231040, "max": 231236 }, "VmHWM": { "n": 171, "median": 233640, "min": 233556, "max": 233796 }, "VmSwap": { "n": 171, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "114005": { "read_bytes": 0, "write_bytes": 0, "rchar": 1058746, "wchar": 1081936 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 4803731456, "read_mib_per_s": { "n": 170, "median": 0.039062965532897995, "min": 0.03515613616445191, "max": 304.21866078198127 }, "write_bytes": 13086720, "write_mib_per_s": { "n": 170, "median": 0.0039062402148725903, "min": 0, "max": 2.1171840447536683 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "needles-16k-ctx32768", "scope": "docker_container", "samples": 22, "start_utc": "2026-10-12T02:16:29.849468+00:00", "end_utc": "2026-10-12T02:16:50.849756+00:00", "sample_interval_s": { "n": 21, "median": 0.9999995059988578, "min": 0.9997253089986771, "max": 1.0003477340014797 }, "collector_duration_s": { "n": 22, "median": 0.04059428950040456, "min": 0.03799347299900546, "max": 0.04259070000080101 }, "gpu_utilization_pct": { "n": 22, "median": 100.0, "min": 98.0, "max": 100.0 }, "vram_used_mib": { "n": 22, "median": 15504.0, "min": 15504.0, "max": 15504.0 }, "gpu_power_w": { "n": 22, "median": 146.445, "min": 78.92, "max": 153.73 }, "gpu_temperature_c": { "n": 22, "median": 50.0, "min": 44.0, "max": 52.0 }, "cpu_utilization_pct": { "n": 22, "median": 11.355628339596276, "min": 10.742971887550201, "max": 27.421109902067464 }, "proc_memory_available_gib": { "n": 22, "median": 21.324480056762695, "min": 21.22361183166504, "max": 21.327452659606934 }, "proc_memory_total_minus_available_gib": { "n": 22, "median": 30.675519943237305, "min": 30.672547340393066, "max": 30.77638816833496 }, "proc_page_cache_cached_gib": { "n": 22, "median": 17.42008399963379, "min": 17.4197998046875, "max": 17.42031478881836 }, "proc_shmem_gib": { "n": 22, "median": 0.9385337829589844, "min": 0.9385337829589844, "max": 0.9385337829589844 }, "proc_cached_minus_shmem_estimate_gib": { "n": 22, "median": 16.481550216674805, "min": 16.481266021728516, "max": 16.481781005859375 }, "proc_unevictable_gib": { "n": 22, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 22, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 22, "median": 61.87890625, "min": 61.87890625, "max": 61.87890625 }, "pswpin_counter_delta": 57, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 44, "container_memory_current_gib": { "n": 22, "median": 30.597686767578125, "min": 30.59597396850586, "max": 30.70226287841797 }, "container_memory_lifetime_peak_gib": { "n": 22, "median": 32.25005340576172, "min": 32.25005340576172, "max": 32.25005340576172 }, "container_swap_mib": { "n": 22, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 22, "median": 29.509883880615234, "min": 29.508148193359375, "max": 29.61554718017578 }, "container_memory_stat_file_gib": { "n": 22, "median": 0.9600563049316406, "min": 0.960052490234375, "max": 0.9600563049316406 }, "container_memory_stat_shmem_gib": { "n": 22, "median": 0.9383010864257812, "min": 0.9383010864257812, "max": 0.9383010864257812 }, "container_memory_stat_file_mapped_gib": { "n": 22, "median": 0.7607040405273438, "min": 0.7607040405273438, "max": 0.7607040405273438 }, "container_memory_stat_unevictable_gib": { "n": 22, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:6": { "read_bytes": 294719488, "read_mib_per_s": { "n": 21, "median": 0.03906259023458841, "min": 0.039048921362321215, "max": 280.1180629158102 }, "write_bytes": 8192, "write_mib_per_s": { "n": 21, "median": 0.0, "min": 0.0, "max": 0.00781110837292947 } }, "252:0": { "read_bytes": 0, "read_mib_per_s": { "n": 21, "median": 0.0, "min": 0.0, "max": 0.0 }, "write_bytes": 0, "write_mib_per_s": { "n": 21, "median": 0.0, "min": 0.0, "max": 0.0 } }, "252:4": { "read_bytes": 294719488, "read_mib_per_s": { "n": 21, "median": 0.03906259023458841, "min": 0.039048921362321215, "max": 280.1180629158102 }, "write_bytes": 8192, "write_mib_per_s": { "n": 21, "median": 0.0, "min": 0.0, "max": 0.00781110837292947 } }, "252:3": { "read_bytes": 294719488, "read_mib_per_s": { "n": 21, "median": 0.03906259023458841, "min": 0.039048921362321215, "max": 280.1180629158102 }, "write_bytes": 8192, "write_mib_per_s": { "n": 21, "median": 0.0, "min": 0.0, "max": 0.00781110837292947 } }, "259:0": { "read_bytes": 294719488, "read_mib_per_s": { "n": 21, "median": 0.03906259023458841, "min": 0.039048921362321215, "max": 280.1180629158102 }, "write_bytes": 8192, "write_mib_per_s": { "n": 21, "median": 0.0, "min": 0.0, "max": 0.00781110837292947 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 22, "median": 31633848.0, "min": 31632044, "max": 31745308 }, "VmHWM": { "n": 22, "median": 32311920.0, "min": 32311920, "max": 32311920 }, "VmSwap": { "n": 22, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "20554": { "read_bytes": 294719488, "write_bytes": 4096, "rchar": 294928345, "wchar": 2147 } } }, "python": { "VmRSS": { "n": 22, "median": 212232.0, "min": 211524, "max": 212268 }, "VmHWM": { "n": 22, "median": 214072.0, "min": 214072, "max": 214072 }, "VmSwap": { "n": 22, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "20458": { "read_bytes": 0, "write_bytes": 0, "rchar": 134626, "wchar": 199453 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "needles-16k-ctx32768", "scope": "host", "samples": 22, "start_utc": "2026-10-12T02:16:29.697223+00:00", "end_utc": "2026-10-12T02:16:50.697207+00:00", "sample_interval_s": { "n": 21, "median": 0.9999985279991961, "min": 0.9996085349994246, "max": 1.0002254419996461 }, "collector_duration_s": { "n": 22, "median": 0.04606243650050601, "min": 0.04411952500049665, "max": 0.04983770300168544 }, "gpu_utilization_pct": { "n": 22, "median": 100.0, "min": 98.0, "max": 100.0 }, "vram_used_mib": { "n": 22, "median": 15504.0, "min": 15504.0, "max": 15504.0 }, "gpu_power_w": { "n": 22, "median": 146.445, "min": 78.92, "max": 153.73 }, "gpu_temperature_c": { "n": 22, "median": 50.0, "min": 44.0, "max": 52.0 }, "cpu_utilization_pct": { "n": 22, "median": 9.583339988430549, "min": 9.265442404006677, "max": 25.86359610274579 }, "proc_memory_available_gib": { "n": 22, "median": 24.43220329284668, "min": 24.287433624267578, "max": 24.50281524658203 }, "proc_memory_total_minus_available_gib": { "n": 22, "median": 34.32053184509277, "min": 34.24991989135742, "max": 34.465301513671875 }, "proc_page_cache_cached_gib": { "n": 22, "median": 17.673294067382812, "min": 17.672847747802734, "max": 17.673675537109375 }, "proc_shmem_gib": { "n": 22, "median": 1.0046463012695312, "min": 1.004638671875, "max": 1.0046463012695312 }, "proc_cached_minus_shmem_estimate_gib": { "n": 22, "median": 16.66864776611328, "min": 16.668201446533203, "max": 16.669029235839844 }, "proc_unevictable_gib": { "n": 22, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 22, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 22, "median": 62.78515625, "min": 62.78515625, "max": 62.78515625 }, "pswpin_counter_delta": 57, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 44, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 22, "median": 31633848.0, "min": 31632044, "max": 31745308 }, "VmHWM": { "n": 22, "median": 32311920.0, "min": 32311920, "max": 32311920 }, "VmSwap": { "n": 22, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "55503": { "read_bytes": 294887424, "write_bytes": 8192, "rchar": 295096736, "wchar": 3143 } } }, "python": { "VmRSS": { "n": 22, "median": 212232.0, "min": 211568, "max": 212268 }, "VmHWM": { "n": 22, "median": 214072.0, "min": 214072, "max": 214072 }, "VmSwap": { "n": 22, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "55404": { "read_bytes": 0, "write_bytes": 0, "rchar": 134790, "wchar": 199588 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 296812544, "read_mib_per_s": { "n": 21, "median": 0.06250000031252512, "min": 0.03905417556338798, "max": 263.7178290359143 }, "write_bytes": 8126464, "write_mib_per_s": { "n": 21, "median": 0.0039054175563387975, "min": 0, "max": 4.898492950941869 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "needles-32k-ctx32768", "scope": "docker_container", "samples": 1, "start_utc": "2026-10-12T02:16:51.849226+00:00", "end_utc": "2026-10-12T02:16:51.849226+00:00", "sample_interval_s": { "n": 0, "median": null, "min": null, "max": null }, "collector_duration_s": { "n": 1, "median": 0.035620619000837905, "min": 0.035620619000837905, "max": 0.035620619000837905 }, "gpu_utilization_pct": { "n": 1, "median": 71.0, "min": 71.0, "max": 71.0 }, "vram_used_mib": { "n": 1, "median": 15504.0, "min": 15504.0, "max": 15504.0 }, "gpu_power_w": { "n": 1, "median": 134.89, "min": 134.89, "max": 134.89 }, "gpu_temperature_c": { "n": 1, "median": 45.0, "min": 45.0, "max": 45.0 }, "cpu_utilization_pct": { "n": 1, "median": 18.236472945891784, "min": 18.236472945891784, "max": 18.236472945891784 }, "proc_memory_available_gib": { "n": 1, "median": 21.22813892364502, "min": 21.22813892364502, "max": 21.22813892364502 }, "proc_memory_total_minus_available_gib": { "n": 1, "median": 30.77186107635498, "min": 30.77186107635498, "max": 30.77186107635498 }, "proc_page_cache_cached_gib": { "n": 1, "median": 17.42029571533203, "min": 17.42029571533203, "max": 17.42029571533203 }, "proc_shmem_gib": { "n": 1, "median": 0.9384422302246094, "min": 0.9384422302246094, "max": 0.9384422302246094 }, "proc_cached_minus_shmem_estimate_gib": { "n": 1, "median": 16.481853485107422, "min": 16.481853485107422, "max": 16.481853485107422 }, "proc_unevictable_gib": { "n": 1, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 1, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 1, "median": 61.87890625, "min": 61.87890625, "max": 61.87890625 }, "pswpin_counter_delta": 0, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 0, "container_memory_current_gib": { "n": 1, "median": 30.706825256347656, "min": 30.706825256347656, "max": 30.706825256347656 }, "container_memory_lifetime_peak_gib": { "n": 1, "median": 32.25005340576172, "min": 32.25005340576172, "max": 32.25005340576172 }, "container_swap_mib": { "n": 1, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 1, "median": 29.619468688964844, "min": 29.619468688964844, "max": 29.619468688964844 }, "container_memory_stat_file_gib": { "n": 1, "median": 0.9600563049316406, "min": 0.9600563049316406, "max": 0.9600563049316406 }, "container_memory_stat_shmem_gib": { "n": 1, "median": 0.9383010864257812, "min": 0.9383010864257812, "max": 0.9383010864257812 }, "container_memory_stat_file_mapped_gib": { "n": 1, "median": 0.7607040405273438, "min": 0.7607040405273438, "max": 0.7607040405273438 }, "container_memory_stat_unevictable_gib": { "n": 1, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:6": { "read_bytes": 0, "read_mib_per_s": { "n": 0, "median": null, "min": null, "max": null }, "write_bytes": 0, "write_mib_per_s": { "n": 0, "median": null, "min": null, "max": null } }, "252:0": { "read_bytes": 0, "read_mib_per_s": { "n": 0, "median": null, "min": null, "max": null }, "write_bytes": 0, "write_mib_per_s": { "n": 0, "median": null, "min": null, "max": null } }, "252:4": { "read_bytes": 0, "read_mib_per_s": { "n": 0, "median": null, "min": null, "max": null }, "write_bytes": 0, "write_mib_per_s": { "n": 0, "median": null, "min": null, "max": null } }, "252:3": { "read_bytes": 0, "read_mib_per_s": { "n": 0, "median": null, "min": null, "max": null }, "write_bytes": 0, "write_mib_per_s": { "n": 0, "median": null, "min": null, "max": null } }, "259:0": { "read_bytes": 0, "read_mib_per_s": { "n": 0, "median": null, "min": null, "max": null }, "write_bytes": 0, "write_mib_per_s": { "n": 0, "median": null, "min": null, "max": null } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 1, "median": 31748764, "min": 31748764, "max": 31748764 }, "VmHWM": { "n": 1, "median": 32311920, "min": 32311920, "max": 32311920 }, "VmSwap": { "n": 1, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "20554": { "read_bytes": 0, "write_bytes": 0, "rchar": 0, "wchar": 0 } } }, "python": { "VmRSS": { "n": 1, "median": 212224, "min": 212224, "max": 212224 }, "VmHWM": { "n": 1, "median": 214072, "min": 214072, "max": 214072 }, "VmSwap": { "n": 1, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "20458": { "read_bytes": 0, "write_bytes": 0, "rchar": 0, "wchar": 0 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "needles-32k-ctx32768", "scope": "host", "samples": 1, "start_utc": "2026-10-12T02:16:51.697254+00:00", "end_utc": "2026-10-12T02:16:51.697254+00:00", "sample_interval_s": { "n": 0, "median": null, "min": null, "max": null }, "collector_duration_s": { "n": 1, "median": 0.05576896700040379, "min": 0.05576896700040379, "max": 0.05576896700040379 }, "gpu_utilization_pct": { "n": 1, "median": 71.0, "min": 71.0, "max": 71.0 }, "vram_used_mib": { "n": 1, "median": 15504.0, "min": 15504.0, "max": 15504.0 }, "gpu_power_w": { "n": 1, "median": 134.89, "min": 134.89, "max": 134.89 }, "gpu_temperature_c": { "n": 1, "median": 45.0, "min": 45.0, "max": 45.0 }, "cpu_utilization_pct": { "n": 1, "median": 12.98917568692756, "min": 12.98917568692756, "max": 12.98917568692756 }, "proc_memory_available_gib": { "n": 1, "median": 24.427825927734375, "min": 24.427825927734375, "max": 24.427825927734375 }, "proc_memory_total_minus_available_gib": { "n": 1, "median": 34.32490921020508, "min": 34.32490921020508, "max": 34.32490921020508 }, "proc_page_cache_cached_gib": { "n": 1, "median": 17.673728942871094, "min": 17.673728942871094, "max": 17.673728942871094 }, "proc_shmem_gib": { "n": 1, "median": 1.0046463012695312, "min": 1.0046463012695312, "max": 1.0046463012695312 }, "proc_cached_minus_shmem_estimate_gib": { "n": 1, "median": 16.669082641601562, "min": 16.669082641601562, "max": 16.669082641601562 }, "proc_unevictable_gib": { "n": 1, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 1, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 1, "median": 62.78515625, "min": 62.78515625, "max": 62.78515625 }, "pswpin_counter_delta": 0, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 0, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 1, "median": 31748764, "min": 31748764, "max": 31748764 }, "VmHWM": { "n": 1, "median": 32311920, "min": 32311920, "max": 32311920 }, "VmSwap": { "n": 1, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "55503": { "read_bytes": 0, "write_bytes": 0, "rchar": 0, "wchar": 0 } } }, "python": { "VmRSS": { "n": 1, "median": 212224, "min": 212224, "max": 212224 }, "VmHWM": { "n": 1, "median": 214072, "min": 214072, "max": 214072 }, "VmSwap": { "n": 1, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "55404": { "read_bytes": 0, "write_bytes": 0, "rchar": 0, "wchar": 0 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 0, "read_mib_per_s": { "n": 0, "median": null, "min": null, "max": null }, "write_bytes": 0, "write_mib_per_s": { "n": 0, "median": null, "min": null, "max": null } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "needles-32k-ctx65536", "scope": "docker_container", "samples": 46, "start_utc": "2026-10-12T02:23:27.813446+00:00", "end_utc": "2026-10-12T02:24:12.813777+00:00", "sample_interval_s": { "n": 45, "median": 1.000000664000254, "min": 0.9999649749988748, "max": 1.0000340469996445 }, "collector_duration_s": { "n": 46, "median": 0.03967845250008395, "min": 0.036003969000375946, "max": 0.04855156600024202 }, "gpu_utilization_pct": { "n": 46, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 46, "median": 15486.0, "min": 15486.0, "max": 15486.0 }, "gpu_power_w": { "n": 46, "median": 150.14, "min": 63.12, "max": 152.62 }, "gpu_temperature_c": { "n": 46, "median": 49.0, "min": 41.0, "max": 52.0 }, "cpu_utilization_pct": { "n": 46, "median": 11.382928448796712, "min": 10.96579476861167, "max": 38.77755511022044 }, "proc_memory_available_gib": { "n": 46, "median": 20.250330448150635, "min": 20.137006759643555, "max": 20.360400199890137 }, "proc_memory_total_minus_available_gib": { "n": 46, "median": 31.749669551849365, "min": 31.639599800109863, "max": 31.862993240356445 }, "proc_page_cache_cached_gib": { "n": 46, "median": 45.925397872924805, "min": 45.92463684082031, "max": 45.9259033203125 }, "proc_shmem_gib": { "n": 46, "median": 29.222476959228516, "min": 29.222476959228516, "max": 29.222476959228516 }, "proc_cached_minus_shmem_estimate_gib": { "n": 46, "median": 16.70292091369629, "min": 16.702159881591797, "max": 16.703426361083984 }, "proc_unevictable_gib": { "n": 46, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 46, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 46, "median": 54.703125, "min": 54.703125, "max": 54.703125 }, "pswpin_counter_delta": 6, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 11, "container_memory_current_gib": { "n": 46, "median": 36.19914627075195, "min": 36.08482360839844, "max": 36.30951690673828 }, "container_memory_lifetime_peak_gib": { "n": 46, "median": 36.9876708984375, "min": 36.9876708984375, "max": 36.9876708984375 }, "container_swap_mib": { "n": 46, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 46, "median": 2.2957191467285156, "min": 2.1819915771484375, "max": 2.4059486389160156 }, "container_memory_stat_file_gib": { "n": 46, "median": 33.762535095214844, "min": 33.76227951049805, "max": 33.762535095214844 }, "container_memory_stat_shmem_gib": { "n": 46, "median": 29.222198486328125, "min": 29.222198486328125, "max": 29.222198486328125 }, "container_memory_stat_file_mapped_gib": { "n": 46, "median": 29.173648834228516, "min": 29.173507690429688, "max": 29.173648834228516 }, "container_memory_stat_unevictable_gib": { "n": 46, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 1133907968, "read_mib_per_s": { "n": 45, "median": 0.03906256414072702, "min": 0.039062043013177734, "max": 465.08472781795405 }, "write_bytes": 0, "write_mib_per_s": { "n": 45, "median": 0.0, "min": 0.0, "max": 0.0 } }, "252:3": { "read_bytes": 1133907968, "read_mib_per_s": { "n": 45, "median": 0.03906256414072702, "min": 0.039062043013177734, "max": 465.08472781795405 }, "write_bytes": 0, "write_mib_per_s": { "n": 45, "median": 0.0, "min": 0.0, "max": 0.0 } }, "252:6": { "read_bytes": 1133903872, "read_mib_per_s": { "n": 45, "median": 0.03906256414072702, "min": 0.039062043013177734, "max": 465.0808216109223 }, "write_bytes": 0, "write_mib_per_s": { "n": 45, "median": 0.0, "min": 0.0, "max": 0.0 } }, "259:0": { "read_bytes": 1133907968, "read_mib_per_s": { "n": 45, "median": 0.03906256414072702, "min": 0.039062043013177734, "max": 465.08472781795405 }, "write_bytes": 0, "write_mib_per_s": { "n": 45, "median": 0.0, "min": 0.0, "max": 0.0 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 46, "median": 32956316.0, "min": 32837096, "max": 33072036 }, "VmHWM": { "n": 46, "median": 33514628.0, "min": 33514628, "max": 33514628 }, "VmSwap": { "n": 46, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "26431": { "read_bytes": 1134538752, "write_bytes": 4096, "rchar": 1134628125, "wchar": 2554 } } }, "python": { "VmRSS": { "n": 46, "median": 219124.0, "min": 219072, "max": 219196 }, "VmHWM": { "n": 46, "median": 219124.0, "min": 219072, "max": 219196 }, "VmSwap": { "n": 46, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "26259": { "read_bytes": 0, "write_bytes": 0, "rchar": 279921, "wchar": 282438 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "needles-32k-ctx65536", "scope": "host", "samples": 46, "start_utc": "2026-10-12T02:23:27.697242+00:00", "end_utc": "2026-10-12T02:24:12.697216+00:00", "sample_interval_s": { "n": 45, "median": 0.9999997559989424, "min": 0.9997586640001828, "max": 1.0002381239992246 }, "collector_duration_s": { "n": 46, "median": 0.046098185000118974, "min": 0.042395624001073884, "max": 0.05812232199969003 }, "gpu_utilization_pct": { "n": 46, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 46, "median": 15486.0, "min": 15486.0, "max": 15486.0 }, "gpu_power_w": { "n": 46, "median": 150.14, "min": 63.12, "max": 152.62 }, "gpu_temperature_c": { "n": 46, "median": 49.0, "min": 41.0, "max": 52.0 }, "cpu_utilization_pct": { "n": 46, "median": 9.519834645043325, "min": 9.265442404006677, "max": 31.57894736842105 }, "proc_memory_available_gib": { "n": 46, "median": 23.202348709106445, "min": 22.878883361816406, "max": 23.31822967529297 }, "proc_memory_total_minus_available_gib": { "n": 46, "median": 35.55038642883301, "min": 35.434505462646484, "max": 35.87385177612305 }, "proc_page_cache_cached_gib": { "n": 46, "median": 46.184553146362305, "min": 46.18326187133789, "max": 46.18523025512695 }, "proc_shmem_gib": { "n": 46, "median": 29.288589477539062, "min": 29.288589477539062, "max": 29.28870391845703 }, "proc_cached_minus_shmem_estimate_gib": { "n": 46, "median": 16.895931243896484, "min": 16.894668579101562, "max": 16.89664077758789 }, "proc_unevictable_gib": { "n": 46, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 46, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 46, "median": 55.609375, "min": 55.609375, "max": 55.609375 }, "pswpin_counter_delta": 4, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 9, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 46, "median": 32956314.0, "min": 32828836, "max": 33187000 }, "VmHWM": { "n": 46, "median": 33514628.0, "min": 33514628, "max": 33514628 }, "VmSwap": { "n": 46, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65735": { "read_bytes": 1174659072, "write_bytes": 4096, "rchar": 1174687005, "wchar": 2554 } } }, "python": { "VmRSS": { "n": 46, "median": 219124.0, "min": 219072, "max": 219196 }, "VmHWM": { "n": 46, "median": 219124.0, "min": 219072, "max": 219196 }, "VmSwap": { "n": 46, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65561": { "read_bytes": 0, "write_bytes": 0, "rchar": 279921, "wchar": 282438 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 1177829376, "read_mib_per_s": { "n": 45, "median": 0.03906264320366322, "min": 0.03906201848249233, "max": 452.3006922372147 }, "write_bytes": 5111808, "write_mib_per_s": { "n": 45, "median": 0.0039062493203091482, "min": 0, "max": 3.5429646649646633 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "needles-64k-ctx131072", "scope": "docker_container", "samples": 83, "start_utc": "2026-10-12T02:57:24.757918+00:00", "end_utc": "2026-10-12T02:58:46.757825+00:00", "sample_interval_s": { "n": 82, "median": 1.0000002695005605, "min": 0.9996730990005744, "max": 1.0002303619985469 }, "collector_duration_s": { "n": 83, "median": 0.039132578000135254, "min": 0.03412678200038499, "max": 0.041569408000214025 }, "gpu_utilization_pct": { "n": 83, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 83, "median": 15472.0, "min": 15402.0, "max": 15480.0 }, "gpu_power_w": { "n": 83, "median": 148.79, "min": 39.01, "max": 153.27 }, "gpu_temperature_c": { "n": 83, "median": 48.0, "min": 37.0, "max": 51.0 }, "cpu_utilization_pct": { "n": 83, "median": 11.38861138861139, "min": 1.901901901901902, "max": 33.90216154721274 }, "proc_memory_available_gib": { "n": 83, "median": 19.24630641937256, "min": 18.94316864013672, "max": 19.755096435546875 }, "proc_memory_total_minus_available_gib": { "n": 83, "median": 32.75369358062744, "min": 32.244903564453125, "max": 33.05683135986328 }, "proc_page_cache_cached_gib": { "n": 83, "median": 46.285240173339844, "min": 46.28373336791992, "max": 46.2861442565918 }, "proc_shmem_gib": { "n": 83, "median": 30.187644958496094, "min": 30.187557220458984, "max": 30.187644958496094 }, "proc_cached_minus_shmem_estimate_gib": { "n": 83, "median": 16.09759521484375, "min": 16.096176147460938, "max": 16.098499298095703 }, "proc_unevictable_gib": { "n": 83, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 83, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 83, "median": 53.859375, "min": 53.859375, "max": 53.859375 }, "pswpin_counter_delta": 4, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 4, "container_memory_current_gib": { "n": 83, "median": 32.65568542480469, "min": 32.163692474365234, "max": 32.95793914794922 }, "container_memory_lifetime_peak_gib": { "n": 83, "median": 33.967227935791016, "min": 33.967227935791016, "max": 33.967227935791016 }, "container_swap_mib": { "n": 83, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 83, "median": 2.3273391723632812, "min": 1.8368186950683594, "max": 2.6300048828125 }, "container_memory_stat_file_gib": { "n": 83, "median": 30.187408447265625, "min": 30.187408447265625, "max": 30.187408447265625 }, "container_memory_stat_shmem_gib": { "n": 83, "median": 30.187366485595703, "min": 30.187366485595703, "max": 30.187366485595703 }, "container_memory_stat_file_mapped_gib": { "n": 83, "median": 29.942607879638672, "min": 29.942588806152344, "max": 29.942607879638672 }, "container_memory_stat_unevictable_gib": { "n": 83, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 2392997888, "read_mib_per_s": { "n": 82, "median": 0.03906265589906306, "min": 0.03515633750410993, "max": 426.4341836213265 }, "write_bytes": 12288, "write_mib_per_s": { "n": 82, "median": 0.0, "min": 0.0, "max": 0.007815054749208081 } }, "252:3": { "read_bytes": 2392997888, "read_mib_per_s": { "n": 82, "median": 0.03906265589906306, "min": 0.03515633750410993, "max": 426.4341836213265 }, "write_bytes": 12288, "write_mib_per_s": { "n": 82, "median": 0.0, "min": 0.0, "max": 0.007815054749208081 } }, "252:6": { "read_bytes": 2392997888, "read_mib_per_s": { "n": 82, "median": 0.03906265589906306, "min": 0.03515633750410993, "max": 426.4341836213265 }, "write_bytes": 12288, "write_mib_per_s": { "n": 82, "median": 0.0, "min": 0.0, "max": 0.007815054749208081 } }, "259:0": { "read_bytes": 2392997888, "read_mib_per_s": { "n": 82, "median": 0.03906265589906306, "min": 0.03515633750410993, "max": 426.43808984094784 }, "write_bytes": 12288, "write_mib_per_s": { "n": 82, "median": 0.0, "min": 0.0, "max": 0.007815054749208081 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 83, "median": 34001996, "min": 33472804, "max": 34318256 }, "VmHWM": { "n": 83, "median": 34287084, "min": 34177852, "max": 34318256 }, "VmSwap": { "n": 83, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "63181": { "read_bytes": 2392997888, "write_bytes": 4096, "rchar": 2393919547, "wchar": 3316 } } }, "python": { "VmRSS": { "n": 83, "median": 219784, "min": 213832, "max": 221864 }, "VmHWM": { "n": 83, "median": 220352, "min": 213832, "max": 222316 }, "VmSwap": { "n": 83, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "63086": { "read_bytes": 0, "write_bytes": 0, "rchar": 513547, "wchar": 834612 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "needles-64k-ctx131072", "scope": "host", "samples": 83, "start_utc": "2026-10-12T02:57:24.884176+00:00", "end_utc": "2026-10-12T02:58:46.884185+00:00", "sample_interval_s": { "n": 82, "median": 0.9999999555002432, "min": 0.9995827239999926, "max": 1.000414672000261 }, "collector_duration_s": { "n": 83, "median": 0.04553038099948026, "min": 0.0427113810001174, "max": 0.055218890998730785 }, "gpu_utilization_pct": { "n": 83, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 83, "median": 15472.0, "min": 15402.0, "max": 15480.0 }, "gpu_power_w": { "n": 83, "median": 148.79, "min": 39.01, "max": 153.27 }, "gpu_temperature_c": { "n": 83, "median": 48.0, "min": 37.0, "max": 51.0 }, "cpu_utilization_pct": { "n": 83, "median": 9.515859766277128, "min": 2.00836820083682, "max": 25.09124087591241 }, "proc_memory_available_gib": { "n": 83, "median": 22.38507080078125, "min": 22.184494018554688, "max": 23.01769256591797 }, "proc_memory_total_minus_available_gib": { "n": 83, "median": 36.3676643371582, "min": 35.735042572021484, "max": 36.568241119384766 }, "proc_page_cache_cached_gib": { "n": 83, "median": 46.56454849243164, "min": 46.56279373168945, "max": 46.56584930419922 }, "proc_shmem_gib": { "n": 83, "median": 30.253761291503906, "min": 30.25366973876953, "max": 30.253761291503906 }, "proc_cached_minus_shmem_estimate_gib": { "n": 83, "median": 16.310787200927734, "min": 16.309123992919922, "max": 16.312088012695312 }, "proc_unevictable_gib": { "n": 83, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 83, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 83, "median": 54.765625, "min": 54.765625, "max": 54.765625 }, "pswpin_counter_delta": 2, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 2, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 83, "median": 34001992, "min": 33472804, "max": 34318256 }, "VmHWM": { "n": 83, "median": 34287084, "min": 34177852, "max": 34318256 }, "VmSwap": { "n": 83, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "114103": { "read_bytes": 2392997888, "write_bytes": 4096, "rchar": 2393919547, "wchar": 3316 } } }, "python": { "VmRSS": { "n": 83, "median": 219784, "min": 213864, "max": 221864 }, "VmHWM": { "n": 83, "median": 220352, "min": 213864, "max": 222316 }, "VmSwap": { "n": 83, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "114005": { "read_bytes": 0, "write_bytes": 0, "rchar": 513547, "wchar": 834612 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 2399731712, "read_mib_per_s": { "n": 82, "median": 0.0390629641852005, "min": 0.0351563915396383, "max": 360.43704207789756 }, "write_bytes": 6901760, "write_mib_per_s": { "n": 82, "median": 0.003907427350376533, "min": 0, "max": 1.4882920222587916 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "needles-8k-ctx32768", "scope": "docker_container", "samples": 11, "start_utc": "2026-10-12T02:16:18.849192+00:00", "end_utc": "2026-10-12T02:16:28.849405+00:00", "sample_interval_s": { "n": 10, "median": 1.0000004840003385, "min": 0.9997751060000155, "max": 1.0002225089992862 }, "collector_duration_s": { "n": 11, "median": 0.039755776000674814, "min": 0.036672288000772824, "max": 0.041999773999123136 }, "gpu_utilization_pct": { "n": 11, "median": 100.0, "min": 59.0, "max": 100.0 }, "vram_used_mib": { "n": 11, "median": 15504.0, "min": 15504.0, "max": 15504.0 }, "gpu_power_w": { "n": 11, "median": 141.86, "min": 18.26, "max": 144.15 }, "gpu_temperature_c": { "n": 11, "median": 46.0, "min": 40.0, "max": 48.0 }, "cpu_utilization_pct": { "n": 11, "median": 11.646586345381525, "min": 11.122244488977955, "max": 24.67248908296943 }, "proc_memory_available_gib": { "n": 11, "median": 21.329002380371094, "min": 21.22036075592041, "max": 21.332813262939453 }, "proc_memory_total_minus_available_gib": { "n": 11, "median": 30.670997619628906, "min": 30.667186737060547, "max": 30.77963924407959 }, "proc_page_cache_cached_gib": { "n": 11, "median": 17.39868927001953, "min": 17.398578643798828, "max": 17.39883041381836 }, "proc_shmem_gib": { "n": 11, "median": 0.9385337829589844, "min": 0.9385337829589844, "max": 0.9385337829589844 }, "proc_cached_minus_shmem_estimate_gib": { "n": 11, "median": 16.460155487060547, "min": 16.460044860839844, "max": 16.460296630859375 }, "proc_unevictable_gib": { "n": 11, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 11, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 11, "median": 61.87890625, "min": 61.87890625, "max": 61.87890625 }, "pswpin_counter_delta": 0, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 4, "container_memory_current_gib": { "n": 11, "median": 30.59326171875, "min": 30.592689514160156, "max": 30.70291519165039 }, "container_memory_lifetime_peak_gib": { "n": 11, "median": 32.25005340576172, "min": 32.25005340576172, "max": 32.25005340576172 }, "container_swap_mib": { "n": 11, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 11, "median": 29.50594711303711, "min": 29.505916595458984, "max": 29.61553955078125 }, "container_memory_stat_file_gib": { "n": 11, "median": 0.9600486755371094, "min": 0.9600486755371094, "max": 0.960052490234375 }, "container_memory_stat_shmem_gib": { "n": 11, "median": 0.9383010864257812, "min": 0.9383010864257812, "max": 0.9383010864257812 }, "container_memory_stat_file_mapped_gib": { "n": 11, "median": 0.7607002258300781, "min": 0.7607002258300781, "max": 0.7607040405273438 }, "container_memory_stat_unevictable_gib": { "n": 11, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:6": { "read_bytes": 28860416, "read_mib_per_s": { "n": 10, "median": 0.039062605234770856, "min": 0.039053810175779476, "max": 25.417759383899263 }, "write_bytes": 0, "write_mib_per_s": { "n": 10, "median": 0.0, "min": 0.0, "max": 0.0 } }, "252:0": { "read_bytes": 0, "read_mib_per_s": { "n": 10, "median": 0.0, "min": 0.0, "max": 0.0 }, "write_bytes": 0, "write_mib_per_s": { "n": 10, "median": 0.0, "min": 0.0, "max": 0.0 } }, "252:4": { "read_bytes": 28860416, "read_mib_per_s": { "n": 10, "median": 0.039062605234770856, "min": 0.039053810175779476, "max": 25.417759383899263 }, "write_bytes": 0, "write_mib_per_s": { "n": 10, "median": 0.0, "min": 0.0, "max": 0.0 } }, "252:3": { "read_bytes": 28860416, "read_mib_per_s": { "n": 10, "median": 0.039062605234770856, "min": 0.039053810175779476, "max": 25.417759383899263 }, "write_bytes": 0, "write_mib_per_s": { "n": 10, "median": 0.0, "min": 0.0, "max": 0.0 } }, "259:0": { "read_bytes": 28860416, "read_mib_per_s": { "n": 10, "median": 0.039062605234770856, "min": 0.039053810175779476, "max": 25.417759383899263 }, "write_bytes": 0, "write_mib_per_s": { "n": 10, "median": 0.0, "min": 0.0, "max": 0.0 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 11, "median": 31630380, "min": 31630380, "max": 31745296 }, "VmHWM": { "n": 11, "median": 32311920, "min": 32311920, "max": 32311920 }, "VmSwap": { "n": 11, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "20554": { "read_bytes": 29421568, "write_bytes": 4096, "rchar": 29550160, "wchar": 2048 } } }, "python": { "VmRSS": { "n": 11, "median": 211568, "min": 211536, "max": 211568 }, "VmHWM": { "n": 11, "median": 214072, "min": 214072, "max": 214072 }, "VmSwap": { "n": 11, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "20458": { "read_bytes": 0, "write_bytes": 0, "rchar": 59624, "wchar": 62376 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ2_XS", "phase": "needles-8k-ctx32768", "scope": "host", "samples": 11, "start_utc": "2026-10-12T02:16:18.697220+00:00", "end_utc": "2026-10-12T02:16:28.697266+00:00", "sample_interval_s": { "n": 10, "median": 1.0000003084996933, "min": 0.9999856079994061, "max": 1.0000138240011438 }, "collector_duration_s": { "n": 11, "median": 0.046527160000550793, "min": 0.0452845449999586, "max": 0.06316382300065015 }, "gpu_utilization_pct": { "n": 11, "median": 100.0, "min": 59.0, "max": 100.0 }, "vram_used_mib": { "n": 11, "median": 15504.0, "min": 15504.0, "max": 15504.0 }, "gpu_power_w": { "n": 11, "median": 141.86, "min": 18.26, "max": 144.15 }, "gpu_temperature_c": { "n": 11, "median": 46.0, "min": 40.0, "max": 48.0 }, "cpu_utilization_pct": { "n": 11, "median": 10.091743119266056, "min": 9.341117597998332, "max": 17.970049916805323 }, "proc_memory_available_gib": { "n": 11, "median": 24.438800811767578, "min": 24.39879608154297, "max": 24.555984497070312 }, "proc_memory_total_minus_available_gib": { "n": 11, "median": 34.313934326171875, "min": 34.19675064086914, "max": 34.353939056396484 }, "proc_page_cache_cached_gib": { "n": 11, "median": 17.651668548583984, "min": 17.651351928710938, "max": 17.651893615722656 }, "proc_shmem_gib": { "n": 11, "median": 1.0046463012695312, "min": 1.0046463012695312, "max": 1.0046577453613281 }, "proc_cached_minus_shmem_estimate_gib": { "n": 11, "median": 16.647022247314453, "min": 16.646705627441406, "max": 16.647235870361328 }, "proc_unevictable_gib": { "n": 11, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 11, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 11, "median": 62.78515625, "min": 62.78515625, "max": 62.78515625 }, "pswpin_counter_delta": 0, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 4, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 11, "median": 31630380, "min": 31630380, "max": 31644208 }, "VmHWM": { "n": 11, "median": 32311920, "min": 32311920, "max": 32311920 }, "VmSwap": { "n": 11, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "55503": { "read_bytes": 99696640, "write_bytes": 4096, "rchar": 99825232, "wchar": 2048 } } }, "python": { "VmRSS": { "n": 11, "median": 211568, "min": 211536, "max": 211568 }, "VmHWM": { "n": 11, "median": 214072, "min": 214072, "max": 214072 }, "VmSwap": { "n": 11, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "55404": { "read_bytes": 0, "write_bytes": 0, "rchar": 66626, "wchar": 62376 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 99954688, "read_mib_per_s": { "n": 10, "median": 0.060546570963016313, "min": 0.03906234742248582, "max": 92.39326056543844 }, "write_bytes": 491520, "write_mib_per_s": { "n": 10, "median": 0.0039062498828724276, "min": 0, "max": 0.36718242406974966 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "practical", "scope": "docker_container", "samples": 768, "start_utc": "2026-10-12T03:12:21.783955+00:00", "end_utc": "2026-10-12T03:25:08.784211+00:00", "sample_interval_s": { "n": 767, "median": 0.9999998879993655, "min": 0.9996187639990239, "max": 1.0003815000000031 }, "collector_duration_s": { "n": 768, "median": 0.04395804799878533, "min": 0.035776290998910554, "max": 0.052584061999368714 }, "gpu_utilization_pct": { "n": 768, "median": 93.0, "min": 42.0, "max": 100.0 }, "vram_used_mib": { "n": 768, "median": 15520.0, "min": 15520.0, "max": 15520.0 }, "gpu_power_w": { "n": 768, "median": 105.455, "min": 64.59, "max": 146.11 }, "gpu_temperature_c": { "n": 768, "median": 43.0, "min": 39.0, "max": 50.0 }, "cpu_utilization_pct": { "n": 768, "median": 56.83417775730689, "min": 10.542168674698795, "max": 67.90744466800805 }, "proc_memory_available_gib": { "n": 768, "median": 13.114488124847412, "min": 12.782267570495605, "max": 13.232686042785645 }, "proc_memory_total_minus_available_gib": { "n": 768, "median": 38.88551187515259, "min": 38.767313957214355, "max": 39.217732429504395 }, "proc_page_cache_cached_gib": { "n": 768, "median": 45.56583786010742, "min": 45.5568962097168, "max": 45.57476806640625 }, "proc_shmem_gib": { "n": 768, "median": 36.314029693603516, "min": 36.314029693603516, "max": 36.31403732299805 }, "proc_cached_minus_shmem_estimate_gib": { "n": 768, "median": 9.251808166503906, "min": 9.242866516113281, "max": 9.260738372802734 }, "proc_unevictable_gib": { "n": 768, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 768, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 768, "median": 53.6796875, "min": 53.6796875, "max": 53.6953125 }, "pswpin_counter_delta": 5, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 15, "container_memory_current_gib": { "n": 768, "median": 39.73598861694336, "min": 39.618980407714844, "max": 40.0657844543457 }, "container_memory_lifetime_peak_gib": { "n": 768, "median": 42.38410568237305, "min": 42.38410568237305, "max": 42.38410568237305 }, "container_swap_mib": { "n": 768, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 768, "median": 2.3202896118164062, "min": 2.204273223876953, "max": 2.649517059326172 }, "container_memory_stat_file_gib": { "n": 768, "median": 37.24778366088867, "min": 37.24762725830078, "max": 37.247825622558594 }, "container_memory_stat_shmem_gib": { "n": 768, "median": 36.31375503540039, "min": 36.31375503540039, "max": 36.31375503540039 }, "container_memory_stat_file_mapped_gib": { "n": 768, "median": 36.09004592895508, "min": 36.09002685546875, "max": 36.09004592895508 }, "container_memory_stat_unevictable_gib": { "n": 768, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 1071927296, "read_mib_per_s": { "n": 767, "median": 0.7070326131586091, "min": 0.035155782498392386, "max": 7.289056187671174 }, "write_bytes": 163840, "write_mib_per_s": { "n": 767, "median": 0.0, "min": 0.0, "max": 0.0117189346552457 } }, "252:3": { "read_bytes": 1071927296, "read_mib_per_s": { "n": 767, "median": 0.7070326131586091, "min": 0.035155782498392386, "max": 7.289056187671174 }, "write_bytes": 163840, "write_mib_per_s": { "n": 767, "median": 0.0, "min": 0.0, "max": 0.0117189346552457 } }, "252:6": { "read_bytes": 1071927296, "read_mib_per_s": { "n": 767, "median": 0.7070326131586091, "min": 0.035155782498392386, "max": 7.289056187671174 }, "write_bytes": 163840, "write_mib_per_s": { "n": 767, "median": 0.0, "min": 0.0, "max": 0.0117189346552457 } }, "259:0": { "read_bytes": 1071927296, "read_mib_per_s": { "n": 767, "median": 0.7070326131586091, "min": 0.035155782498392386, "max": 7.289056187671174 }, "write_bytes": 163840, "write_mib_per_s": { "n": 767, "median": 0.0, "min": 0.0, "max": 0.0117189346552457 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 768, "median": 40415156.0, "min": 40297692, "max": 40759904 }, "VmHWM": { "n": 768, "median": 40968636.0, "min": 40968636, "max": 40968636 }, "VmSwap": { "n": 768, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "68203": { "read_bytes": 1071927296, "write_bytes": 172032, "rchar": 1074317174, "wchar": 349666 } } }, "python": { "VmRSS": { "n": 768, "median": 224360.0, "min": 220132, "max": 224836 }, "VmHWM": { "n": 768, "median": 224360.0, "min": 220132, "max": 224836 }, "VmSwap": { "n": 768, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "68107": { "read_bytes": 0, "write_bytes": 0, "rchar": 5047955, "wchar": 2186664 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "practical", "scope": "host", "samples": 768, "start_utc": "2026-10-12T03:12:21.220054+00:00", "end_utc": "2026-10-12T03:25:08.220160+00:00", "sample_interval_s": { "n": 767, "median": 1.0000000159998308, "min": 0.9996959389991389, "max": 1.0003128460011794 }, "collector_duration_s": { "n": 768, "median": 0.05627136299972335, "min": 0.0444519640004728, "max": 0.07445190399994317 }, "gpu_utilization_pct": { "n": 768, "median": 93.0, "min": 40.0, "max": 100.0 }, "vram_used_mib": { "n": 768, "median": 15520.0, "min": 15520.0, "max": 15520.0 }, "gpu_power_w": { "n": 768, "median": 105.265, "min": 65.02, "max": 144.3 }, "gpu_temperature_c": { "n": 768, "median": 43.0, "min": 39.0, "max": 51.0 }, "cpu_utilization_pct": { "n": 768, "median": 47.61904761904762, "min": 9.189640768588138, "max": 57.28643216080402 }, "proc_memory_available_gib": { "n": 768, "median": 16.261465072631836, "min": 15.857059478759766, "max": 16.490184783935547 }, "proc_memory_total_minus_available_gib": { "n": 768, "median": 42.49127006530762, "min": 42.262550354003906, "max": 42.89567565917969 }, "proc_page_cache_cached_gib": { "n": 768, "median": 45.86266899108887, "min": 45.84303283691406, "max": 45.87541961669922 }, "proc_shmem_gib": { "n": 768, "median": 36.38014602661133, "min": 36.38014602661133, "max": 36.380348205566406 }, "proc_cached_minus_shmem_estimate_gib": { "n": 768, "median": 9.482522964477539, "min": 9.462886810302734, "max": 9.49527359008789 }, "proc_unevictable_gib": { "n": 768, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 768, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 768, "median": 54.5859375, "min": 54.5859375, "max": 54.6015625 }, "pswpin_counter_delta": 5, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 16, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 768, "median": 40415156.0, "min": 40297692, "max": 40759904 }, "VmHWM": { "n": 768, "median": 40968636.0, "min": 40968636, "max": 40968636 }, "VmSwap": { "n": 768, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "122130": { "read_bytes": 1072128000, "write_bytes": 172032, "rchar": 1074519098, "wchar": 350377 } } }, "python": { "VmRSS": { "n": 768, "median": 224360.0, "min": 220132, "max": 224836 }, "VmHWM": { "n": 768, "median": 224360.0, "min": 220132, "max": 224836 }, "VmSwap": { "n": 768, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "122032": { "read_bytes": 0, "write_bytes": 0, "rchar": 5042394, "wchar": 2187687 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 1140203520, "read_mib_per_s": { "n": 767, "median": 0.8867309975280517, "min": 0.03515632762518182, "max": 9.917770374753164 }, "write_bytes": 92090368, "write_mib_per_s": { "n": 767, "median": 0.01562501267188627, "min": 0, "max": 5.578120565392852 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "dossier", "scope": "docker_container", "samples": 118, "start_utc": "2026-10-12T03:25:09.784341+00:00", "end_utc": "2026-10-12T03:27:06.784078+00:00", "sample_interval_s": { "n": 117, "median": 1.0000009240011423, "min": 0.9995221640001546, "max": 1.0004678829991462 }, "collector_duration_s": { "n": 118, "median": 0.04422225849975803, "min": 0.03752183299911849, "max": 0.050524518001111574 }, "gpu_utilization_pct": { "n": 118, "median": 91.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 118, "median": 15520.0, "min": 15520.0, "max": 15520.0 }, "gpu_power_w": { "n": 118, "median": 107.805, "min": 65.68, "max": 144.36 }, "gpu_temperature_c": { "n": 118, "median": 44.0, "min": 40.0, "max": 50.0 }, "cpu_utilization_pct": { "n": 118, "median": 57.142857142857146, "min": 10.954773869346734, "max": 67.33870967741936 }, "proc_memory_available_gib": { "n": 118, "median": 12.644939422607422, "min": 12.639042854309082, "max": 13.085127830505371 }, "proc_memory_total_minus_available_gib": { "n": 118, "median": 39.35506057739258, "min": 38.91487216949463, "max": 39.36095714569092 }, "proc_page_cache_cached_gib": { "n": 118, "median": 45.58668518066406, "min": 45.58528137207031, "max": 45.587982177734375 }, "proc_shmem_gib": { "n": 118, "median": 36.314029693603516, "min": 36.314029693603516, "max": 36.314029693603516 }, "proc_cached_minus_shmem_estimate_gib": { "n": 118, "median": 9.272655487060547, "min": 9.271251678466797, "max": 9.27395248413086 }, "proc_unevictable_gib": { "n": 118, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 118, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 118, "median": 53.6796875, "min": 53.6796875, "max": 53.6796875 }, "pswpin_counter_delta": 0, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 4, "container_memory_current_gib": { "n": 118, "median": 40.06865119934082, "min": 39.6265754699707, "max": 40.06904220581055 }, "container_memory_lifetime_peak_gib": { "n": 118, "median": 42.38410568237305, "min": 42.38410568237305, "max": 42.38410568237305 }, "container_swap_mib": { "n": 118, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 118, "median": 2.6523170471191406, "min": 2.21112060546875, "max": 2.6525535583496094 }, "container_memory_stat_file_gib": { "n": 118, "median": 37.24782943725586, "min": 37.247825622558594, "max": 37.247833251953125 }, "container_memory_stat_shmem_gib": { "n": 118, "median": 36.31375503540039, "min": 36.31375503540039, "max": 36.31375503540039 }, "container_memory_stat_file_mapped_gib": { "n": 118, "median": 36.09004592895508, "min": 36.09004592895508, "max": 36.09004592895508 }, "container_memory_stat_unevictable_gib": { "n": 118, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 190119936, "read_mib_per_s": { "n": 117, "median": 1.1015695588572214, "min": 0.03515580007606462, "max": 17.18746249693138 }, "write_bytes": 12288, "write_mib_per_s": { "n": 117, "median": 0.0, "min": 0.0, "max": 0.00781248567970532 } }, "252:3": { "read_bytes": 190119936, "read_mib_per_s": { "n": 117, "median": 1.1015695588572214, "min": 0.03515580007606462, "max": 17.18746249693138 }, "write_bytes": 12288, "write_mib_per_s": { "n": 117, "median": 0.0, "min": 0.0, "max": 0.00781248567970532 } }, "252:6": { "read_bytes": 190119936, "read_mib_per_s": { "n": 117, "median": 1.1015695588572214, "min": 0.03515580007606462, "max": 17.18746249693138 }, "write_bytes": 12288, "write_mib_per_s": { "n": 117, "median": 0.0, "min": 0.0, "max": 0.00781248567970532 } }, "259:0": { "read_bytes": 190119936, "read_mib_per_s": { "n": 117, "median": 1.1015695588572214, "min": 0.03515580007606462, "max": 17.18746249693138 }, "write_bytes": 12288, "write_mib_per_s": { "n": 117, "median": 0.0, "min": 0.0, "max": 0.00781248567970532 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 118, "median": 40762204.0, "min": 40300240, "max": 40762204 }, "VmHWM": { "n": 118, "median": 40968636.0, "min": 40968636, "max": 40968636 }, "VmSwap": { "n": 118, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "68203": { "read_bytes": 190119936, "write_bytes": 20480, "rchar": 191678341, "wchar": 48575 } } }, "python": { "VmRSS": { "n": 118, "median": 225468.0, "min": 224804, "max": 225716 }, "VmHWM": { "n": 118, "median": 225468.0, "min": 224804, "max": 225716 }, "VmSwap": { "n": 118, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "68107": { "read_bytes": 0, "write_bytes": 0, "rchar": 764115, "wchar": 1509894 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "dossier", "scope": "host", "samples": 118, "start_utc": "2026-10-12T03:25:09.220015+00:00", "end_utc": "2026-10-12T03:27:06.220133+00:00", "sample_interval_s": { "n": 117, "median": 1.000000436000846, "min": 0.9999579029990855, "max": 1.000035775001379 }, "collector_duration_s": { "n": 118, "median": 0.05644977649990324, "min": 0.045222453998576384, "max": 0.06443492899961711 }, "gpu_utilization_pct": { "n": 118, "median": 94.0, "min": 60.0, "max": 100.0 }, "vram_used_mib": { "n": 118, "median": 15520.0, "min": 15520.0, "max": 15520.0 }, "gpu_power_w": { "n": 118, "median": 107.37, "min": 63.56, "max": 143.72 }, "gpu_temperature_c": { "n": 118, "median": 44.0, "min": 41.0, "max": 49.0 }, "cpu_utilization_pct": { "n": 118, "median": 47.93320251100762, "min": 9.265442404006677, "max": 56.67506297229219 }, "proc_memory_available_gib": { "n": 118, "median": 15.810819625854492, "min": 15.681648254394531, "max": 16.22928237915039 }, "proc_memory_total_minus_available_gib": { "n": 118, "median": 42.94191551208496, "min": 42.52345275878906, "max": 43.07108688354492 }, "proc_page_cache_cached_gib": { "n": 118, "median": 45.88791465759277, "min": 45.87548065185547, "max": 45.89185333251953 }, "proc_shmem_gib": { "n": 118, "median": 36.38014602661133, "min": 36.38014602661133, "max": 36.3829345703125 }, "proc_cached_minus_shmem_estimate_gib": { "n": 118, "median": 9.507766723632812, "min": 9.49533462524414, "max": 9.509647369384766 }, "proc_unevictable_gib": { "n": 118, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 118, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 118, "median": 54.5859375, "min": 54.5859375, "max": 54.5859375 }, "pswpin_counter_delta": 3, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 13, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 118, "median": 40762204.0, "min": 40300240, "max": 40762204 }, "VmHWM": { "n": 118, "median": 40968636.0, "min": 40968636, "max": 40968636 }, "VmSwap": { "n": 118, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "122130": { "read_bytes": 189685760, "write_bytes": 20480, "rchar": 191244462, "wchar": 48307 } } }, "python": { "VmRSS": { "n": 118, "median": 225468.0, "min": 224792, "max": 225716 }, "VmHWM": { "n": 118, "median": 225468.0, "min": 224792, "max": 225716 }, "VmSwap": { "n": 118, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "122032": { "read_bytes": 0, "write_bytes": 0, "rchar": 758632, "wchar": 1510339 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 210255872, "read_mib_per_s": { "n": 117, "median": 1.1406277694439833, "min": 0.03515614875027523, "max": 17.281695090057465 }, "write_bytes": 10178560, "write_mib_per_s": { "n": 117, "median": 0.0039062114691325424, "min": 0, "max": 2.1368030319236704 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "humaneval", "scope": "docker_container", "samples": 1164, "start_utc": "2026-10-12T03:27:07.784333+00:00", "end_utc": "2026-10-12T03:46:30.784031+00:00", "sample_interval_s": { "n": 1163, "median": 1.0000000379986886, "min": 0.9996411249994708, "max": 1.0003639810001914 }, "collector_duration_s": { "n": 1164, "median": 0.045280807000381174, "min": 0.03662332699968829, "max": 0.05606809500022791 }, "gpu_utilization_pct": { "n": 1164, "median": 90.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 1164, "median": 15520.0, "min": 15520.0, "max": 15520.0 }, "gpu_power_w": { "n": 1164, "median": 96.03, "min": 39.69, "max": 128.98 }, "gpu_temperature_c": { "n": 1164, "median": 42.0, "min": 39.0, "max": 45.0 }, "cpu_utilization_pct": { "n": 1164, "median": 57.429718875502004, "min": 16.265060240963855, "max": 68.40521564694082 }, "proc_memory_available_gib": { "n": 1164, "median": 13.005033493041992, "min": 12.762073516845703, "max": 13.11612319946289 }, "proc_memory_total_minus_available_gib": { "n": 1164, "median": 38.99496650695801, "min": 38.88387680053711, "max": 39.2379264831543 }, "proc_page_cache_cached_gib": { "n": 1164, "median": 45.67804718017578, "min": 45.630916595458984, "max": 45.691627502441406 }, "proc_shmem_gib": { "n": 1164, "median": 36.314029693603516, "min": 36.314029693603516, "max": 36.31403350830078 }, "proc_cached_minus_shmem_estimate_gib": { "n": 1164, "median": 9.364017486572266, "min": 9.316886901855469, "max": 9.37759780883789 }, "proc_unevictable_gib": { "n": 1164, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 1164, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 1164, "median": 53.5859375, "min": 53.53125, "max": 53.6796875 }, "pswpin_counter_delta": 17, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 227, "container_memory_current_gib": { "n": 1164, "median": 39.73982238769531, "min": 39.629371643066406, "max": 40.06867599487305 }, "container_memory_lifetime_peak_gib": { "n": 1164, "median": 42.38410568237305, "min": 42.38410568237305, "max": 42.38410568237305 }, "container_swap_mib": { "n": 1164, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 1164, "median": 2.32379150390625, "min": 2.214153289794922, "max": 2.6525115966796875 }, "container_memory_stat_file_gib": { "n": 1164, "median": 37.2481689453125, "min": 37.247833251953125, "max": 37.24826431274414 }, "container_memory_stat_shmem_gib": { "n": 1164, "median": 36.31375503540039, "min": 36.31375503540039, "max": 36.31375503540039 }, "container_memory_stat_file_mapped_gib": { "n": 1164, "median": 36.09011459350586, "min": 36.09004592895508, "max": 36.09011459350586 }, "container_memory_stat_unevictable_gib": { "n": 1164, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 2437685248, "read_mib_per_s": { "n": 1163, "median": 1.8241795666225185, "min": 0.039060747930153863, "max": 13.133181634344297 }, "write_bytes": 344064, "write_mib_per_s": { "n": 1163, "median": 0.0, "min": 0.0, "max": 0.011719051554643067 } }, "252:3": { "read_bytes": 2437685248, "read_mib_per_s": { "n": 1163, "median": 1.8241795666225185, "min": 0.039060747930153863, "max": 13.133181634344297 }, "write_bytes": 344064, "write_mib_per_s": { "n": 1163, "median": 0.0, "min": 0.0, "max": 0.011719051554643067 } }, "252:6": { "read_bytes": 2437685248, "read_mib_per_s": { "n": 1163, "median": 1.8241795666225185, "min": 0.039060747930153863, "max": 13.133181634344297 }, "write_bytes": 344064, "write_mib_per_s": { "n": 1163, "median": 0.0, "min": 0.0, "max": 0.011719051554643067 } }, "259:0": { "read_bytes": 2437685248, "read_mib_per_s": { "n": 1163, "median": 1.8241795666225185, "min": 0.039060747930153863, "max": 13.133181634344297 }, "write_bytes": 344064, "write_mib_per_s": { "n": 1163, "median": 0.0, "min": 0.0, "max": 0.011719051554643067 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 1164, "median": 40417548.0, "min": 40302540, "max": 40762204 }, "VmHWM": { "n": 1164, "median": 40968636.0, "min": 40968636, "max": 40968636 }, "VmSwap": { "n": 1164, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "68203": { "read_bytes": 2437685248, "write_bytes": 352256, "rchar": 2438415853, "wchar": 694947 } } }, "python": { "VmRSS": { "n": 1164, "median": 225720.0, "min": 225672, "max": 225724 }, "VmHWM": { "n": 1164, "median": 225720.0, "min": 225672, "max": 225724 }, "VmSwap": { "n": 1164, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "68107": { "read_bytes": 0, "write_bytes": 0, "rchar": 7829023, "wchar": 477558 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "humaneval", "scope": "host", "samples": 1164, "start_utc": "2026-10-12T03:27:07.220027+00:00", "end_utc": "2026-10-12T03:46:30.220110+00:00", "sample_interval_s": { "n": 1163, "median": 0.9999995280013536, "min": 0.9998926919997757, "max": 1.000111506000394 }, "collector_duration_s": { "n": 1164, "median": 0.058568653499605716, "min": 0.04570666400104528, "max": 0.260055378999823 }, "gpu_utilization_pct": { "n": 1164, "median": 90.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 1164, "median": 15520.0, "min": 15520.0, "max": 15520.0 }, "gpu_power_w": { "n": 1164, "median": 96.11, "min": 51.1, "max": 131.79 }, "gpu_temperature_c": { "n": 1164, "median": 42.0, "min": 39.0, "max": 44.0 }, "cpu_utilization_pct": { "n": 1164, "median": 48.24561281064874, "min": 8.915054667788057, "max": 58.20770519262982 }, "proc_memory_available_gib": { "n": 1164, "median": 16.15581512451172, "min": 15.81131362915039, "max": 16.321033477783203 }, "proc_memory_total_minus_available_gib": { "n": 1164, "median": 42.596920013427734, "min": 42.43170166015625, "max": 42.94142150878906 }, "proc_page_cache_cached_gib": { "n": 1164, "median": 45.98533821105957, "min": 45.889808654785156, "max": 46.05254364013672 }, "proc_shmem_gib": { "n": 1164, "median": 36.38014602661133, "min": 36.38008117675781, "max": 36.39004135131836 }, "proc_cached_minus_shmem_estimate_gib": { "n": 1164, "median": 9.605157852172852, "min": 9.509662628173828, "max": 9.67239761352539 }, "proc_unevictable_gib": { "n": 1164, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 1164, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 1164, "median": 54.4921875, "min": 54.4375, "max": 54.5859375 }, "pswpin_counter_delta": 17, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 590, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 1164, "median": 40417548.0, "min": 40302540, "max": 40762204 }, "VmHWM": { "n": 1164, "median": 40968636.0, "min": 40968636, "max": 40968636 }, "VmSwap": { "n": 1164, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "122130": { "read_bytes": 2437513216, "write_bytes": 352256, "rchar": 2438243817, "wchar": 694713 } } }, "python": { "VmRSS": { "n": 1164, "median": 225720.0, "min": 225672, "max": 225720 }, "VmHWM": { "n": 1164, "median": 225720.0, "min": 225672, "max": 225720 }, "VmSwap": { "n": 1164, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "122032": { "read_bytes": 0, "write_bytes": 0, "rchar": 7828786, "wchar": 477498 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 2661376000, "read_mib_per_s": { "n": 1163, "median": 1.914068081422688, "min": 0.035156219941446076, "max": 73.93739545248529 }, "write_bytes": 103759872, "write_mib_per_s": { "n": 1163, "median": 0.007812595352730423, "min": 0, "max": 4.074212989060765 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "official-optional-65536", "scope": "docker_container", "samples": 190, "start_utc": "2026-10-12T03:07:12.784257+00:00", "end_utc": "2026-10-12T03:10:21.784154+00:00", "sample_interval_s": { "n": 189, "median": 1.0000004459998308, "min": 0.9996566879999591, "max": 1.0003511189988785 }, "collector_duration_s": { "n": 190, "median": 0.04029176799940615, "min": 0.03585828399991442, "max": 0.0514429270006076 }, "gpu_utilization_pct": { "n": 190, "median": 100.0, "min": 3.0, "max": 100.0 }, "vram_used_mib": { "n": 190, "median": 15518.0, "min": 15428.0, "max": 15518.0 }, "gpu_power_w": { "n": 190, "median": 104.265, "min": 40.4, "max": 141.71 }, "gpu_temperature_c": { "n": 190, "median": 44.5, "min": 38.0, "max": 53.0 }, "cpu_utilization_pct": { "n": 190, "median": 11.841448038895804, "min": 6.393606393606394, "max": 60.98294884653962 }, "proc_memory_available_gib": { "n": 190, "median": 13.152563095092773, "min": 12.587141990661621, "max": 13.675822257995605 }, "proc_memory_total_minus_available_gib": { "n": 190, "median": 38.84743690490723, "min": 38.324177742004395, "max": 39.41285800933838 }, "proc_page_cache_cached_gib": { "n": 190, "median": 45.54359245300293, "min": 45.483306884765625, "max": 45.54743957519531 }, "proc_shmem_gib": { "n": 190, "median": 36.31404495239258, "min": 36.26514434814453, "max": 36.31404495239258 }, "proc_cached_minus_shmem_estimate_gib": { "n": 190, "median": 9.229547500610352, "min": 9.218162536621094, "max": 9.233394622802734 }, "proc_unevictable_gib": { "n": 190, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 190, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 190, "median": 53.703125, "min": 53.6953125, "max": 53.7890625 }, "pswpin_counter_delta": 2, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 137, "container_memory_current_gib": { "n": 190, "median": 39.69181442260742, "min": 39.181396484375, "max": 40.257747650146484 }, "container_memory_lifetime_peak_gib": { "n": 190, "median": 42.38410568237305, "min": 42.38410568237305, "max": 42.38410568237305 }, "container_swap_mib": { "n": 190, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 190, "median": 2.2766170501708984, "min": 1.8207550048828125, "max": 2.8418006896972656 }, "container_memory_stat_file_gib": { "n": 190, "median": 37.24729919433594, "min": 37.19394302368164, "max": 37.247314453125 }, "container_memory_stat_shmem_gib": { "n": 190, "median": 36.31375503540039, "min": 36.26495361328125, "max": 36.31375503540039 }, "container_memory_stat_file_mapped_gib": { "n": 190, "median": 36.089866638183594, "min": 36.038509368896484, "max": 36.089866638183594 }, "container_memory_stat_unevictable_gib": { "n": 190, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 951660544, "read_mib_per_s": { "n": 189, "median": 0.03906286176112068, "min": 0.035155804611137355, "max": 136.98477307790128 }, "write_bytes": 65536, "write_mib_per_s": { "n": 189, "median": 0.0, "min": 0.0, "max": 0.015625000671860817 } }, "252:3": { "read_bytes": 951660544, "read_mib_per_s": { "n": 189, "median": 0.03906286176112068, "min": 0.035155804611137355, "max": 136.98477307790128 }, "write_bytes": 65536, "write_mib_per_s": { "n": 189, "median": 0.0, "min": 0.0, "max": 0.015625000671860817 } }, "252:6": { "read_bytes": 951660544, "read_mib_per_s": { "n": 189, "median": 0.03906286176112068, "min": 0.035155804611137355, "max": 136.98477307790128 }, "write_bytes": 65536, "write_mib_per_s": { "n": 189, "median": 0.0, "min": 0.0, "max": 0.015625000671860817 } }, "259:0": { "read_bytes": 951660544, "read_mib_per_s": { "n": 189, "median": 0.03906286176112068, "min": 0.035155804611137355, "max": 136.98477307790128 }, "write_bytes": 65536, "write_mib_per_s": { "n": 189, "median": 0.0, "min": 0.0, "max": 0.015625000671860817 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 190, "median": 40377572.0, "min": 39828164, "max": 40970028 }, "VmHWM": { "n": 190, "median": 40968536.0, "min": 40659900, "max": 40970028 }, "VmSwap": { "n": 190, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "68203": { "read_bytes": 951517184, "write_bytes": 65536, "rchar": 948439412, "wchar": 64764 } } }, "python": { "VmRSS": { "n": 190, "median": 215824.0, "min": 213820, "max": 218040 }, "VmHWM": { "n": 190, "median": 215824.0, "min": 213820, "max": 218304 }, "VmSwap": { "n": 190, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "68107": { "read_bytes": 143360, "write_bytes": 0, "rchar": 1242236, "wchar": 1503828 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "official-optional-65536", "scope": "host", "samples": 190, "start_utc": "2026-10-12T03:07:12.220070+00:00", "end_utc": "2026-10-12T03:10:21.220021+00:00", "sample_interval_s": { "n": 189, "median": 0.9999998870007403, "min": 0.999687907000407, "max": 1.0003017409999302 }, "collector_duration_s": { "n": 190, "median": 0.05500905349981622, "min": 0.04535005199977604, "max": 0.06933316799950262 }, "gpu_utilization_pct": { "n": 190, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 190, "median": 15518.0, "min": 15428.0, "max": 15518.0 }, "gpu_power_w": { "n": 190, "median": 104.89500000000001, "min": 39.7, "max": 140.97 }, "gpu_temperature_c": { "n": 190, "median": 44.0, "min": 37.0, "max": 52.0 }, "cpu_utilization_pct": { "n": 190, "median": 10.18363939899833, "min": 4.606365159128978, "max": 52.13389121338912 }, "proc_memory_available_gib": { "n": 190, "median": 16.28888511657715, "min": 15.725852966308594, "max": 16.849693298339844 }, "proc_memory_total_minus_available_gib": { "n": 190, "median": 42.463850021362305, "min": 41.90304183959961, "max": 43.02688217163086 }, "proc_page_cache_cached_gib": { "n": 190, "median": 45.82844161987305, "min": 45.7642936706543, "max": 45.83320617675781 }, "proc_shmem_gib": { "n": 190, "median": 36.38016128540039, "min": 36.331260681152344, "max": 36.38022232055664 }, "proc_cached_minus_shmem_estimate_gib": { "n": 190, "median": 9.448280334472656, "min": 9.433032989501953, "max": 9.453044891357422 }, "proc_unevictable_gib": { "n": 190, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 190, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 190, "median": 54.609375, "min": 54.6015625, "max": 54.6953125 }, "pswpin_counter_delta": 2, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 141, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 190, "median": 40377588.0, "min": 39828164, "max": 40970064 }, "VmHWM": { "n": 190, "median": 40968536.0, "min": 40659900, "max": 40970064 }, "VmSwap": { "n": 190, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "122130": { "read_bytes": 951492608, "write_bytes": 65536, "rchar": 948414385, "wchar": 64764 } } }, "python": { "VmRSS": { "n": 190, "median": 215824.0, "min": 187856, "max": 218040 }, "VmHWM": { "n": 190, "median": 215824.0, "min": 187856, "max": 218304 }, "VmSwap": { "n": 190, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "122032": { "read_bytes": 196608, "write_bytes": 0, "rchar": 1350138, "wchar": 1504145 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 976564224, "read_mib_per_s": { "n": 189, "median": 0.042968682796909206, "min": 0.035156202398461796, "max": 139.04787141700552 }, "write_bytes": 21590016, "write_mib_per_s": { "n": 189, "median": 0.003906293816897997, "min": 0, "max": 3.5820859412896735 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "official-standard-32768", "scope": "docker_container", "samples": 120, "start_utc": "2026-10-12T03:03:05.104313+00:00", "end_utc": "2026-10-12T03:05:04.104089+00:00", "sample_interval_s": { "n": 119, "median": 0.9999998870007403, "min": 0.9996774999999616, "max": 1.0003395959993213 }, "collector_duration_s": { "n": 120, "median": 0.040719021500081, "min": 0.03654859400012356, "max": 0.052755033000721596 }, "gpu_utilization_pct": { "n": 120, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 120, "median": 15528.0, "min": 15528.0, "max": 15528.0 }, "gpu_power_w": { "n": 120, "median": 99.425, "min": 71.04, "max": 142.45 }, "gpu_temperature_c": { "n": 120, "median": 45.0, "min": 40.0, "max": 54.0 }, "cpu_utilization_pct": { "n": 120, "median": 13.058767734253598, "min": 10.832497492477433, "max": 66.03015075376885 }, "proc_memory_available_gib": { "n": 120, "median": 14.131758213043213, "min": 13.554661750793457, "max": 14.269352912902832 }, "proc_memory_total_minus_available_gib": { "n": 120, "median": 37.86824178695679, "min": 37.73064708709717, "max": 38.44533824920654 }, "proc_page_cache_cached_gib": { "n": 120, "median": 11.98300552368164, "min": 11.955692291259766, "max": 11.984756469726562 }, "proc_shmem_gib": { "n": 120, "median": 0.9617195129394531, "min": 0.9421920776367188, "max": 0.9617195129394531 }, "proc_cached_minus_shmem_estimate_gib": { "n": 120, "median": 11.021286010742188, "min": 11.013500213623047, "max": 11.02303695678711 }, "proc_unevictable_gib": { "n": 120, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 120, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 120, "median": 53.7890625, "min": 53.7890625, "max": 53.7890625 }, "pswpin_counter_delta": 3, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 125, "container_memory_current_gib": { "n": 120, "median": 41.49087905883789, "min": 41.349021911621094, "max": 42.065467834472656 }, "container_memory_lifetime_peak_gib": { "n": 120, "median": 43.719032287597656, "min": 43.719032287597656, "max": 43.719032287597656 }, "container_swap_mib": { "n": 120, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 120, "median": 36.6562442779541, "min": 36.53395462036133, "max": 37.229034423828125 }, "container_memory_stat_file_gib": { "n": 120, "median": 4.74505615234375, "min": 4.725444793701172, "max": 4.745338439941406 }, "container_memory_stat_shmem_gib": { "n": 120, "median": 0.9614295959472656, "min": 0.9420013427734375, "max": 0.9614295959472656 }, "container_memory_stat_file_mapped_gib": { "n": 120, "median": 0.7921676635742188, "min": 0.7725715637207031, "max": 0.7922592163085938 }, "container_memory_stat_unevictable_gib": { "n": 120, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 840085504, "read_mib_per_s": { "n": 119, "median": 0.039064141358025155, "min": 0.03515651423631642, "max": 123.01448190127198 }, "write_bytes": 45056, "write_mib_per_s": { "n": 119, "median": 0.0, "min": 0.0, "max": 0.011719336564509254 } }, "252:3": { "read_bytes": 840085504, "read_mib_per_s": { "n": 119, "median": 0.039064141358025155, "min": 0.03515651423631642, "max": 123.01448190127198 }, "write_bytes": 45056, "write_mib_per_s": { "n": 119, "median": 0.0, "min": 0.0, "max": 0.011719336564509254 } }, "252:6": { "read_bytes": 840085504, "read_mib_per_s": { "n": 119, "median": 0.039064141358025155, "min": 0.03515651423631642, "max": 123.01448190127198 }, "write_bytes": 45056, "write_mib_per_s": { "n": 119, "median": 0.0, "min": 0.0, "max": 0.011719336564509254 } }, "259:0": { "read_bytes": 840085504, "read_mib_per_s": { "n": 119, "median": 0.039064141358025155, "min": 0.03515651423631642, "max": 123.01448190127198 }, "write_bytes": 45056, "write_mib_per_s": { "n": 119, "median": 0.0, "min": 0.0, "max": 0.011719336564509254 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 120, "median": 39353024.0, "min": 39205092, "max": 39952492 }, "VmHWM": { "n": 120, "median": 39432820.0, "min": 39432820, "max": 39952492 }, "VmSwap": { "n": 120, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65309": { "read_bytes": 840085504, "write_bytes": 49152, "rchar": 840626633, "wchar": 47388 } } }, "python": { "VmRSS": { "n": 120, "median": 216108.0, "min": 213996, "max": 216400 }, "VmHWM": { "n": 120, "median": 216108.0, "min": 213996, "max": 216400 }, "VmSwap": { "n": 120, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65213": { "read_bytes": 0, "write_bytes": 0, "rchar": 804403, "wchar": 949697 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "official-standard-32768", "scope": "host", "samples": 120, "start_utc": "2026-10-12T03:03:05.220046+00:00", "end_utc": "2026-10-12T03:05:04.220028+00:00", "sample_interval_s": { "n": 119, "median": 0.9999999569990905, "min": 0.9997207319993322, "max": 1.000289450001219 }, "collector_duration_s": { "n": 120, "median": 0.05561223750009958, "min": 0.047040208999533206, "max": 0.0654813559995091 }, "gpu_utilization_pct": { "n": 120, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 120, "median": 15528.0, "min": 15528.0, "max": 15528.0 }, "gpu_power_w": { "n": 120, "median": 99.425, "min": 71.04, "max": 142.45 }, "gpu_temperature_c": { "n": 120, "median": 45.0, "min": 40.0, "max": 54.0 }, "cpu_utilization_pct": { "n": 120, "median": 11.54906937394247, "min": 9.341117597998332, "max": 55.57418273260687 }, "proc_memory_available_gib": { "n": 120, "median": 17.02586555480957, "min": 16.519641876220703, "max": 17.215320587158203 }, "proc_memory_total_minus_available_gib": { "n": 120, "median": 41.72686958312988, "min": 41.53741455078125, "max": 42.23309326171875 }, "proc_page_cache_cached_gib": { "n": 120, "median": 12.26546859741211, "min": 12.234893798828125, "max": 12.267810821533203 }, "proc_shmem_gib": { "n": 120, "median": 1.0278358459472656, "min": 1.0083084106445312, "max": 1.0278434753417969 }, "proc_cached_minus_shmem_estimate_gib": { "n": 120, "median": 11.237632751464844, "min": 11.226585388183594, "max": 11.239974975585938 }, "proc_unevictable_gib": { "n": 120, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 120, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 120, "median": 54.6953125, "min": 54.6953125, "max": 54.6953125 }, "pswpin_counter_delta": 3, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 125, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 120, "median": 39351428.0, "min": 39206168, "max": 39952492 }, "VmHWM": { "n": 120, "median": 39432820.0, "min": 39432820, "max": 39952492 }, "VmSwap": { "n": 120, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "117886": { "read_bytes": 839725056, "write_bytes": 49152, "rchar": 840265616, "wchar": 46500 } } }, "python": { "VmRSS": { "n": 120, "median": 216108.0, "min": 213964, "max": 216400 }, "VmHWM": { "n": 120, "median": 216108.0, "min": 213964, "max": 216400 }, "VmSwap": { "n": 120, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "117788": { "read_bytes": 0, "write_bytes": 0, "rchar": 798203, "wchar": 949446 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 858103808, "read_mib_per_s": { "n": 119, "median": 0.09765676562777628, "min": 0.03905119663108753, "max": 138.08441344061725 }, "write_bytes": 16179200, "write_mib_per_s": { "n": 119, "median": 0.015620478652435012, "min": 0, "max": 3.6484239023302107 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "needles-128k-ctx131072", "scope": "docker_container", "samples": 192, "start_utc": "2026-10-12T03:49:57.747681+00:00", "end_utc": "2026-10-12T03:53:08.747515+00:00", "sample_interval_s": { "n": 191, "median": 0.9999999690007826, "min": 0.9996014549997199, "max": 1.0004181780004728 }, "collector_duration_s": { "n": 192, "median": 0.039560185000482306, "min": 0.035437983000520035, "max": 0.04630331800035492 }, "gpu_utilization_pct": { "n": 192, "median": 100.0, "min": 50.0, "max": 100.0 }, "vram_used_mib": { "n": 192, "median": 15522.0, "min": 15518.0, "max": 15522.0 }, "gpu_power_w": { "n": 192, "median": 134.595, "min": 59.36, "max": 147.51 }, "gpu_temperature_c": { "n": 192, "median": 48.0, "min": 42.0, "max": 52.0 }, "cpu_utilization_pct": { "n": 192, "median": 11.411411411411411, "min": 10.753768844221106, "max": 36.21621621621622 }, "proc_memory_available_gib": { "n": 192, "median": 11.814711093902588, "min": 11.48316764831543, "max": 12.267498970031738 }, "proc_memory_total_minus_available_gib": { "n": 192, "median": 40.18528890609741, "min": 39.73250102996826, "max": 40.51683235168457 }, "proc_page_cache_cached_gib": { "n": 192, "median": 46.23712730407715, "min": 46.23503112792969, "max": 46.239219665527344 }, "proc_shmem_gib": { "n": 192, "median": 37.278900146484375, "min": 37.278900146484375, "max": 37.27890396118164 }, "proc_cached_minus_shmem_estimate_gib": { "n": 192, "median": 8.958227157592773, "min": 8.956130981445312, "max": 8.960319519042969 }, "proc_unevictable_gib": { "n": 192, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 192, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 192, "median": 53.53125, "min": 53.5234375, "max": 53.53125 }, "pswpin_counter_delta": 2, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 8, "container_memory_current_gib": { "n": 192, "median": 42.76172637939453, "min": 42.30430221557617, "max": 42.998870849609375 }, "container_memory_lifetime_peak_gib": { "n": 192, "median": 44.598018646240234, "min": 44.598018646240234, "max": 44.598018646240234 }, "container_swap_mib": { "n": 192, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 192, "median": 2.6507797241210938, "min": 2.193897247314453, "max": 2.8871917724609375 }, "container_memory_stat_file_gib": { "n": 192, "median": 39.94003677368164, "min": 39.94003677368164, "max": 39.94003677368164 }, "container_memory_stat_shmem_gib": { "n": 192, "median": 37.278621673583984, "min": 37.278621673583984, "max": 37.278621673583984 }, "container_memory_stat_file_mapped_gib": { "n": 192, "median": 37.03480911254883, "min": 37.03480911254883, "max": 37.03480911254883 }, "container_memory_stat_unevictable_gib": { "n": 192, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 4788334592, "read_mib_per_s": { "n": 191, "median": 0.03906269218840263, "min": 0.03514155457497433, "max": 291.5459299815857 }, "write_bytes": 8192, "write_mib_per_s": { "n": 191, "median": 0.0, "min": 0.0, "max": 0.0039062503906280134 } }, "252:3": { "read_bytes": 4788334592, "read_mib_per_s": { "n": 191, "median": 0.03906269218840263, "min": 0.03514155457497433, "max": 291.5459299815857 }, "write_bytes": 8192, "write_mib_per_s": { "n": 191, "median": 0.0, "min": 0.0, "max": 0.0039062503906280134 } }, "252:6": { "read_bytes": 4788334592, "read_mib_per_s": { "n": 191, "median": 0.03906269218840263, "min": 0.03514155457497433, "max": 291.5459299815857 }, "write_bytes": 8192, "write_mib_per_s": { "n": 191, "median": 0.0, "min": 0.0, "max": 0.0039062503906280134 } }, "259:0": { "read_bytes": 4788334592, "read_mib_per_s": { "n": 191, "median": 0.03906269218840263, "min": 0.03514155457497433, "max": 291.5459299815857 }, "write_bytes": 8192, "write_mib_per_s": { "n": 191, "median": 0.0, "min": 0.0, "max": 0.0039062503906280134 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 192, "median": 41769576.0, "min": 41290532, "max": 42017360 }, "VmHWM": { "n": 192, "median": 42114156.0, "min": 41748624, "max": 42130752 }, "VmSwap": { "n": 192, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "106486": { "read_bytes": 4788334592, "write_bytes": 8192, "rchar": 4789499785, "wchar": 3532 } } }, "python": { "VmRSS": { "n": 192, "median": 229872.0, "min": 229684, "max": 229876 }, "VmHWM": { "n": 192, "median": 232288.0, "min": 232116, "max": 232308 }, "VmSwap": { "n": 192, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "106392": { "read_bytes": 0, "write_bytes": 0, "rchar": 1188021, "wchar": 1081868 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "needles-128k-ctx131072", "scope": "host", "samples": 192, "start_utc": "2026-10-12T03:49:57.822097+00:00", "end_utc": "2026-10-12T03:53:08.822086+00:00", "sample_interval_s": { "n": 191, "median": 0.9999985720005498, "min": 0.9995836460002465, "max": 1.0004176889997325 }, "collector_duration_s": { "n": 192, "median": 0.04827098899932025, "min": 0.04311611499906576, "max": 0.059688454000934144 }, "gpu_utilization_pct": { "n": 192, "median": 100.0, "min": 50.0, "max": 100.0 }, "vram_used_mib": { "n": 192, "median": 15522.0, "min": 15518.0, "max": 15522.0 }, "gpu_power_w": { "n": 192, "median": 134.595, "min": 59.36, "max": 147.51 }, "gpu_temperature_c": { "n": 192, "median": 48.0, "min": 42.0, "max": 52.0 }, "cpu_utilization_pct": { "n": 192, "median": 9.50792326939116, "min": 9.181969949916528, "max": 27.669039145907472 }, "proc_memory_available_gib": { "n": 192, "median": 14.930574417114258, "min": 14.547428131103516, "max": 15.331840515136719 }, "proc_memory_total_minus_available_gib": { "n": 192, "median": 43.822160720825195, "min": 43.420894622802734, "max": 44.20530700683594 }, "proc_page_cache_cached_gib": { "n": 192, "median": 46.59866905212402, "min": 46.59526062011719, "max": 46.60169982910156 }, "proc_shmem_gib": { "n": 192, "median": 37.34501647949219, "min": 37.344818115234375, "max": 37.34502410888672 }, "proc_cached_minus_shmem_estimate_gib": { "n": 192, "median": 9.253652572631836, "min": 9.250442504882812, "max": 9.256683349609375 }, "proc_unevictable_gib": { "n": 192, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 192, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 192, "median": 54.4375, "min": 54.4296875, "max": 54.4375 }, "pswpin_counter_delta": 2, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 8, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 192, "median": 41769576.0, "min": 41290532, "max": 42017360 }, "VmHWM": { "n": 192, "median": 42114156.0, "min": 41748624, "max": 42130752 }, "VmSwap": { "n": 192, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "173203": { "read_bytes": 4788334592, "write_bytes": 8192, "rchar": 4789499785, "wchar": 3532 } } }, "python": { "VmRSS": { "n": 192, "median": 229872.0, "min": 229684, "max": 229876 }, "VmHWM": { "n": 192, "median": 232288.0, "min": 232116, "max": 232308 }, "VmSwap": { "n": 192, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "173107": { "read_bytes": 0, "write_bytes": 0, "rchar": 1188021, "wchar": 1081868 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 4803346432, "read_mib_per_s": { "n": 191, "median": 0.03906507130209897, "min": 0.03515390393389727, "max": 304.9674518318032 }, "write_bytes": 19005440, "write_mib_per_s": { "n": 191, "median": 0.0039062978638678145, "min": 0, "max": 2.1991916317677527 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "needles-16k-ctx32768", "scope": "docker_container", "samples": 25, "start_utc": "2026-10-12T03:05:18.104405+00:00", "end_utc": "2026-10-12T03:05:42.104075+00:00", "sample_interval_s": { "n": 24, "median": 1.0000005865003914, "min": 0.9999611960010952, "max": 1.000026566998713 }, "collector_duration_s": { "n": 25, "median": 0.04091764999975567, "min": 0.03830356400067103, "max": 0.04462049100038712 }, "gpu_utilization_pct": { "n": 25, "median": 100.0, "min": 2.0, "max": 100.0 }, "vram_used_mib": { "n": 25, "median": 15528.0, "min": 15528.0, "max": 15528.0 }, "gpu_power_w": { "n": 25, "median": 132.1, "min": 90.21, "max": 144.86 }, "gpu_temperature_c": { "n": 25, "median": 46.0, "min": 40.0, "max": 49.0 }, "cpu_utilization_pct": { "n": 25, "median": 11.365902293120637, "min": 10.943775100401606, "max": 25.0 }, "proc_memory_available_gib": { "n": 25, "median": 14.20644760131836, "min": 14.097275733947754, "max": 14.20866584777832 }, "proc_memory_total_minus_available_gib": { "n": 25, "median": 37.79355239868164, "min": 37.79133415222168, "max": 37.902724266052246 }, "proc_page_cache_cached_gib": { "n": 25, "median": 11.985877990722656, "min": 11.985572814941406, "max": 11.986167907714844 }, "proc_shmem_gib": { "n": 25, "median": 0.9617080688476562, "min": 0.9617080688476562, "max": 0.9617080688476562 }, "proc_cached_minus_shmem_estimate_gib": { "n": 25, "median": 11.024169921875, "min": 11.02386474609375, "max": 11.024459838867188 }, "proc_unevictable_gib": { "n": 25, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 25, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 25, "median": 53.7890625, "min": 53.7890625, "max": 53.7890625 }, "pswpin_counter_delta": 8, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 8, "container_memory_current_gib": { "n": 25, "median": 41.420047760009766, "min": 41.4163818359375, "max": 41.52983474731445 }, "container_memory_lifetime_peak_gib": { "n": 25, "median": 43.719032287597656, "min": 43.719032287597656, "max": 43.719032287597656 }, "container_swap_mib": { "n": 25, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 25, "median": 36.58539962768555, "min": 36.58130645751953, "max": 36.69499206542969 }, "container_memory_stat_file_gib": { "n": 25, "median": 4.7453460693359375, "min": 4.745342254638672, "max": 4.7453460693359375 }, "container_memory_stat_shmem_gib": { "n": 25, "median": 0.9614295959472656, "min": 0.9614295959472656, "max": 0.9614295959472656 }, "container_memory_stat_file_mapped_gib": { "n": 25, "median": 0.7922630310058594, "min": 0.7922630310058594, "max": 0.7922630310058594 }, "container_memory_stat_unevictable_gib": { "n": 25, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 295145472, "read_mib_per_s": { "n": 24, "median": 0.03906256033225034, "min": 0.03906220429914977, "max": 248.89819984411773 }, "write_bytes": 4096, "write_mib_per_s": { "n": 24, "median": 0.0, "min": 0.0, "max": 0.003906264898489532 } }, "252:3": { "read_bytes": 295145472, "read_mib_per_s": { "n": 24, "median": 0.03906256033225034, "min": 0.03906220429914977, "max": 248.90210602908365 }, "write_bytes": 4096, "write_mib_per_s": { "n": 24, "median": 0.0, "min": 0.0, "max": 0.003906264898489532 } }, "252:6": { "read_bytes": 295145472, "read_mib_per_s": { "n": 24, "median": 0.03906256033225034, "min": 0.03906220429914977, "max": 248.89819984411773 }, "write_bytes": 4096, "write_mib_per_s": { "n": 24, "median": 0.0, "min": 0.0, "max": 0.003906264898489532 } }, "259:0": { "read_bytes": 295145472, "read_mib_per_s": { "n": 24, "median": 0.03906256033225034, "min": 0.03906220429914977, "max": 248.90210602908365 }, "write_bytes": 4096, "write_mib_per_s": { "n": 24, "median": 0.0, "min": 0.0, "max": 0.003906264898489532 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 25, "median": 39277076, "min": 39273324, "max": 39391992 }, "VmHWM": { "n": 25, "median": 39951464, "min": 39951464, "max": 39951464 }, "VmSwap": { "n": 25, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65309": { "read_bytes": 295145472, "write_bytes": 8192, "rchar": 295356161, "wchar": 3170 } } }, "python": { "VmRSS": { "n": 25, "median": 216948, "min": 216408, "max": 216948 }, "VmHWM": { "n": 25, "median": 216948, "min": 216408, "max": 216948 }, "VmSwap": { "n": 25, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65213": { "read_bytes": 0, "write_bytes": 0, "rchar": 154465, "wchar": 199648 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "needles-16k-ctx32768", "scope": "host", "samples": 25, "start_utc": "2026-10-12T03:05:18.220029+00:00", "end_utc": "2026-10-12T03:05:42.220028+00:00", "sample_interval_s": { "n": 24, "median": 1.0000007255002856, "min": 0.9997737419998884, "max": 1.000231342000916 }, "collector_duration_s": { "n": 25, "median": 0.048429839000164066, "min": 0.044945252000616165, "max": 0.05951922300118895 }, "gpu_utilization_pct": { "n": 25, "median": 100.0, "min": 2.0, "max": 100.0 }, "vram_used_mib": { "n": 25, "median": 15528.0, "min": 15528.0, "max": 15528.0 }, "gpu_power_w": { "n": 25, "median": 132.1, "min": 90.21, "max": 144.86 }, "gpu_temperature_c": { "n": 25, "median": 46.0, "min": 40.0, "max": 49.0 }, "cpu_utilization_pct": { "n": 25, "median": 9.492089925062448, "min": 9.257714762301918, "max": 20.18581081081081 }, "proc_memory_available_gib": { "n": 25, "median": 17.104835510253906, "min": 17.035335540771484, "max": 17.1759033203125 }, "proc_memory_total_minus_available_gib": { "n": 25, "median": 41.64789962768555, "min": 41.57683181762695, "max": 41.71739959716797 }, "proc_page_cache_cached_gib": { "n": 25, "median": 12.26889419555664, "min": 12.268447875976562, "max": 12.269279479980469 }, "proc_shmem_gib": { "n": 25, "median": 1.0278244018554688, "min": 1.0278244018554688, "max": 1.0278244018554688 }, "proc_cached_minus_shmem_estimate_gib": { "n": 25, "median": 11.241069793701172, "min": 11.240623474121094, "max": 11.241455078125 }, "proc_unevictable_gib": { "n": 25, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 25, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 25, "median": 54.6953125, "min": 54.6953125, "max": 54.6953125 }, "pswpin_counter_delta": 8, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 8, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 25, "median": 39277076, "min": 39273324, "max": 39391992 }, "VmHWM": { "n": 25, "median": 39951464, "min": 39951464, "max": 39951464 }, "VmSwap": { "n": 25, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "117886": { "read_bytes": 295145472, "write_bytes": 8192, "rchar": 295356161, "wchar": 3170 } } }, "python": { "VmRSS": { "n": 25, "median": 216948, "min": 216408, "max": 216948 }, "VmHWM": { "n": 25, "median": 216948, "min": 216408, "max": 216948 }, "VmSwap": { "n": 25, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "117788": { "read_bytes": 0, "write_bytes": 0, "rchar": 154465, "wchar": 199648 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 296767488, "read_mib_per_s": { "n": 24, "median": 0.03906259712912422, "min": 0.03906189325161618, "max": 208.42735269665374 }, "write_bytes": 7413760, "write_mib_per_s": { "n": 24, "median": 0.009765679840853794, "min": 0, "max": 5.949280081131761 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "needles-32k-ctx32768", "scope": "docker_container", "samples": 0 }, { "quant": "IQ3_XXS", "phase": "needles-32k-ctx32768", "scope": "host", "samples": 0 }, { "quant": "IQ3_XXS", "phase": "needles-32k-ctx65536", "scope": "docker_container", "samples": 52, "start_utc": "2026-10-12T03:10:22.783909+00:00", "end_utc": "2026-10-12T03:11:13.784141+00:00", "sample_interval_s": { "n": 51, "median": 1.000000097999873, "min": 0.999978028001351, "max": 1.0000260779997916 }, "collector_duration_s": { "n": 52, "median": 0.03979446450011892, "min": 0.033805303000917775, "max": 0.044797747001211974 }, "gpu_utilization_pct": { "n": 52, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 52, "median": 15518.0, "min": 15518.0, "max": 15518.0 }, "gpu_power_w": { "n": 52, "median": 133.63, "min": 76.28, "max": 145.59 }, "gpu_temperature_c": { "n": 52, "median": 47.0, "min": 40.0, "max": 50.0 }, "cpu_utilization_pct": { "n": 52, "median": 11.40593405273379, "min": 10.9, "max": 44.07630522088353 }, "proc_memory_available_gib": { "n": 52, "median": 13.12401294708252, "min": 12.903347969055176, "max": 13.234234809875488 }, "proc_memory_total_minus_available_gib": { "n": 52, "median": 38.87598705291748, "min": 38.76576519012451, "max": 39.096652030944824 }, "proc_page_cache_cached_gib": { "n": 52, "median": 45.548484802246094, "min": 45.5474739074707, "max": 45.54904556274414 }, "proc_shmem_gib": { "n": 52, "median": 36.31403350830078, "min": 36.31400680541992, "max": 36.31403350830078 }, "proc_cached_minus_shmem_estimate_gib": { "n": 52, "median": 9.234451293945312, "min": 9.233467102050781, "max": 9.23501205444336 }, "proc_unevictable_gib": { "n": 52, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 52, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 52, "median": 53.6953125, "min": 53.6953125, "max": 53.6953125 }, "pswpin_counter_delta": 4, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 11, "container_memory_current_gib": { "n": 52, "median": 39.72865867614746, "min": 39.615848541259766, "max": 39.9486083984375 }, "container_memory_lifetime_peak_gib": { "n": 52, "median": 42.38410568237305, "min": 42.38410568237305, "max": 42.38410568237305 }, "container_swap_mib": { "n": 52, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 52, "median": 2.313495635986328, "min": 2.200145721435547, "max": 2.533031463623047 }, "container_memory_stat_file_gib": { "n": 52, "median": 37.247615814208984, "min": 37.247314453125, "max": 37.247615814208984 }, "container_memory_stat_shmem_gib": { "n": 52, "median": 36.31375503540039, "min": 36.31375503540039, "max": 36.31375503540039 }, "container_memory_stat_file_mapped_gib": { "n": 52, "median": 36.09002685546875, "min": 36.089866638183594, "max": 36.09002685546875 }, "container_memory_stat_unevictable_gib": { "n": 52, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 1175400448, "read_mib_per_s": { "n": 51, "median": 0.03906270625111051, "min": 0.03906148135469737, "max": 352.3262513199578 }, "write_bytes": 4096, "write_mib_per_s": { "n": 51, "median": 0.0, "min": 0.0, "max": 0.003906272535284193 } }, "252:3": { "read_bytes": 1175400448, "read_mib_per_s": { "n": 51, "median": 0.03906270625111051, "min": 0.03906148135469737, "max": 352.3262513199578 }, "write_bytes": 4096, "write_mib_per_s": { "n": 51, "median": 0.0, "min": 0.0, "max": 0.003906272535284193 } }, "252:6": { "read_bytes": 1175400448, "read_mib_per_s": { "n": 51, "median": 0.03906270625111051, "min": 0.03906148135469737, "max": 352.3262513199578 }, "write_bytes": 4096, "write_mib_per_s": { "n": 51, "median": 0.0, "min": 0.0, "max": 0.003906272535284193 } }, "259:0": { "read_bytes": 1175400448, "read_mib_per_s": { "n": 51, "median": 0.03906270625111051, "min": 0.03906148135469737, "max": 352.3262513199578 }, "write_bytes": 4096, "write_mib_per_s": { "n": 51, "median": 0.0, "min": 0.0, "max": 0.003906272535284193 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 52, "median": 40411868.0, "min": 40293012, "max": 40642440 }, "VmHWM": { "n": 52, "median": 40968636.0, "min": 40968636, "max": 40968636 }, "VmSwap": { "n": 52, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "68203": { "read_bytes": 1175400448, "write_bytes": 4096, "rchar": 1175522582, "wchar": 2561 } } }, "python": { "VmRSS": { "n": 52, "median": 220112.0, "min": 218004, "max": 220148 }, "VmHWM": { "n": 52, "median": 220112.0, "min": 218304, "max": 220148 }, "VmSwap": { "n": 52, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "68107": { "read_bytes": 0, "write_bytes": 0, "rchar": 319077, "wchar": 423038 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "needles-32k-ctx65536", "scope": "host", "samples": 52, "start_utc": "2026-10-12T03:10:22.220018+00:00", "end_utc": "2026-10-12T03:11:13.220025+00:00", "sample_interval_s": { "n": 51, "median": 0.9999998879993655, "min": 0.9997398689993133, "max": 1.0002603949997138 }, "collector_duration_s": { "n": 52, "median": 0.04959709499962628, "min": 0.045224825998957385, "max": 0.05983588399976725 }, "gpu_utilization_pct": { "n": 52, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 52, "median": 15518.0, "min": 15518.0, "max": 15518.0 }, "gpu_power_w": { "n": 52, "median": 132.03, "min": 54.21, "max": 143.92 }, "gpu_temperature_c": { "n": 52, "median": 47.0, "min": 41.0, "max": 50.0 }, "cpu_utilization_pct": { "n": 52, "median": 9.587343046070401, "min": 9.265442404006677, "max": 30.195712954333644 }, "proc_memory_available_gib": { "n": 52, "median": 16.214309692382812, "min": 16.003684997558594, "max": 16.331504821777344 }, "proc_memory_total_minus_available_gib": { "n": 52, "median": 42.53842544555664, "min": 42.42123031616211, "max": 42.74905014038086 }, "proc_page_cache_cached_gib": { "n": 52, "median": 45.834503173828125, "min": 45.8332405090332, "max": 45.8353157043457 }, "proc_shmem_gib": { "n": 52, "median": 36.380149841308594, "min": 36.380149841308594, "max": 36.38016128540039 }, "proc_cached_minus_shmem_estimate_gib": { "n": 52, "median": 9.454353332519531, "min": 9.453079223632812, "max": 9.45516586303711 }, "proc_unevictable_gib": { "n": 52, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 52, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 52, "median": 54.6015625, "min": 54.6015625, "max": 54.6015625 }, "pswpin_counter_delta": 4, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 11, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 52, "median": 40411868.0, "min": 40293004, "max": 40642072 }, "VmHWM": { "n": 52, "median": 40968636.0, "min": 40968636, "max": 40968636 }, "VmSwap": { "n": 52, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "122130": { "read_bytes": 1175564288, "write_bytes": 4096, "rchar": 1175686873, "wchar": 3702 } } }, "python": { "VmRSS": { "n": 52, "median": 220112.0, "min": 218040, "max": 220148 }, "VmHWM": { "n": 52, "median": 220112.0, "min": 218304, "max": 220148 }, "VmSwap": { "n": 52, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "122032": { "read_bytes": 0, "write_bytes": 0, "rchar": 355376, "wchar": 423295 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 1179656192, "read_mib_per_s": { "n": 51, "median": 0.039062569882926786, "min": 0.03905233096828869, "max": 467.5823127345523 }, "write_bytes": 9633792, "write_mib_per_s": { "n": 51, "median": 0.0039062466171871243, "min": 0, "max": 6.000000251991889 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "needles-64k-ctx131072", "scope": "docker_container", "samples": 93, "start_utc": "2026-10-12T03:48:24.747357+00:00", "end_utc": "2026-10-12T03:49:56.747527+00:00", "sample_interval_s": { "n": 92, "median": 0.9999999689989636, "min": 0.9996023639996565, "max": 1.0004114729999856 }, "collector_duration_s": { "n": 93, "median": 0.04001153399985924, "min": 0.03568214799997804, "max": 0.045773014000587864 }, "gpu_utilization_pct": { "n": 93, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 93, "median": 15518.0, "min": 15432.0, "max": 15518.0 }, "gpu_power_w": { "n": 93, "median": 134.4, "min": 39.09, "max": 146.6 }, "gpu_temperature_c": { "n": 93, "median": 48.0, "min": 39.0, "max": 53.0 }, "cpu_utilization_pct": { "n": 93, "median": 11.377245508982035, "min": 1.402805611222445, "max": 34.72850678733032 }, "proc_memory_available_gib": { "n": 93, "median": 12.141942024230957, "min": 11.84128189086914, "max": 12.653338432312012 }, "proc_memory_total_minus_available_gib": { "n": 93, "median": 39.85805797576904, "min": 39.34666156768799, "max": 40.15871810913086 }, "proc_page_cache_cached_gib": { "n": 93, "median": 46.233802795410156, "min": 46.231178283691406, "max": 46.23485565185547 }, "proc_shmem_gib": { "n": 93, "median": 37.278900146484375, "min": 37.278812408447266, "max": 37.278900146484375 }, "proc_cached_minus_shmem_estimate_gib": { "n": 93, "median": 8.954902648925781, "min": 8.95236587524414, "max": 8.955955505371094 }, "proc_unevictable_gib": { "n": 93, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 93, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 93, "median": 53.53125, "min": 53.53125, "max": 53.53125 }, "pswpin_counter_delta": 0, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 14, "container_memory_current_gib": { "n": 93, "median": 42.43195343017578, "min": 41.940345764160156, "max": 42.733551025390625 }, "container_memory_lifetime_peak_gib": { "n": 93, "median": 44.598018646240234, "min": 44.598018646240234, "max": 44.598018646240234 }, "container_swap_mib": { "n": 93, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 93, "median": 2.3208580017089844, "min": 1.8308601379394531, "max": 2.622303009033203 }, "container_memory_stat_file_gib": { "n": 93, "median": 39.940032958984375, "min": 39.940025329589844, "max": 39.940032958984375 }, "container_memory_stat_shmem_gib": { "n": 93, "median": 37.278621673583984, "min": 37.278621673583984, "max": 37.278621673583984 }, "container_memory_stat_file_mapped_gib": { "n": 93, "median": 37.03480911254883, "min": 37.0348014831543, "max": 37.03480911254883 }, "container_memory_stat_unevictable_gib": { "n": 93, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 2393210880, "read_mib_per_s": { "n": 92, "median": 0.039062685352453635, "min": 0.03515625108988022, "max": 423.3920220524419 }, "write_bytes": 8192, "write_mib_per_s": { "n": 92, "median": 0.0, "min": 0.0, "max": 0.0039062814963468 } }, "252:3": { "read_bytes": 2393210880, "read_mib_per_s": { "n": 92, "median": 0.039062685352453635, "min": 0.03515625108988022, "max": 423.3920220524419 }, "write_bytes": 8192, "write_mib_per_s": { "n": 92, "median": 0.0, "min": 0.0, "max": 0.0039062814963468 } }, "252:6": { "read_bytes": 2393210880, "read_mib_per_s": { "n": 92, "median": 0.039062685352453635, "min": 0.03515625108988022, "max": 423.3920220524419 }, "write_bytes": 8192, "write_mib_per_s": { "n": 92, "median": 0.0, "min": 0.0, "max": 0.0039062814963468 } }, "259:0": { "read_bytes": 2393210880, "read_mib_per_s": { "n": 92, "median": 0.039062685352453635, "min": 0.03515625108988022, "max": 423.3920220524419 }, "write_bytes": 8192, "write_mib_per_s": { "n": 92, "median": 0.0, "min": 0.0, "max": 0.0039062814963468 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 93, "median": 41432528, "min": 40903488, "max": 41747916 }, "VmHWM": { "n": 93, "median": 41733456, "min": 41308296, "max": 41747916 }, "VmSwap": { "n": 93, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "106486": { "read_bytes": 2393210880, "write_bytes": 8192, "rchar": 2394127504, "wchar": 3253 } } }, "python": { "VmRSS": { "n": 93, "median": 220780, "min": 213864, "max": 221516 }, "VmHWM": { "n": 93, "median": 220780, "min": 213864, "max": 221516 }, "VmSwap": { "n": 93, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "106392": { "read_bytes": 0, "write_bytes": 0, "rchar": 578172, "wchar": 834613 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "needles-64k-ctx131072", "scope": "host", "samples": 93, "start_utc": "2026-10-12T03:48:24.822068+00:00", "end_utc": "2026-10-12T03:49:56.822159+00:00", "sample_interval_s": { "n": 92, "median": 0.999999759500497, "min": 0.9997616710006696, "max": 1.00033255299968 }, "collector_duration_s": { "n": 93, "median": 0.049144388000058825, "min": 0.04349535300025309, "max": 0.058755724001457565 }, "gpu_utilization_pct": { "n": 93, "median": 100.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 93, "median": 15518.0, "min": 15432.0, "max": 15518.0 }, "gpu_power_w": { "n": 93, "median": 134.4, "min": 39.09, "max": 146.6 }, "gpu_temperature_c": { "n": 93, "median": 48.0, "min": 39.0, "max": 53.0 }, "cpu_utilization_pct": { "n": 93, "median": 9.50792326939116, "min": 1.5037593984962405, "max": 27.106227106227106 }, "proc_memory_available_gib": { "n": 93, "median": 15.1815185546875, "min": 14.950599670410156, "max": 15.691104888916016 }, "proc_memory_total_minus_available_gib": { "n": 93, "median": 43.57121658325195, "min": 43.06163024902344, "max": 43.8021354675293 }, "proc_page_cache_cached_gib": { "n": 93, "median": 46.593780517578125, "min": 46.59080123901367, "max": 46.59529113769531 }, "proc_shmem_gib": { "n": 93, "median": 37.34501647949219, "min": 37.34492874145508, "max": 37.34502029418945 }, "proc_cached_minus_shmem_estimate_gib": { "n": 93, "median": 9.248764038085938, "min": 9.245872497558594, "max": 9.250274658203125 }, "proc_unevictable_gib": { "n": 93, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 93, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 93, "median": 54.4375, "min": 54.4375, "max": 54.4375 }, "pswpin_counter_delta": 0, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 14, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 93, "median": 41432528, "min": 40903488, "max": 41747916 }, "VmHWM": { "n": 93, "median": 41733456, "min": 41308296, "max": 41747916 }, "VmSwap": { "n": 93, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "173203": { "read_bytes": 2393214976, "write_bytes": 8192, "rchar": 2394131600, "wchar": 3253 } } }, "python": { "VmRSS": { "n": 93, "median": 220780, "min": 213896, "max": 221516 }, "VmHWM": { "n": 93, "median": 220780, "min": 213896, "max": 221516 }, "VmSwap": { "n": 93, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "173107": { "read_bytes": 0, "write_bytes": 0, "rchar": 578172, "wchar": 834613 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 2402181120, "read_mib_per_s": { "n": 92, "median": 0.03906472889678562, "min": 0.035149426512106945, "max": 379.571290342034 }, "write_bytes": 14012416, "write_mib_per_s": { "n": 92, "median": 0.0039063382456300715, "min": 0, "max": 3.5351445062487525 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "needles-8k-ctx32768", "scope": "docker_container", "samples": 13, "start_utc": "2026-10-12T03:05:05.104226+00:00", "end_utc": "2026-10-12T03:05:17.104292+00:00", "sample_interval_s": { "n": 12, "median": 0.9999995035004758, "min": 0.9999698560004617, "max": 1.0000207709999813 }, "collector_duration_s": { "n": 13, "median": 0.0392236520001461, "min": 0.036512417000267305, "max": 0.04315564100033953 }, "gpu_utilization_pct": { "n": 13, "median": 100.0, "min": 45.0, "max": 100.0 }, "vram_used_mib": { "n": 13, "median": 15528.0, "min": 15528.0, "max": 15528.0 }, "gpu_power_w": { "n": 13, "median": 125.75, "min": 74.78, "max": 137.58 }, "gpu_temperature_c": { "n": 13, "median": 45.0, "min": 40.0, "max": 47.0 }, "cpu_utilization_pct": { "n": 13, "median": 11.4, "min": 11.033099297893681, "max": 34.51511991657977 }, "proc_memory_available_gib": { "n": 13, "median": 14.207257270812988, "min": 13.555563926696777, "max": 14.208550453186035 }, "proc_memory_total_minus_available_gib": { "n": 13, "median": 37.79274272918701, "min": 37.791449546813965, "max": 38.44443607330322 }, "proc_page_cache_cached_gib": { "n": 13, "median": 11.985416412353516, "min": 11.984790802001953, "max": 11.985549926757812 }, "proc_shmem_gib": { "n": 13, "median": 0.9617080688476562, "min": 0.9617080688476562, "max": 0.9617195129394531 }, "proc_cached_minus_shmem_estimate_gib": { "n": 13, "median": 11.02370834350586, "min": 11.0230712890625, "max": 11.023841857910156 }, "proc_unevictable_gib": { "n": 13, "median": 0.0, "min": 0.0, "max": 0.0 }, "proc_mlocked_gib": { "n": 13, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 13, "median": 53.7890625, "min": 53.7890625, "max": 53.7890625 }, "pswpin_counter_delta": 5, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 7, "container_memory_current_gib": { "n": 13, "median": 41.41611862182617, "min": 41.41572952270508, "max": 42.065433502197266 }, "container_memory_lifetime_peak_gib": { "n": 13, "median": 43.719032287597656, "min": 43.719032287597656, "max": 43.719032287597656 }, "container_swap_mib": { "n": 13, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_anon_gib": { "n": 13, "median": 36.58130645751953, "min": 36.58127212524414, "max": 37.22920608520508 }, "container_memory_stat_file_gib": { "n": 13, "median": 4.745338439941406, "min": 4.745338439941406, "max": 4.745342254638672 }, "container_memory_stat_shmem_gib": { "n": 13, "median": 0.9614295959472656, "min": 0.9614295959472656, "max": 0.9614295959472656 }, "container_memory_stat_file_mapped_gib": { "n": 13, "median": 0.7922592163085938, "min": 0.7922592163085938, "max": 0.7922630310058594 }, "container_memory_stat_unevictable_gib": { "n": 13, "median": 0.0, "min": 0.0, "max": 0.0 }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_memory_event_deltas": { "low": 0, "high": 0, "max": 0, "oom": 0, "oom_kill": 0, "oom_group_kill": 0, "sock_throttled": 0 }, "container_io_devices": { "252:4": { "read_bytes": 335917056, "read_mib_per_s": { "n": 12, "median": 0.03906267216873958, "min": 0.035156204824290924, "max": 171.73564554922015 }, "write_bytes": 12288, "write_mib_per_s": { "n": 12, "median": 0.0, "min": 0.0, "max": 0.011719103260643278 } }, "252:3": { "read_bytes": 335917056, "read_mib_per_s": { "n": 12, "median": 0.03906267216873958, "min": 0.035156204824290924, "max": 171.73564554922015 }, "write_bytes": 12288, "write_mib_per_s": { "n": 12, "median": 0.0, "min": 0.0, "max": 0.011719103260643278 } }, "252:6": { "read_bytes": 335917056, "read_mib_per_s": { "n": 12, "median": 0.03906267216873958, "min": 0.035156204824290924, "max": 171.73564554922015 }, "write_bytes": 12288, "write_mib_per_s": { "n": 12, "median": 0.0, "min": 0.0, "max": 0.011719103260643278 } }, "259:0": { "read_bytes": 335917056, "read_mib_per_s": { "n": 12, "median": 0.03906267216873958, "min": 0.035156204824290924, "max": 171.73564554922015 }, "write_bytes": 12288, "write_mib_per_s": { "n": 12, "median": 0.0, "min": 0.0, "max": 0.011719103260643278 } } }, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 13, "median": 39273320, "min": 39273316, "max": 39952700 }, "VmHWM": { "n": 13, "median": 39951464, "min": 39951464, "max": 39952700 }, "VmSwap": { "n": 13, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65309": { "read_bytes": 335917056, "write_bytes": 4096, "rchar": 336012351, "wchar": 3060 } } }, "python": { "VmRSS": { "n": 13, "median": 216408, "min": 216376, "max": 216408 }, "VmHWM": { "n": 13, "median": 216408, "min": 216376, "max": 216408 }, "VmSwap": { "n": 13, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "65213": { "read_bytes": 0, "write_bytes": 0, "rchar": 103647, "wchar": 93505 } } } }, "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, { "quant": "IQ3_XXS", "phase": "needles-8k-ctx32768", "scope": "host", "samples": 13, "start_utc": "2026-10-12T03:05:05.220069+00:00", "end_utc": "2026-10-12T03:05:17.220031+00:00", "sample_interval_s": { "n": 12, "median": 1.0000003764998837, "min": 0.9999850109998079, "max": 1.0000078499997471 }, "collector_duration_s": { "n": 13, "median": 0.04855150299954403, "min": 0.04691218900006788, "max": 0.053959446000590106 }, "gpu_utilization_pct": { "n": 13, "median": 100.0, "min": 45.0, "max": 100.0 }, "vram_used_mib": { "n": 13, "median": 15528.0, "min": 15528.0, "max": 15528.0 }, "gpu_power_w": { "n": 13, "median": 125.75, "min": 74.78, "max": 137.58 }, "gpu_temperature_c": { "n": 13, "median": 45.0, "min": 40.0, "max": 47.0 }, "cpu_utilization_pct": { "n": 13, "median": 9.5, "min": 9.341117597998332, "max": 26.70753064798599 }, "proc_memory_available_gib": { "n": 13, "median": 16.977256774902344, "min": 16.538272857666016, "max": 17.045543670654297 }, "proc_memory_total_minus_available_gib": { "n": 13, "median": 41.77547836303711, "min": 41.707191467285156, "max": 42.21446228027344 }, "proc_page_cache_cached_gib": { "n": 13, "median": 12.26822280883789, "min": 12.267841339111328, "max": 12.268417358398438 }, "proc_shmem_gib": { "n": 13, "median": 1.0278244018554688, "min": 1.0278244018554688, "max": 1.0278358459472656 }, "proc_cached_minus_shmem_estimate_gib": { "n": 13, "median": 11.240398406982422, "min": 11.240005493164062, "max": 11.240592956542969 }, "proc_unevictable_gib": { "n": 13, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "proc_mlocked_gib": { "n": 13, "median": 0.16986465454101562, "min": 0.16986465454101562, "max": 0.16986465454101562 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 13, "median": 54.6953125, "min": 54.6953125, "max": 54.6953125 }, "pswpin_counter_delta": 5, "pswpout_counter_delta": 0, "pgmajfault_counter_delta": 7, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 13, "median": 39273320, "min": 39273316, "max": 39952700 }, "VmHWM": { "n": 13, "median": 39951464, "min": 39951464, "max": 39952700 }, "VmSwap": { "n": 13, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "117886": { "read_bytes": 335122432, "write_bytes": 4096, "rchar": 335217727, "wchar": 3012 } } }, "python": { "VmRSS": { "n": 13, "median": 216408, "min": 216400, "max": 216408 }, "VmHWM": { "n": 13, "median": 216408, "min": 216400, "max": 216408 }, "VmSwap": { "n": 13, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "117788": { "read_bytes": 0, "write_bytes": 0, "rchar": 103599, "wchar": 93505 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 336445440, "read_mib_per_s": { "n": 12, "median": 0.039062504394574524, "min": 0.03906219336179199, "max": 236.4176061635442 }, "write_bytes": 6852608, "write_mib_per_s": { "n": 12, "median": 0.003906250439457453, "min": 0, "max": 6.371189247756859 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." } ], "whole_host_capture_summary": { "scope": "host", "samples": 6915, "start_utc": "2026-10-12T01:58:40.697161+00:00", "end_utc": "2026-10-12T03:53:55.822063+00:00", "sample_interval_s": { "n": 6914, "median": 0.9999999470001057, "min": 0.9995504540002003, "max": 1.601938630999939 }, "collector_duration_s": { "n": 6915, "median": 0.05377806300020893, "min": 0.041707828000653535, "max": 1.6017648659999395 }, "gpu_utilization_pct": { "n": 6915, "median": 91.0, "min": 0.0, "max": 100.0 }, "vram_used_mib": { "n": 6915, "median": 15494.0, "min": 2.0, "max": 15528.0 }, "gpu_power_w": { "n": 6915, "median": 98.38, "min": 9.72, "max": 155.87 }, "gpu_temperature_c": { "n": 6915, "median": 42.0, "min": 33.0, "max": 56.0 }, "cpu_utilization_pct": { "n": 6914, "median": 30.43872091576459, "min": 0.6672226855713094, "max": 68.58813700918964 }, "proc_memory_available_gib": { "n": 6915, "median": 22.684284210205078, "min": 14.547428131103516, "max": 56.00506591796875 }, "proc_memory_total_minus_available_gib": { "n": 6915, "median": 36.068450927734375, "min": 2.747669219970703, "max": 44.20530700683594 }, "proc_page_cache_cached_gib": { "n": 6915, "median": 46.04438400268555, "min": 9.323848724365234, "max": 52.09363555908203 }, "proc_shmem_gib": { "n": 6915, "median": 29.288589477539062, "min": 0.06619644165039062, "max": 37.34502410888672 }, "proc_cached_minus_shmem_estimate_gib": { "n": 6915, "median": 16.626373291015625, "min": 9.24447250366211, "max": 52.02708053588867 }, "proc_unevictable_gib": { "n": 6915, "median": 0.16986465454101562, "min": 0.1689300537109375, "max": 0.16991043090820312 }, "proc_mlocked_gib": { "n": 6915, "median": 0.16986465454101562, "min": 0.1689300537109375, "max": 0.16991043090820312 }, "page_cache_note": "Raw Linux Cached includes shmem/tmpfs; Cached-minus-Shmem is an approximation, not a complete evictable-cache accounting or model-specific cache measurement.", "proc_swap_used_mib": { "n": 6915, "median": 54.6953125, "min": 0.375, "max": 75.9765625 }, "pswpin_counter_delta": 21659, "pswpout_counter_delta": 22729, "pgmajfault_counter_delta": 141854, "container_memory_current_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_lifetime_peak_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_swap_mib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_anon_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_shmem_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_file_mapped_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_unevictable_gib": { "n": 0, "median": null, "min": null, "max": null }, "container_memory_stat_note": "memory.stat file includes shmem; these categories overlap and must not be added together as independent totals.", "container_memory_events_last": null, "container_memory_event_deltas": null, "container_io_devices": {}, "container_io_note": "Cgroup charged block I/O per device number; logical and physical layers must not be summed. Includes all processes in the inference container.", "processes": { "strata": { "VmRSS": { "n": 6344, "median": 33886090.0, "min": 102104, "max": 42017360 }, "VmHWM": { "n": 6344, "median": 34287084.0, "min": 102104, "max": 42130752 }, "VmSwap": { "n": 6344, "median": 0.0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "55503": { "read_bytes": 45740769280, "write_bytes": 77824, "rchar": 3928768600, "wchar": 258511787 }, "65735": { "read_bytes": 50092437504, "write_bytes": 569344, "rchar": 8133884858, "wchar": 267278079 }, "114103": { "read_bytes": 46893277184, "write_bytes": 32768, "rchar": 9615907281, "wchar": 260447759 }, "117886": { "read_bytes": 53223936000, "write_bytes": 81920, "rchar": 4758348494, "wchar": 260656209 }, "122130": { "read_bytes": 53699129344, "write_bytes": 651264, "rchar": 8444539112, "wchar": 255140250 }, "173203": { "read_bytes": 56915202048, "write_bytes": 36864, "rchar": 9617105833, "wchar": 259094257 } } }, "python": { "VmRSS": { "n": 6353, "median": 224244, "min": 35556, "max": 231236 }, "VmHWM": { "n": 6353, "median": 224244, "min": 35556, "max": 233796 }, "VmSwap": { "n": 6353, "median": 0, "min": 0, "max": 0 }, "io_deltas_by_pid": { "55404": { "read_bytes": 1040384, "write_bytes": 0, "rchar": 4828417, "wchar": 1251554 }, "65561": { "read_bytes": 741376, "write_bytes": 0, "rchar": 14598741, "wchar": 5857072 }, "114005": { "read_bytes": 0, "write_bytes": 4096, "rchar": 12298146, "wchar": 2457730 }, "117788": { "read_bytes": 12238848, "write_bytes": 4096, "rchar": 13420242, "wchar": 1364637 }, "122032": { "read_bytes": 331776, "write_bytes": 4096, "rchar": 16227866, "wchar": 6189459 }, "173107": { "read_bytes": 2719744, "write_bytes": 4096, "rchar": 15569277, "wchar": 2457736 } } } }, "whole_physical_nvme": { "nvme0n1": { "read_bytes": 630837379072, "read_mib_per_s": { "n": 6914, "median": 1.3730506667631623, "min": 0, "max": 2335.398574105028 }, "write_bytes": 148308545536, "write_mib_per_s": { "n": 6914, "median": 0.015624666186835248, "min": 0, "max": 1637.9312805236616 } } }, "disk_attribution": "Entire physical-device traffic from diskstats; includes all host activity. Not model-specific traffic.", "memory_note": "Host procfs and Docker cgroup measurements have separate scopes. Cgroup memory includes charged cache and kernel memory; lifetime peak may precede the selected window. LXC procfs is virtualized." }, "long_context_rankings": { "successful_warmed_dossier_speed": [ "strata-flash-next-iq2_xs", "strata-flash-next-iq3_xxs" ], "full_marker_recall_qualified_practical_coding": [ "strata-flash-next-iq2_xs", "strata-flash-next-iq3_xxs" ], "scope": "Separate dossier evaluation; unchanged practical/coding index; no pooled ranking with report4. All three nonempty stop responses required for speed eligibility; full-marker qualification additionally requires 14/14 markers each." }, "report_url": "https://report5.firetussin.com/", "conclusions": [ { "classification": "verified_measurement", "text": "Highest original v2.1 balanced score: IQ2_XS. This is limited to the tested dataset and configuration." } ], "historical_reference": { "url": "https://report4.firetussin.com/", "model_family": "Qwen3.8-27B", "source": "Original verified local canonical scores, recalculated from raw results; live endpoint returned HTTP 403. Recalculated reference accompanies report5; original report4 publication remains unchanged.", "benchmark_data_sha256": "34c731084d5265c67be57f7b58d242d6f2a7e5a540758a78655b3964bdf39bef", "models": [ { "id": "qwen38-gsq-rco-iq3s-mtp", "label": "GSQ-RCO \u00b7 IQ3_S \u00b7 MTP", "name": "Qwen3.8-27B GSQ-RCO IQ3_S MTP", "provider": "GSQ-RCO", "quant": "IQ3_S", "repo_id": "ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF", "revision": "d562806dbafae37109975e970aae91b43e73b440", "filename": "Qwen3.8-27B-GSQ-RCO-IQ3_S-mtp.gguf", "sha256": "58FD826723939933DC86F45B7FE04545CBC2DE1C70F6FE2CDD3858C87A98C12F", "source_url": "https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF/tree/d562806dbafae37109975e970aae91b43e73b440", "file_bytes": 12120016960, "file_gib": 11.288, "mtp": true, "selected_adaptive": { "mtp": 3, "ubatch": 256 }, "practical_profile": { "ctx": 73728, "ubatch": 256, "mtp": 2 }, "balanced_score": 97.7, "category_scores": { "business": 100.0, "research": 100.0, "knowledge": 100.0, "agent_tools": 100.0, "practical_sysadmin": 95.0, "executive_content": 89.285, "humaneval_plus": 90.24 }, "role_scores": { "CFO / Analyst": 99.46, "Researcher / Scout": 98.93, "Hermes Coordinator": 97.65, "Librarian / Knowledge": 98.39, "Forge / Coder": 94.12 }, "applied_weighted_score": 98.53, "humaneval_plus_passed": 148, "humaneval_base_passed": 152, "humaneval_evaluated": 164, "humaneval_plus_percent": 90.24, "humaneval_base_percent": 92.68, "adaptive_seconds": 31.5127, "fixed_seconds": 33.9472, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 84.3482, "adaptive_repetitions_seconds": [ 84.3482, 31.489, 31.5127 ], "adaptive_generated_tokens": [ 2271, 2271, 2271 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 73.77, "adaptive_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 875 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 875 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 875 } ], "fixed_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 861 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 776 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 861 } ], "free_vram_mib": 1628.0, "vram_gib": 14.331, "peak_temperature_c": 63.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 3, "practical_repeat_identical_tasks": 24, "run_id": "20261009-141105-local-5070ti-v2.1-full", "scorer_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a", "scorer_rebuilt": false, "scorer_recovery": null, "run_completed_at": "2026-10-09T16:38:39.681447+00:00", "rank": 1, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 6, "validated_heuristic_mean": 98.6, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5863, "reasoning_characters": 1877, "wall_time_s": 85.4623, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 861 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 861 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "fd7c9527138870c1fa4c9266dbbbd7564ac8a9e0afe066fb14f2454ef22823cc" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5441, "reasoning_characters": 1877, "wall_time_s": 31.916, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 776 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 776 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "4fe72cb87a889b854ffc70987ee343921123d67c89a16242eef63a7ef32e549a" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5863, "reasoning_characters": 1877, "wall_time_s": 33.9472, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 861 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 861 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "fd7c9527138870c1fa4c9266dbbbd7564ac8a9e0afe066fb14f2454ef22823cc" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 6003, "reasoning_characters": 1877, "wall_time_s": 84.3482, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 875 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 875 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "2881e353a9123a10d21ebc36fcb8c280b415edf4313b7f44ce37b9ce22ed7104" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 6003, "reasoning_characters": 1877, "wall_time_s": 31.489, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 875 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 875 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "2881e353a9123a10d21ebc36fcb8c280b415edf4313b7f44ce37b9ce22ed7104" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 6003, "reasoning_characters": 1877, "wall_time_s": 31.5127, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 875 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 875 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "2881e353a9123a10d21ebc36fcb8c280b415edf4313b7f44ce37b9ce22ed7104" } ] }, "long_context_recall_qualified": true }, { "id": "qwen38-hauhaucs-uncensored-q2-k-p", "label": "HauhauCS Aggressive \u00b7 Q2_K_P \u00b7 MTP", "name": "Qwen3.8-27B HauhauCS Uncensored Q2_K_P MTP", "provider": "HauhauCS Aggressive", "quant": "Q2_K_P", "repo_id": "HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF", "revision": "993a5971fda8f30dd1b7eb2654792ba4415c7460", "filename": "Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-Q2_K_P.gguf", "sha256": "F778A15FA0393B38E473B954B39A793543FC424838E7AF8C0F6BD27B6CC11CA7", "source_url": "https://huggingface.co/HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF/tree/993a5971fda8f30dd1b7eb2654792ba4415c7460", "file_bytes": 10676422784, "file_gib": 9.943, "mtp": true, "selected_adaptive": { "mtp": 3, "ubatch": 512 }, "practical_profile": { "ctx": 73728, "ubatch": 256, "mtp": 2 }, "balanced_score": 97.61, "category_scores": { "business": 100.0, "research": 100.0, "knowledge": 100.0, "agent_tools": 100.0, "practical_sysadmin": 100.0, "executive_content": 90.0775, "humaneval_plus": 85.98 }, "role_scores": { "CFO / Analyst": 99.5, "Researcher / Scout": 99.01, "Hermes Coordinator": 97.81, "Librarian / Knowledge": 98.51, "Forge / Coder": 92.99 }, "applied_weighted_score": 98.9, "humaneval_plus_passed": 141, "humaneval_base_passed": 144, "humaneval_evaluated": 164, "humaneval_plus_percent": 85.98, "humaneval_base_percent": 87.8, "adaptive_seconds": 6.8971, "fixed_seconds": 7.7443, "successful_long_context_output": false, "speed_rank_eligible": false, "long_context_empty_finals": 6, "adaptive_first_request_seconds": 56.98, "adaptive_repetitions_seconds": [ 56.98, 6.8971, 6.8881 ], "adaptive_generated_tokens": [ 515, 515, 515 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 83.37, "adaptive_quality": [ { "overall_hermes": 32.9, "retrieval_facts": "0/14", "word_count": 0 }, { "overall_hermes": 32.9, "retrieval_facts": "0/14", "word_count": 0 }, { "overall_hermes": 32.9, "retrieval_facts": "0/14", "word_count": 0 } ], "fixed_quality": [ { "overall_hermes": 32.9, "retrieval_facts": "0/14", "word_count": 0 }, { "overall_hermes": 32.9, "retrieval_facts": "0/14", "word_count": 0 }, { "overall_hermes": 32.9, "retrieval_facts": "0/14", "word_count": 0 } ], "free_vram_mib": 3026.0, "vram_gib": 12.966, "peak_temperature_c": 64.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 0, 0, 0 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 0, "practical_repeat_identical_tasks": 24, "run_id": "20261010-021726-v2.1-qwen38-hauhaucs-uncensored-q2-k-p", "scorer_image_id": "sha256:a1ce362575e724db75b4bf9a744fc51a0b16e2c4a267aeca076217b0971272f5", "scorer_rebuilt": true, "scorer_recovery": { "replacement_image_id": "sha256:a1ce362575e724db75b4bf9a744fc51a0b16e2c4a267aeca076217b0971272f5", "dockerfile_sha256": "a2a2f40af7d5046ed496855fb3bb0a07dbd80e7f3a1da355454c0ff211eaa4a9", "reason": "Original pinned scorer image disappeared from host; rebuilt retained identical Dockerfile. Same pinned Python base and EvalPlus 0.3.1; transitive dependencies were not fully locked in original build.", "isolation_probe_passed": true, "recovered_at": "2026-10-10T00:31:47.125849+00:00", "original_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a" }, "run_completed_at": "2026-10-10T02:59:59.458159+00:00", "rank": 2, "long_context_recalculation": { "requests": 6, "empty_finals": 6, "valid_completed_finals": 0, "full_marker_recall_completions": 0, "validated_heuristic_mean": 0, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": true, "final_characters": 0, "reasoning_characters": 1788, "wall_time_s": 58.0078, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 32.9, "retrieval_facts": "0/14", "word_count": 0 }, "validated_quality": { "overall_hermes": 0, "retrieval_facts": "0/14", "word_count": 0 }, "valid_completed_final": false, "literal_marker_hits": 0, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": true, "final_characters": 0, "reasoning_characters": 1788, "wall_time_s": 7.7351, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 32.9, "retrieval_facts": "0/14", "word_count": 0 }, "validated_quality": { "overall_hermes": 0, "retrieval_facts": "0/14", "word_count": 0 }, "valid_completed_final": false, "literal_marker_hits": 0, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": true, "final_characters": 0, "reasoning_characters": 1788, "wall_time_s": 7.7443, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 32.9, "retrieval_facts": "0/14", "word_count": 0 }, "validated_quality": { "overall_hermes": 0, "retrieval_facts": "0/14", "word_count": 0 }, "valid_completed_final": false, "literal_marker_hits": 0, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": true, "final_characters": 0, "reasoning_characters": 1788, "wall_time_s": 56.98, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 32.9, "retrieval_facts": "0/14", "word_count": 0 }, "validated_quality": { "overall_hermes": 0, "retrieval_facts": "0/14", "word_count": 0 }, "valid_completed_final": false, "literal_marker_hits": 0, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": true, "final_characters": 0, "reasoning_characters": 1788, "wall_time_s": 6.8971, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 32.9, "retrieval_facts": "0/14", "word_count": 0 }, "validated_quality": { "overall_hermes": 0, "retrieval_facts": "0/14", "word_count": 0 }, "valid_completed_final": false, "literal_marker_hits": 0, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": true, "final_characters": 0, "reasoning_characters": 1788, "wall_time_s": 6.8881, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 32.9, "retrieval_facts": "0/14", "word_count": 0 }, "validated_quality": { "overall_hermes": 0, "retrieval_facts": "0/14", "word_count": 0 }, "valid_completed_final": false, "literal_marker_hits": 0, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855" } ] }, "long_context_recall_qualified": false }, { "id": "qwen38-pi-iq2m-control", "label": "Pi control \u00b7 IQ2_M", "name": "Qwen3.8-27B Pi IQ2_M control", "provider": "Pi control", "quant": "IQ2_M", "repo_id": "firetussin/Qwen3.8-27B-pi-RCO-allocation-transfer-GGUF", "revision": "8e4712137774900181d9e7b833eb82a039445bc4", "filename": "Qwen3.8-27B-pi-llama-IQ2_M-base-imatrix.gguf", "sha256": "11DA0AC94BE157B2911A97DAAC130E36692164AABA0C9DAA0D144A9FA7460EC6", "source_url": "https://huggingface.co/firetussin/Qwen3.8-27B-pi-RCO-allocation-transfer-GGUF/tree/8e4712137774900181d9e7b833eb82a039445bc4", "file_bytes": 10004593632, "file_gib": 9.318, "mtp": false, "selected_adaptive": { "mtp": 0, "ubatch": 512 }, "practical_profile": { "ctx": 73728, "ubatch": 1024, "mtp": 0 }, "balanced_score": 97.58, "category_scores": { "business": 100.0, "research": 100.0, "knowledge": 100.0, "agent_tools": 100.0, "practical_sysadmin": 100.0, "executive_content": 92.855, "humaneval_plus": 82.93 }, "role_scores": { "CFO / Analyst": 99.64, "Researcher / Scout": 99.29, "Hermes Coordinator": 98.07, "Librarian / Knowledge": 98.93, "Forge / Coder": 91.47 }, "applied_weighted_score": 99.21, "humaneval_plus_passed": 136, "humaneval_base_passed": 142, "humaneval_evaluated": 164, "humaneval_plus_percent": 82.93, "humaneval_base_percent": 86.59, "adaptive_seconds": 39.5328, "fixed_seconds": 39.5408, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 93.7959, "adaptive_repetitions_seconds": [ 93.7959, 39.5123, 39.5328 ], "adaptive_generated_tokens": [ 1620, 1620, 1620 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 41.55, "adaptive_quality": [ { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 } ], "fixed_quality": [ { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 } ], "free_vram_mib": 4726.0, "vram_gib": 11.306, "peak_temperature_c": 64.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 0, "practical_repeat_identical_tasks": 24, "run_id": "20261009-205329-v2.1-qwen38-pi-iq2m-control", "scorer_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a", "scorer_rebuilt": false, "scorer_recovery": null, "run_completed_at": "2026-10-09T21:51:24.521769+00:00", "rank": 3, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 6, "validated_heuristic_mean": 97.1, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3546, "reasoning_characters": 1752, "wall_time_s": 93.8659, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "f7ba67206334032578fb041feba7e4b00b9555d3a77f4518de31865ff52f03cf" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3546, "reasoning_characters": 1752, "wall_time_s": 39.5141, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "f7ba67206334032578fb041feba7e4b00b9555d3a77f4518de31865ff52f03cf" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3546, "reasoning_characters": 1752, "wall_time_s": 39.5408, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "f7ba67206334032578fb041feba7e4b00b9555d3a77f4518de31865ff52f03cf" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3546, "reasoning_characters": 1752, "wall_time_s": 93.7959, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "f7ba67206334032578fb041feba7e4b00b9555d3a77f4518de31865ff52f03cf" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3546, "reasoning_characters": 1752, "wall_time_s": 39.5123, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "f7ba67206334032578fb041feba7e4b00b9555d3a77f4518de31865ff52f03cf" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3546, "reasoning_characters": 1752, "wall_time_s": 39.5328, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 474 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "f7ba67206334032578fb041feba7e4b00b9555d3a77f4518de31865ff52f03cf" } ] }, "long_context_recall_qualified": true }, { "id": "qwen38-pi-rco-transfer-iq3xxs", "label": "Pi transfer \u00b7 IQ3_XXS", "name": "Qwen3.8-27B Pi RCO allocation-transfer IQ3_XXS", "provider": "Pi transfer", "quant": "IQ3_XXS", "repo_id": "firetussin/Qwen3.8-27B-pi-RCO-allocation-transfer-GGUF", "revision": "8e4712137774900181d9e7b833eb82a039445bc4", "filename": "Qwen3.8-27B-pi-RCO-allocation-transfer-IQ3_XXS.gguf", "sha256": "35134E868FC680F30B9764125B8B07FD359C935DEEFCCB64DB9025FB2936883C", "source_url": "https://huggingface.co/firetussin/Qwen3.8-27B-pi-RCO-allocation-transfer-GGUF/tree/8e4712137774900181d9e7b833eb82a039445bc4", "file_bytes": 10094357472, "file_gib": 9.401, "mtp": false, "selected_adaptive": { "mtp": 0, "ubatch": 512 }, "practical_profile": { "ctx": 73728, "ubatch": 512, "mtp": 0 }, "balanced_score": 97.58, "category_scores": { "business": 100.0, "research": 97.5, "knowledge": 100.0, "agent_tools": 100.0, "practical_sysadmin": 100.0, "executive_content": 93.65, "humaneval_plus": 87.2 }, "role_scores": { "CFO / Analyst": 99.18, "Researcher / Scout": 98.36, "Hermes Coordinator": 97.91, "Librarian / Knowledge": 98.42, "Forge / Coder": 93.47 }, "applied_weighted_score": 98.74, "humaneval_plus_passed": 143, "humaneval_base_passed": 149, "humaneval_evaluated": 164, "humaneval_plus_percent": 87.2, "humaneval_base_percent": 90.85, "adaptive_seconds": 47.0664, "fixed_seconds": 47.0577, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 97.6601, "adaptive_repetitions_seconds": [ 97.6601, 47.056, 47.0664 ], "adaptive_generated_tokens": [ 1935, 1935, 1935 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 41.58, "adaptive_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 } ], "fixed_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 } ], "free_vram_mib": 4284.0, "vram_gib": 11.737, "peak_temperature_c": 65.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 0, "practical_repeat_identical_tasks": 24, "run_id": "20261009-141105-local-5070ti-v2.1-full", "scorer_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a", "scorer_rebuilt": false, "scorer_recovery": null, "run_completed_at": "2026-10-09T16:38:39.681447+00:00", "rank": 3, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 6, "validated_heuristic_mean": 98.6, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4960, "reasoning_characters": 1899, "wall_time_s": 97.7204, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "465c28eed4aeea70c731bef834cbda76a37a911d53df798da2810814e616736e" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4960, "reasoning_characters": 1899, "wall_time_s": 47.0404, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "465c28eed4aeea70c731bef834cbda76a37a911d53df798da2810814e616736e" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4960, "reasoning_characters": 1899, "wall_time_s": 47.0577, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "465c28eed4aeea70c731bef834cbda76a37a911d53df798da2810814e616736e" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4960, "reasoning_characters": 1899, "wall_time_s": 97.6601, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "465c28eed4aeea70c731bef834cbda76a37a911d53df798da2810814e616736e" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4960, "reasoning_characters": 1899, "wall_time_s": 47.056, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "465c28eed4aeea70c731bef834cbda76a37a911d53df798da2810814e616736e" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4960, "reasoning_characters": 1899, "wall_time_s": 47.0664, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 708 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "465c28eed4aeea70c731bef834cbda76a37a911d53df798da2810814e616736e" } ] }, "long_context_recall_qualified": true }, { "id": "qwen38-gsq-rco-iq3xxs-mtp", "label": "GSQ-RCO \u00b7 IQ3_XXS \u00b7 MTP", "name": "Qwen3.8-27B GSQ-RCO IQ3_XXS MTP", "provider": "GSQ-RCO", "quant": "IQ3_XXS", "repo_id": "ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF", "revision": "d562806dbafae37109975e970aae91b43e73b440", "filename": "Qwen3.8-27B-GSQ-RCO-IQ3_XXS-mtp.gguf", "sha256": "63F29A2189A6B4CC31F81E093D3856AD293A5583114439F41AE1ED7AF4093262", "source_url": "https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF/tree/d562806dbafae37109975e970aae91b43e73b440", "file_bytes": 10442827840, "file_gib": 9.726, "mtp": true, "selected_adaptive": { "mtp": 2, "ubatch": 256 }, "practical_profile": { "ctx": 73728, "ubatch": 1024, "mtp": 3 }, "balanced_score": 97.37, "category_scores": { "business": 100.0, "research": 100.0, "knowledge": 100.0, "agent_tools": 100.0, "practical_sysadmin": 100.0, "executive_content": 86.5075, "humaneval_plus": 87.2 }, "role_scores": { "CFO / Analyst": 99.33, "Researcher / Scout": 98.65, "Hermes Coordinator": 97.34, "Librarian / Knowledge": 97.98, "Forge / Coder": 93.6 }, "applied_weighted_score": 98.5, "humaneval_plus_passed": 143, "humaneval_base_passed": 150, "humaneval_evaluated": 164, "humaneval_plus_percent": 87.2, "humaneval_base_percent": 91.46, "adaptive_seconds": 30.5218, "fixed_seconds": 33.1036, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 85.8776, "adaptive_repetitions_seconds": [ 85.8776, 30.5215, 30.5218 ], "adaptive_generated_tokens": [ 2008, 2008, 2008 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 67.37, "adaptive_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 696 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 696 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 696 } ], "fixed_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 798 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 798 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 798 } ], "free_vram_mib": 3376.0, "vram_gib": 12.624, "peak_temperature_c": 63.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 0, "practical_repeat_identical_tasks": 24, "run_id": "20261009-234120-v2.1-qwen38-gsq-rco-iq3xxs-mtp", "scorer_image_id": "sha256:a1ce362575e724db75b4bf9a744fc51a0b16e2c4a267aeca076217b0971272f5", "scorer_rebuilt": true, "scorer_recovery": { "replacement_image_id": "sha256:a1ce362575e724db75b4bf9a744fc51a0b16e2c4a267aeca076217b0971272f5", "dockerfile_sha256": "a2a2f40af7d5046ed496855fb3bb0a07dbd80e7f3a1da355454c0ff211eaa4a9", "reason": "Original pinned scorer image disappeared from host; rebuilt retained identical Dockerfile. Same pinned Python base and EvalPlus 0.3.1; transitive dependencies were not fully locked in original build.", "isolation_probe_passed": true, "recovered_at": "2026-10-10T00:31:47.125849+00:00", "original_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a" }, "run_completed_at": "2026-10-10T00:34:16.518209+00:00", "rank": 5, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 6, "validated_heuristic_mean": 98.6, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5643, "reasoning_characters": 2106, "wall_time_s": 87.2117, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 798 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 798 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "3d76ef71c6f34cee5f6aa7dcd8cf4792bfacfe3384df4b6ca677ef40ce51144c" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5643, "reasoning_characters": 2106, "wall_time_s": 33.1036, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 798 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 798 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "3d76ef71c6f34cee5f6aa7dcd8cf4792bfacfe3384df4b6ca677ef40ce51144c" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5643, "reasoning_characters": 2106, "wall_time_s": 33.0983, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 798 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 798 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "3d76ef71c6f34cee5f6aa7dcd8cf4792bfacfe3384df4b6ca677ef40ce51144c" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5152, "reasoning_characters": 1948, "wall_time_s": 85.8776, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 696 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 696 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "add2db7b9513495b1234a7de4a57c10740d659375dae90fc58274a4487323bce" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5152, "reasoning_characters": 1948, "wall_time_s": 30.5215, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 696 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 696 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "add2db7b9513495b1234a7de4a57c10740d659375dae90fc58274a4487323bce" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5152, "reasoning_characters": 1948, "wall_time_s": 30.5218, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 696 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 696 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "add2db7b9513495b1234a7de4a57c10740d659375dae90fc58274a4487323bce" } ] }, "long_context_recall_qualified": true }, { "id": "qwen38-gsq-rco-iq3xxs", "label": "GSQ-RCO \u00b7 IQ3_XXS", "name": "Qwen3.8-27B GSQ-RCO IQ3_XXS", "provider": "GSQ-RCO", "quant": "IQ3_XXS", "repo_id": "ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF", "revision": "d562806dbafae37109975e970aae91b43e73b440", "filename": "Qwen3.8-27B-GSQ-RCO-IQ3_XXS.gguf", "sha256": "FDFCB6A29B11188956DFBFD904223588A6C1B77EB250C3E8A36E1BD269DF91F7", "source_url": "https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF/tree/d562806dbafae37109975e970aae91b43e73b440", "file_bytes": 10094357632, "file_gib": 9.401, "mtp": false, "selected_adaptive": { "mtp": 0, "ubatch": 512 }, "practical_profile": { "ctx": 73728, "ubatch": 1024, "mtp": 0 }, "balanced_score": 97.31, "category_scores": { "business": 100.0, "research": 100.0, "knowledge": 100.0, "agent_tools": 100.0, "practical_sysadmin": 100.0, "executive_content": 86.5075, "humaneval_plus": 86.59 }, "role_scores": { "CFO / Analyst": 99.33, "Researcher / Scout": 98.65, "Hermes Coordinator": 97.31, "Librarian / Knowledge": 97.98, "Forge / Coder": 93.3 }, "applied_weighted_score": 98.5, "humaneval_plus_passed": 142, "humaneval_base_passed": 148, "humaneval_evaluated": 164, "humaneval_plus_percent": 86.59, "humaneval_base_percent": 90.24, "adaptive_seconds": 55.2577, "fixed_seconds": 55.3026, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 105.8465, "adaptive_repetitions_seconds": [ 105.8465, 55.2577, 55.2509 ], "adaptive_generated_tokens": [ 2275, 2275, 2275 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 41.57, "adaptive_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 } ], "fixed_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 } ], "free_vram_mib": 4284.0, "vram_gib": 11.737, "peak_temperature_c": 65.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 0, "practical_repeat_identical_tasks": 24, "run_id": "20261009-183243-v2.1-qwen38-gsq-rco-iq3xxs", "scorer_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a", "scorer_rebuilt": false, "scorer_recovery": null, "run_completed_at": "2026-10-09T19:28:02.076048+00:00", "rank": 6, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 6, "validated_heuristic_mean": 98.6, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 6307, "reasoning_characters": 2076, "wall_time_s": 105.9478, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "b1598242a6048b85ebfcd0c2ca018e59ccd49c66bc27400c63de114660224250" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 6307, "reasoning_characters": 2076, "wall_time_s": 55.2699, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "b1598242a6048b85ebfcd0c2ca018e59ccd49c66bc27400c63de114660224250" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 6307, "reasoning_characters": 2076, "wall_time_s": 55.3026, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "b1598242a6048b85ebfcd0c2ca018e59ccd49c66bc27400c63de114660224250" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 6307, "reasoning_characters": 2076, "wall_time_s": 105.8465, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "b1598242a6048b85ebfcd0c2ca018e59ccd49c66bc27400c63de114660224250" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 6307, "reasoning_characters": 2076, "wall_time_s": 55.2577, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "b1598242a6048b85ebfcd0c2ca018e59ccd49c66bc27400c63de114660224250" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 6307, "reasoning_characters": 2076, "wall_time_s": 55.2509, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 822 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "b1598242a6048b85ebfcd0c2ca018e59ccd49c66bc27400c63de114660224250" } ] }, "long_context_recall_qualified": true }, { "id": "qwen38-unsloth-ud-q2-k-xl", "label": "Unsloth \u00b7 UD-Q2_K_XL", "name": "Qwen3.8-27B Unsloth UD-Q2_K_XL", "provider": "Unsloth", "quant": "UD-Q2_K_XL", "repo_id": "unsloth/Qwen3.8-27B-GGUF", "revision": "4ca720788d1e01f1bff70c033e0d0028fd02e502", "filename": "Qwen3.8-27B-UD-Q2_K_XL.gguf", "sha256": "FD4730DD8AAD070517978752B63D530AEB1740D2283CAB9FA24F1E404032DDB0", "source_url": "https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/tree/4ca720788d1e01f1bff70c033e0d0028fd02e502", "file_bytes": 9828981664, "file_gib": 9.154, "mtp": false, "selected_adaptive": { "mtp": 0, "ubatch": 512 }, "practical_profile": { "ctx": 73728, "ubatch": 1024, "mtp": 0 }, "balanced_score": 97.09, "category_scores": { "business": 100.0, "research": 100.0, "knowledge": 100.0, "agent_tools": 97.5, "practical_sysadmin": 100.0, "executive_content": 89.285, "humaneval_plus": 85.37 }, "role_scores": { "CFO / Analyst": 99.21, "Researcher / Scout": 98.55, "Hermes Coordinator": 96.91, "Librarian / Knowledge": 97.89, "Forge / Coder": 92.31 }, "applied_weighted_score": 98.39, "humaneval_plus_passed": 140, "humaneval_base_passed": 144, "humaneval_evaluated": 164, "humaneval_plus_percent": 85.37, "humaneval_base_percent": 87.8, "adaptive_seconds": 42.3193, "fixed_seconds": 42.2704, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 92.595, "adaptive_repetitions_seconds": [ 92.595, 42.3117, 42.3193 ], "adaptive_generated_tokens": [ 1772, 1772, 1772 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 42.41, "adaptive_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 } ], "fixed_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 } ], "free_vram_mib": 4954.0, "vram_gib": 11.083, "peak_temperature_c": 65.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 0, "practical_repeat_identical_tasks": 24, "run_id": "20261009-163846-v2.1-qwen38-unsloth-ud-q2-k-xl", "scorer_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a", "scorer_rebuilt": false, "scorer_recovery": null, "run_completed_at": "2026-10-09T17:32:58.393493+00:00", "rank": 7, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 6, "validated_heuristic_mean": 98.6, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4457, "reasoning_characters": 2142, "wall_time_s": 92.5988, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "81a3c503a7afeda8c7d5abfda9703ddd9df10c2a9ebf93aac390805b809ae15b" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4457, "reasoning_characters": 2142, "wall_time_s": 42.2704, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "81a3c503a7afeda8c7d5abfda9703ddd9df10c2a9ebf93aac390805b809ae15b" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4457, "reasoning_characters": 2142, "wall_time_s": 42.2634, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "81a3c503a7afeda8c7d5abfda9703ddd9df10c2a9ebf93aac390805b809ae15b" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4457, "reasoning_characters": 2142, "wall_time_s": 92.595, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "81a3c503a7afeda8c7d5abfda9703ddd9df10c2a9ebf93aac390805b809ae15b" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4457, "reasoning_characters": 2142, "wall_time_s": 42.3117, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "81a3c503a7afeda8c7d5abfda9703ddd9df10c2a9ebf93aac390805b809ae15b" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4457, "reasoning_characters": 2142, "wall_time_s": 42.3193, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 636 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "81a3c503a7afeda8c7d5abfda9703ddd9df10c2a9ebf93aac390805b809ae15b" } ] }, "long_context_recall_qualified": true }, { "id": "qwen38-orcarouter-uncensored-iq3-xxs", "label": "OrcaRouter \u00b7 IQ3_XXS \u00b7 MTP", "name": "Qwen3.8-27B orcarouter Uncensored IQ3_XXS MTP", "provider": "OrcaRouter", "quant": "IQ3_XXS", "repo_id": "orcarouter/Qwen3.8-27B-Uncensored-GGUF", "revision": "fc437a3374c9977bdc339a8ec106dcd9a0357001", "filename": "Qwen3.8-27B-Uncensored-IQ3_XXS.gguf", "sha256": "0D91AD7B2C57F545A030B1D391375A5F9A822423B90923D91CF744E717E80D03", "source_url": "https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored-GGUF/tree/fc437a3374c9977bdc339a8ec106dcd9a0357001", "file_bytes": 11637692000, "file_gib": 10.838, "mtp": true, "selected_adaptive": { "mtp": 3, "ubatch": 256 }, "practical_profile": { "ctx": 73728, "ubatch": 256, "mtp": 2 }, "balanced_score": 97.0, "category_scores": { "business": 100.0, "research": 100.0, "knowledge": 100.0, "agent_tools": 97.5, "practical_sysadmin": 100.0, "executive_content": 86.5075, "humaneval_plus": 87.2 }, "role_scores": { "CFO / Analyst": 99.08, "Researcher / Scout": 98.28, "Hermes Coordinator": 96.59, "Librarian / Knowledge": 97.48, "Forge / Coder": 93.22 }, "applied_weighted_score": 98.08, "humaneval_plus_passed": 143, "humaneval_base_passed": 149, "humaneval_evaluated": 164, "humaneval_plus_percent": 87.2, "humaneval_base_percent": 90.85, "adaptive_seconds": 26.2816, "fixed_seconds": 32.8604, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 78.7592, "adaptive_repetitions_seconds": [ 78.7592, 26.2816, 26.2473 ], "adaptive_generated_tokens": [ 1847, 1847, 1847 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 72.35, "adaptive_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 692 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 692 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 692 } ], "fixed_quality": [ { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 775 }, { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 775 }, { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 775 } ], "free_vram_mib": 2436.0, "vram_gib": 13.542, "peak_temperature_c": 64.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 0, "practical_repeat_identical_tasks": 24, "run_id": "20261010-034645-v2.1-qwen38-orcarouter-uncensored-iq3-xxs", "scorer_image_id": "sha256:a1ce362575e724db75b4bf9a744fc51a0b16e2c4a267aeca076217b0971272f5", "scorer_rebuilt": true, "scorer_recovery": { "replacement_image_id": "sha256:a1ce362575e724db75b4bf9a744fc51a0b16e2c4a267aeca076217b0971272f5", "dockerfile_sha256": "a2a2f40af7d5046ed496855fb3bb0a07dbd80e7f3a1da355454c0ff211eaa4a9", "reason": "Original pinned scorer image disappeared from host; rebuilt retained identical Dockerfile. Same pinned Python base and EvalPlus 0.3.1; transitive dependencies were not fully locked in original build.", "isolation_probe_passed": true, "recovered_at": "2026-10-10T00:31:47.125849+00:00", "original_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a" }, "run_completed_at": "2026-10-10T04:31:32.813593+00:00", "rank": 8, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 6, "validated_heuristic_mean": 97.85, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5619, "reasoning_characters": 2044, "wall_time_s": 84.1473, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 775 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 775 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "47a9d3a8edac4ce0b68988416af5974ab84a1074c633f0bc51193ea4942d22df" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5619, "reasoning_characters": 2044, "wall_time_s": 32.8604, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 775 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 775 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "47a9d3a8edac4ce0b68988416af5974ab84a1074c633f0bc51193ea4942d22df" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5619, "reasoning_characters": 2044, "wall_time_s": 32.8547, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 775 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 775 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "47a9d3a8edac4ce0b68988416af5974ab84a1074c633f0bc51193ea4942d22df" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4913, "reasoning_characters": 2057, "wall_time_s": 78.7592, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 692 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 692 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "e3902a7e641f55e47a2502ce970bb701d8a564f296b2f5a2dc17777fb10d8934" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4913, "reasoning_characters": 2057, "wall_time_s": 26.2816, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 692 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 692 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "e3902a7e641f55e47a2502ce970bb701d8a564f296b2f5a2dc17777fb10d8934" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4913, "reasoning_characters": 2057, "wall_time_s": 26.2473, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 692 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 692 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "e3902a7e641f55e47a2502ce970bb701d8a564f296b2f5a2dc17777fb10d8934" } ] }, "long_context_recall_qualified": true }, { "id": "qwen38-hauhaucs-uncensored-iq3-xs", "label": "HauhauCS Aggressive \u00b7 IQ3_XS \u00b7 MTP", "name": "Qwen3.8-27B HauhauCS Uncensored IQ3_XS MTP", "provider": "HauhauCS Aggressive", "quant": "IQ3_XS", "repo_id": "HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF", "revision": "993a5971fda8f30dd1b7eb2654792ba4415c7460", "filename": "Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-IQ3_XS.gguf", "sha256": "E22208325E05D8561B51829048E52512DA57C2BF8D54866199C4126C45BD3504", "source_url": "https://huggingface.co/HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF/tree/993a5971fda8f30dd1b7eb2654792ba4415c7460", "file_bytes": 12175558784, "file_gib": 11.339, "mtp": true, "selected_adaptive": { "mtp": 3, "ubatch": 1024 }, "practical_profile": { "ctx": 73728, "ubatch": 256, "mtp": 2 }, "balanced_score": 96.77, "category_scores": { "business": 95.0, "research": 100.0, "knowledge": 100.0, "agent_tools": 100.0, "practical_sysadmin": 95.0, "executive_content": 93.65, "humaneval_plus": 86.59 }, "role_scores": { "CFO / Analyst": 97.18, "Researcher / Scout": 98.86, "Hermes Coordinator": 98.13, "Librarian / Knowledge": 98.8, "Forge / Coder": 92.3 }, "applied_weighted_score": 97.91, "humaneval_plus_passed": 142, "humaneval_base_passed": 148, "humaneval_evaluated": 164, "humaneval_plus_percent": 86.59, "humaneval_base_percent": 90.24, "adaptive_seconds": 28.0724, "fixed_seconds": 34.379, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 77.6575, "adaptive_repetitions_seconds": [ 77.6575, 28.062, 28.0724 ], "adaptive_generated_tokens": [ 1899, 1899, 1899 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 69.44, "adaptive_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 752 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 752 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 752 } ], "fixed_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 809 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 800 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 817 } ], "free_vram_mib": 1434.0, "vram_gib": 14.521, "peak_temperature_c": 62.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 3, "practical_repeat_identical_tasks": 24, "run_id": "20261010-030000-v2.1-qwen38-hauhaucs-uncensored-iq3-xs", "scorer_image_id": "sha256:a1ce362575e724db75b4bf9a744fc51a0b16e2c4a267aeca076217b0971272f5", "scorer_rebuilt": true, "scorer_recovery": { "replacement_image_id": "sha256:a1ce362575e724db75b4bf9a744fc51a0b16e2c4a267aeca076217b0971272f5", "dockerfile_sha256": "a2a2f40af7d5046ed496855fb3bb0a07dbd80e7f3a1da355454c0ff211eaa4a9", "reason": "Original pinned scorer image disappeared from host; rebuilt retained identical Dockerfile. Same pinned Python base and EvalPlus 0.3.1; transitive dependencies were not fully locked in original build.", "isolation_probe_passed": true, "recovered_at": "2026-10-10T00:31:47.125849+00:00", "original_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a" }, "run_completed_at": "2026-10-10T03:46:36.611105+00:00", "rank": 9, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 6, "validated_heuristic_mean": 98.6, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 6061, "reasoning_characters": 1898, "wall_time_s": 85.4581, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 809 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 809 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "3612c414c8cddda031f06f8f756ecdc4e949b4b710d511f57475f9106cb75d8a" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5954, "reasoning_characters": 1898, "wall_time_s": 34.0571, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 800 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 800 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "8691508d76c7d976e843de18e6309f271f2b629847cc9f3df94039615bfe983a" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 6099, "reasoning_characters": 1898, "wall_time_s": 34.379, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 817 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 817 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "7bb48a09a917b62c470b8ea658a7e2ea967f79472057f4bf06e767291eea4018" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5206, "reasoning_characters": 2181, "wall_time_s": 77.6575, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 752 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 752 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "2b8ac11d350d220dafb883ebef134c9402d371540656b9dc8b28b76b02da57cc" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5206, "reasoning_characters": 2181, "wall_time_s": 28.062, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 752 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 752 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "2b8ac11d350d220dafb883ebef134c9402d371540656b9dc8b28b76b02da57cc" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5206, "reasoning_characters": 2181, "wall_time_s": 28.0724, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 752 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 752 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "2b8ac11d350d220dafb883ebef134c9402d371540656b9dc8b28b76b02da57cc" } ] }, "long_context_recall_qualified": true }, { "id": "qwen38-swift-iq2-s", "label": "Swift \u00b7 IQ2_S", "name": "Swift-Qwen3.8-27B IQ2_S", "provider": "Swift", "quant": "IQ2_S", "repo_id": "ukisai/Swift-Qwen3.8-27B-GGUF", "revision": "f8b396d5df72de36889bd572f3117c2bf70e4911", "filename": "Swift-Qwen3.8-27B-IQ2_S.gguf", "sha256": "9E475E0F1D336F10FECA5B73A3EF1DAB08151D9F5C33E8A75BF911C761E8EF65", "source_url": "https://huggingface.co/ukisai/Swift-Qwen3.8-27B-GGUF/tree/f8b396d5df72de36889bd572f3117c2bf70e4911", "file_bytes": 9896380512, "file_gib": 9.217, "mtp": false, "selected_adaptive": { "mtp": 0, "ubatch": 512 }, "practical_profile": { "ctx": 73728, "ubatch": 1024, "mtp": 0 }, "balanced_score": 96.52, "category_scores": { "business": 100.0, "research": 100.0, "knowledge": 98.215, "agent_tools": 100.0, "practical_sysadmin": 100.0, "executive_content": 90.0775, "humaneval_plus": 78.66 }, "role_scores": { "CFO / Analyst": 99.24, "Researcher / Scout": 98.56, "Hermes Coordinator": 97.0, "Librarian / Knowledge": 97.89, "Forge / Coder": 89.15 }, "applied_weighted_score": 98.5, "humaneval_plus_passed": 129, "humaneval_base_passed": 135, "humaneval_evaluated": 164, "humaneval_plus_percent": 78.66, "humaneval_base_percent": 82.32, "adaptive_seconds": 35.8948, "fixed_seconds": 35.9104, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 89.3587, "adaptive_repetitions_seconds": [ 89.3587, 35.8948, 35.8928 ], "adaptive_generated_tokens": [ 1517, 1517, 1517 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 42.9, "adaptive_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 } ], "fixed_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 } ], "free_vram_mib": 5214.0, "vram_gib": 10.829, "peak_temperature_c": 65.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 0, "practical_repeat_identical_tasks": 24, "run_id": "20261009-141105-local-5070ti-v2.1-full", "scorer_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a", "scorer_rebuilt": false, "scorer_recovery": null, "run_completed_at": "2026-10-09T16:38:39.681447+00:00", "rank": 10, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 6, "validated_heuristic_mean": 98.6, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3697, "reasoning_characters": 1919, "wall_time_s": 89.4533, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "9e2f54d088729d58f1f259b938191038acea90bfacb059e5fc1e87b89b87ca33" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3697, "reasoning_characters": 1919, "wall_time_s": 35.9104, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "9e2f54d088729d58f1f259b938191038acea90bfacb059e5fc1e87b89b87ca33" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3697, "reasoning_characters": 1919, "wall_time_s": 35.8954, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "9e2f54d088729d58f1f259b938191038acea90bfacb059e5fc1e87b89b87ca33" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3697, "reasoning_characters": 1919, "wall_time_s": 89.3587, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "9e2f54d088729d58f1f259b938191038acea90bfacb059e5fc1e87b89b87ca33" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3697, "reasoning_characters": 1919, "wall_time_s": 35.8948, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "9e2f54d088729d58f1f259b938191038acea90bfacb059e5fc1e87b89b87ca33" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3697, "reasoning_characters": 1919, "wall_time_s": 35.8928, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 513 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "9e2f54d088729d58f1f259b938191038acea90bfacb059e5fc1e87b89b87ca33" } ] }, "long_context_recall_qualified": true }, { "id": "qwen38-gsq-rco-iq3s", "label": "GSQ-RCO \u00b7 IQ3_S", "name": "Qwen3.8-27B GSQ-RCO IQ3_S", "provider": "GSQ-RCO", "quant": "IQ3_S", "repo_id": "ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF", "revision": "d562806dbafae37109975e970aae91b43e73b440", "filename": "Qwen3.8-27B-GSQ-RCO-IQ3_S.gguf", "sha256": "64B53B64C7AA39F20A7E54BD80582FE595B1D745624EE8A72E92508C0326D810", "source_url": "https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF/tree/d562806dbafae37109975e970aae91b43e73b440", "file_bytes": 11771546784, "file_gib": 10.963, "mtp": false, "selected_adaptive": { "mtp": 0, "ubatch": 512 }, "practical_profile": { "ctx": 73728, "ubatch": 1024, "mtp": 0 }, "balanced_score": 96.44, "category_scores": { "business": 100.0, "research": 100.0, "knowledge": 100.0, "agent_tools": 90.625, "practical_sysadmin": 100.0, "executive_content": 90.0775, "humaneval_plus": 88.41 }, "role_scores": { "CFO / Analyst": 98.57, "Researcher / Scout": 97.6, "Hermes Coordinator": 95.12, "Librarian / Knowledge": 96.64, "Forge / Coder": 92.8 }, "applied_weighted_score": 97.33, "humaneval_plus_passed": 145, "humaneval_base_passed": 152, "humaneval_evaluated": 164, "humaneval_plus_percent": 88.41, "humaneval_base_percent": 92.68, "adaptive_seconds": 53.7452, "fixed_seconds": 53.7526, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 101.9073, "adaptive_repetitions_seconds": [ 101.9073, 53.7415, 53.7452 ], "adaptive_generated_tokens": [ 2081, 2081, 2081 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 39.11, "adaptive_quality": [ { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 } ], "fixed_quality": [ { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 } ], "free_vram_mib": 2684.0, "vram_gib": 13.3, "peak_temperature_c": 65.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 13, 13, 13 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 0, "practical_repeat_identical_tasks": 24, "run_id": "20261009-173259-v2.1-qwen38-gsq-rco-iq3s", "scorer_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a", "scorer_rebuilt": false, "scorer_recovery": null, "run_completed_at": "2026-10-09T18:32:42.022411+00:00", "rank": 11, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 0, "validated_heuristic_mean": 97.6, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5425, "reasoning_characters": 1877, "wall_time_s": 101.9917, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, "validated_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, "valid_completed_final": true, "literal_marker_hits": 13, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "9cc2f488285d73372b8a0154c85d4ae9fd421408729e0e74db19dfd3faeb16ad" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5425, "reasoning_characters": 1877, "wall_time_s": 53.7439, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, "validated_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, "valid_completed_final": true, "literal_marker_hits": 13, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "9cc2f488285d73372b8a0154c85d4ae9fd421408729e0e74db19dfd3faeb16ad" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5425, "reasoning_characters": 1877, "wall_time_s": 53.7526, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, "validated_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, "valid_completed_final": true, "literal_marker_hits": 13, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "9cc2f488285d73372b8a0154c85d4ae9fd421408729e0e74db19dfd3faeb16ad" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5425, "reasoning_characters": 1877, "wall_time_s": 101.9073, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, "validated_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, "valid_completed_final": true, "literal_marker_hits": 13, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "9cc2f488285d73372b8a0154c85d4ae9fd421408729e0e74db19dfd3faeb16ad" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5425, "reasoning_characters": 1877, "wall_time_s": 53.7415, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, "validated_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, "valid_completed_final": true, "literal_marker_hits": 13, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "9cc2f488285d73372b8a0154c85d4ae9fd421408729e0e74db19dfd3faeb16ad" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5425, "reasoning_characters": 1877, "wall_time_s": 53.7452, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, "validated_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 788 }, "valid_completed_final": true, "literal_marker_hits": 13, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "9cc2f488285d73372b8a0154c85d4ae9fd421408729e0e74db19dfd3faeb16ad" } ] }, "long_context_recall_qualified": false }, { "id": "qwen38-gsq-rco-iq2s-mtp", "label": "GSQ-RCO \u00b7 IQ2_S \u00b7 MTP", "name": "Qwen3.8-27B GSQ-RCO IQ2_S MTP", "provider": "GSQ-RCO", "quant": "IQ2_S", "repo_id": "ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF", "revision": "d562806dbafae37109975e970aae91b43e73b440", "filename": "Qwen3.8-27B-GSQ-RCO-IQ2_S-mtp.gguf", "sha256": "E6406238A5CC0043775CD1963B6F9E5B8707400276E38D9FDE742304906B1330", "source_url": "https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF/tree/d562806dbafae37109975e970aae91b43e73b440", "file_bytes": 9607981120, "file_gib": 8.948, "mtp": true, "selected_adaptive": { "mtp": 2, "ubatch": 1024 }, "practical_profile": { "ctx": 73728, "ubatch": 1024, "mtp": 2 }, "balanced_score": 96.43, "category_scores": { "business": 95.0, "research": 97.5, "knowledge": 100.0, "agent_tools": 100.0, "practical_sysadmin": 95.0, "executive_content": 96.4275, "humaneval_plus": 85.37 }, "role_scores": { "CFO / Analyst": 96.82, "Researcher / Scout": 98.14, "Hermes Coordinator": 97.98, "Librarian / Knowledge": 98.59, "Forge / Coder": 91.56 }, "applied_weighted_score": 97.66, "humaneval_plus_passed": 140, "humaneval_base_passed": 144, "humaneval_evaluated": 164, "humaneval_plus_percent": 85.37, "humaneval_base_percent": 87.8, "adaptive_seconds": 26.9291, "fixed_seconds": 25.8562, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 79.8838, "adaptive_repetitions_seconds": [ 79.8838, 26.9158, 26.9291 ], "adaptive_generated_tokens": [ 1766, 1766, 1766 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 67.38, "adaptive_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 616 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 616 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 616 } ], "fixed_quality": [ { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 595 }, { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 595 }, { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 595 } ], "free_vram_mib": 3660.0, "vram_gib": 12.347, "peak_temperature_c": 63.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 3, "practical_repeat_identical_tasks": 24, "run_id": "20261009-201032-v2.1-qwen38-gsq-rco-iq2s-mtp", "scorer_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a", "scorer_rebuilt": false, "scorer_recovery": null, "run_completed_at": "2026-10-09T20:53:28.495933+00:00", "rank": 12, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 3, "validated_heuristic_mean": 98.1, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4306, "reasoning_characters": 1916, "wall_time_s": 80.5761, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 595 }, "validated_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 595 }, "valid_completed_final": true, "literal_marker_hits": 13, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "2cb23b422e959fbc6593e65b0055aef82438ccd43bd4ecd73706ba41ac7ba42d" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4306, "reasoning_characters": 1916, "wall_time_s": 25.8332, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 595 }, "validated_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 595 }, "valid_completed_final": true, "literal_marker_hits": 13, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "2cb23b422e959fbc6593e65b0055aef82438ccd43bd4ecd73706ba41ac7ba42d" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4306, "reasoning_characters": 1916, "wall_time_s": 25.8562, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 595 }, "validated_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 595 }, "valid_completed_final": true, "literal_marker_hits": 13, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "2cb23b422e959fbc6593e65b0055aef82438ccd43bd4ecd73706ba41ac7ba42d" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4552, "reasoning_characters": 1916, "wall_time_s": 79.8838, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 616 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 616 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "cfa309e0f7fb4b02636cfb4686c0435b1a80968b02c61706d27fd6f9dc5b1a3e" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4552, "reasoning_characters": 1916, "wall_time_s": 26.9158, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 616 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 616 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "cfa309e0f7fb4b02636cfb4686c0435b1a80968b02c61706d27fd6f9dc5b1a3e" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4552, "reasoning_characters": 1916, "wall_time_s": 26.9291, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 616 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 616 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "cfa309e0f7fb4b02636cfb4686c0435b1a80968b02c61706d27fd6f9dc5b1a3e" } ] }, "long_context_recall_qualified": false }, { "id": "qwen38-gsq-rco-iq2s", "label": "GSQ-RCO \u00b7 IQ2_S", "name": "Qwen3.8-27B GSQ-RCO IQ2_S", "provider": "GSQ-RCO", "quant": "IQ2_S", "repo_id": "ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF", "revision": "d562806dbafae37109975e970aae91b43e73b440", "filename": "Qwen3.8-27B-GSQ-RCO-IQ2_S.gguf", "sha256": "16C9802111AA9EF3ACDE465188D6D601F8DB128EE3D828AD983A5CACA4135ECB", "source_url": "https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF/tree/d562806dbafae37109975e970aae91b43e73b440", "file_bytes": 9259510912, "file_gib": 8.624, "mtp": false, "selected_adaptive": { "mtp": 0, "ubatch": 512 }, "practical_profile": { "ctx": 73728, "ubatch": 1024, "mtp": 0 }, "balanced_score": 95.54, "category_scores": { "business": 95.0, "research": 95.0, "knowledge": 100.0, "agent_tools": 100.0, "practical_sysadmin": 100.0, "executive_content": 90.0775, "humaneval_plus": 85.37 }, "role_scores": { "CFO / Analyst": 96.0, "Researcher / Scout": 96.51, "Hermes Coordinator": 96.78, "Librarian / Knowledge": 97.01, "Forge / Coder": 92.44 }, "applied_weighted_score": 96.68, "humaneval_plus_passed": 140, "humaneval_base_passed": 146, "humaneval_evaluated": 164, "humaneval_plus_percent": 85.37, "humaneval_base_percent": 89.02, "adaptive_seconds": 35.96, "fixed_seconds": 35.9615, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 87.1573, "adaptive_repetitions_seconds": [ 87.1573, 35.96, 35.9469 ], "adaptive_generated_tokens": [ 1523, 1523, 1523 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 43.0, "adaptive_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 } ], "fixed_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 } ], "free_vram_mib": 5080.0, "vram_gib": 10.96, "peak_temperature_c": 65.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 0, "practical_repeat_identical_tasks": 24, "run_id": "20261009-215125-v2.1-qwen38-gsq-rco-iq2s", "scorer_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a", "scorer_rebuilt": false, "scorer_recovery": null, "run_completed_at": "2026-10-09T22:46:45.106382+00:00", "rank": 13, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 6, "validated_heuristic_mean": 98.6, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3884, "reasoning_characters": 1790, "wall_time_s": 87.251, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "8b02b9dce6a36de27c79d729f51143c7bbf94633d42dfaf7ae1a5739a8e9a3c7" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3884, "reasoning_characters": 1790, "wall_time_s": 35.9615, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "8b02b9dce6a36de27c79d729f51143c7bbf94633d42dfaf7ae1a5739a8e9a3c7" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3884, "reasoning_characters": 1790, "wall_time_s": 35.9557, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "8b02b9dce6a36de27c79d729f51143c7bbf94633d42dfaf7ae1a5739a8e9a3c7" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3884, "reasoning_characters": 1790, "wall_time_s": 87.1573, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "8b02b9dce6a36de27c79d729f51143c7bbf94633d42dfaf7ae1a5739a8e9a3c7" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3884, "reasoning_characters": 1790, "wall_time_s": 35.96, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "8b02b9dce6a36de27c79d729f51143c7bbf94633d42dfaf7ae1a5739a8e9a3c7" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 3884, "reasoning_characters": 1790, "wall_time_s": 35.9469, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 548 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "8b02b9dce6a36de27c79d729f51143c7bbf94633d42dfaf7ae1a5739a8e9a3c7" } ] }, "long_context_recall_qualified": true }, { "id": "qwen38-hauhaucs-uncensored-iq2-m", "label": "HauhauCS Aggressive \u00b7 IQ2_M \u00b7 MTP", "name": "Qwen3.8-27B HauhauCS Uncensored IQ2_M MTP", "provider": "HauhauCS Aggressive", "quant": "IQ2_M", "repo_id": "HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF", "revision": "993a5971fda8f30dd1b7eb2654792ba4415c7460", "filename": "Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-IQ2_M.gguf", "sha256": "5E14825174E02733B873CC01358787D570B648DD530C061623764ABA743512CE", "source_url": "https://huggingface.co/HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF/tree/993a5971fda8f30dd1b7eb2654792ba4415c7460", "file_bytes": 10319906944, "file_gib": 9.611, "mtp": true, "selected_adaptive": { "mtp": 3, "ubatch": 1024 }, "practical_profile": { "ctx": 73728, "ubatch": 256, "mtp": 2 }, "balanced_score": 95.26, "category_scores": { "business": 91.6675, "research": 100.0, "knowledge": 100.0, "agent_tools": 97.5, "practical_sysadmin": 95.0, "executive_content": 86.5075, "humaneval_plus": 89.02 }, "role_scores": { "CFO / Analyst": 94.91, "Researcher / Scout": 97.44, "Hermes Coordinator": 96.43, "Librarian / Knowledge": 97.06, "Forge / Coder": 93.14 }, "applied_weighted_score": 95.95, "humaneval_plus_passed": 146, "humaneval_base_passed": 149, "humaneval_evaluated": 164, "humaneval_plus_percent": 89.02, "humaneval_base_percent": 90.85, "adaptive_seconds": 25.1881, "fixed_seconds": 30.8051, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 75.5002, "adaptive_repetitions_seconds": [ 75.5002, 25.1881, 25.1864 ], "adaptive_generated_tokens": [ 1817, 1817, 1817 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 74.27, "adaptive_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 675 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 675 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 675 } ], "fixed_quality": [ { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 784 }, { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 784 }, { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 784 } ], "free_vram_mib": 3040.0, "vram_gib": 12.952, "peak_temperature_c": 63.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 3, "practical_repeat_identical_tasks": 24, "run_id": "20261010-013048-v2.1-qwen38-hauhaucs-uncensored-iq2-m", "scorer_image_id": "sha256:a1ce362575e724db75b4bf9a744fc51a0b16e2c4a267aeca076217b0971272f5", "scorer_rebuilt": true, "scorer_recovery": { "replacement_image_id": "sha256:a1ce362575e724db75b4bf9a744fc51a0b16e2c4a267aeca076217b0971272f5", "dockerfile_sha256": "a2a2f40af7d5046ed496855fb3bb0a07dbd80e7f3a1da355454c0ff211eaa4a9", "reason": "Original pinned scorer image disappeared from host; rebuilt retained identical Dockerfile. Same pinned Python base and EvalPlus 0.3.1; transitive dependencies were not fully locked in original build.", "isolation_probe_passed": true, "recovered_at": "2026-10-10T00:31:47.125849+00:00", "original_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a" }, "run_completed_at": "2026-10-10T02:17:25.567390+00:00", "rank": 14, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 3, "validated_heuristic_mean": 98.1, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5549, "reasoning_characters": 2146, "wall_time_s": 83.0046, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 784 }, "validated_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 784 }, "valid_completed_final": true, "literal_marker_hits": 13, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "50644a975b77060d4433944dcd5111b75c440ebbbbc703971f4d446e61a1785d" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5549, "reasoning_characters": 2146, "wall_time_s": 30.8051, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 784 }, "validated_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 784 }, "valid_completed_final": true, "literal_marker_hits": 13, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "50644a975b77060d4433944dcd5111b75c440ebbbbc703971f4d446e61a1785d" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5549, "reasoning_characters": 2146, "wall_time_s": 30.8033, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 784 }, "validated_quality": { "overall_hermes": 97.6, "retrieval_facts": "13/14", "word_count": 784 }, "valid_completed_final": true, "literal_marker_hits": 13, "literal_marker_total": 14, "completed_full_marker_recall": false, "final_sha256": "50644a975b77060d4433944dcd5111b75c440ebbbbc703971f4d446e61a1785d" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4656, "reasoning_characters": 2176, "wall_time_s": 75.5002, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 675 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 675 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "ddc4be6f01d933c27ab2314d11b00163666219b349b4d44399bad08ccd798335" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4656, "reasoning_characters": 2176, "wall_time_s": 25.1881, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 675 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 675 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "ddc4be6f01d933c27ab2314d11b00163666219b349b4d44399bad08ccd798335" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4656, "reasoning_characters": 2176, "wall_time_s": 25.1864, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 675 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 675 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "ddc4be6f01d933c27ab2314d11b00163666219b349b4d44399bad08ccd798335" } ] }, "long_context_recall_qualified": false }, { "id": "qwen38-gsq-rco-iq2xs-mtp", "label": "GSQ-RCO \u00b7 IQ2_XS \u00b7 MTP", "name": "Qwen3.8-27B GSQ-RCO IQ2_XS MTP", "provider": "GSQ-RCO", "quant": "IQ2_XS", "repo_id": "ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF", "revision": "d562806dbafae37109975e970aae91b43e73b440", "filename": "Qwen3.8-27B-GSQ-RCO-IQ2_XS-mtp.gguf", "sha256": "F3369F8D36968AB7DD12F5FF11452680AEF9371AA19C9922D22CCD5E58C2D9C8", "source_url": "https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF/tree/d562806dbafae37109975e970aae91b43e73b440", "file_bytes": 8771311680, "file_gib": 8.169, "mtp": true, "selected_adaptive": { "mtp": 3, "ubatch": 1024 }, "practical_profile": { "ctx": 73728, "ubatch": 512, "mtp": 3 }, "balanced_score": 95.0, "category_scores": { "business": 95.0, "research": 95.2275, "knowledge": 100.0, "agent_tools": 90.625, "practical_sysadmin": 100.0, "executive_content": 96.4275, "humaneval_plus": 87.2 }, "role_scores": { "CFO / Analyst": 95.43, "Researcher / Scout": 95.83, "Hermes Coordinator": 95.06, "Librarian / Knowledge": 96.15, "Forge / Coder": 91.96 }, "applied_weighted_score": 95.87, "humaneval_plus_passed": 143, "humaneval_base_passed": 152, "humaneval_evaluated": 164, "humaneval_plus_percent": 87.2, "humaneval_base_percent": 92.68, "adaptive_seconds": 23.0666, "fixed_seconds": 24.9506, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 75.8903, "adaptive_repetitions_seconds": [ 75.8903, 23.0666, 23.0658 ], "adaptive_generated_tokens": [ 1625, 1625, 1625 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 72.72, "adaptive_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 589 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 589 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 589 } ], "fixed_quality": [ { "overall_hermes": 95.7, "retrieval_facts": "14/14", "word_count": 592 }, { "overall_hermes": 95.7, "retrieval_facts": "14/14", "word_count": 592 }, { "overall_hermes": 95.7, "retrieval_facts": "14/14", "word_count": 592 } ], "free_vram_mib": 4186.0, "vram_gib": 11.833, "peak_temperature_c": 63.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 0, "practical_repeat_identical_tasks": 24, "run_id": "20261009-192803-v2.1-qwen38-gsq-rco-iq2xs-mtp", "scorer_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a", "scorer_rebuilt": false, "scorer_recovery": null, "run_completed_at": "2026-10-09T20:10:31.256321+00:00", "rank": 15, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 6, "validated_heuristic_mean": 97.15, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4235, "reasoning_characters": 1982, "wall_time_s": 79.8992, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 95.7, "retrieval_facts": "14/14", "word_count": 592 }, "validated_quality": { "overall_hermes": 95.7, "retrieval_facts": "14/14", "word_count": 592 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "28f64d91097192c8666b44bd2e2685b47a029aa9ae69454942a6dfc13c76e00a" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4235, "reasoning_characters": 1982, "wall_time_s": 24.9506, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 95.7, "retrieval_facts": "14/14", "word_count": 592 }, "validated_quality": { "overall_hermes": 95.7, "retrieval_facts": "14/14", "word_count": 592 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "28f64d91097192c8666b44bd2e2685b47a029aa9ae69454942a6dfc13c76e00a" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4235, "reasoning_characters": 1982, "wall_time_s": 24.9419, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 95.7, "retrieval_facts": "14/14", "word_count": 592 }, "validated_quality": { "overall_hermes": 95.7, "retrieval_facts": "14/14", "word_count": 592 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "28f64d91097192c8666b44bd2e2685b47a029aa9ae69454942a6dfc13c76e00a" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4180, "reasoning_characters": 1988, "wall_time_s": 75.8903, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 589 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 589 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "2cfbcf1144ead189d26a68b4e311d3aa3e1465b1cf8d27d384649e96aea9a20c" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4180, "reasoning_characters": 1988, "wall_time_s": 23.0666, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 589 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 589 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "2cfbcf1144ead189d26a68b4e311d3aa3e1465b1cf8d27d384649e96aea9a20c" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4180, "reasoning_characters": 1988, "wall_time_s": 23.0658, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 589 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 589 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "2cfbcf1144ead189d26a68b4e311d3aa3e1465b1cf8d27d384649e96aea9a20c" } ] }, "long_context_recall_qualified": true }, { "id": "qwen38-unsloth-ud-iq2-s", "label": "Unsloth \u00b7 UD-IQ2_S", "name": "Qwen3.8-27B Unsloth UD-IQ2_S", "provider": "Unsloth", "quant": "UD-IQ2_S", "repo_id": "unsloth/Qwen3.8-27B-GGUF", "revision": "4ca720788d1e01f1bff70c033e0d0028fd02e502", "filename": "Qwen3.8-27B-UD-IQ2_S.gguf", "sha256": "7897D2C5A5CEE46AEF50895141B2C8A0803C1185F3D03C4FDA4CD137A7AD77FE", "source_url": "https://huggingface.co/unsloth/Qwen3.8-27B-GGUF/tree/4ca720788d1e01f1bff70c033e0d0028fd02e502", "file_bytes": 8371970048, "file_gib": 7.797, "mtp": false, "selected_adaptive": { "mtp": 0, "ubatch": 512 }, "practical_profile": { "ctx": 73728, "ubatch": 1024, "mtp": 0 }, "balanced_score": 94.92, "category_scores": { "business": 100.0, "research": 100.0, "knowledge": 95.0, "agent_tools": 100.0, "practical_sysadmin": 97.5, "executive_content": 87.3, "humaneval_plus": 73.17 }, "role_scores": { "CFO / Analyst": 98.61, "Researcher / Scout": 97.48, "Hermes Coordinator": 95.38, "Librarian / Knowledge": 96.34, "Forge / Coder": 85.58 }, "applied_weighted_score": 97.34, "humaneval_plus_passed": 120, "humaneval_base_passed": 128, "humaneval_evaluated": 164, "humaneval_plus_percent": 73.17, "humaneval_base_percent": 78.05, "adaptive_seconds": 48.487, "fixed_seconds": 48.4827, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 98.7852, "adaptive_repetitions_seconds": [ 98.7852, 48.4782, 48.487 ], "adaptive_generated_tokens": [ 2120, 2120, 2120 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 44.21, "adaptive_quality": [ { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 } ], "fixed_quality": [ { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 } ], "free_vram_mib": 5936.0, "vram_gib": 10.124, "peak_temperature_c": 66.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 3, "practical_repeat_identical_tasks": 24, "run_id": "20261010-003417-v2.1-qwen38-unsloth-ud-iq2-s", "scorer_image_id": "sha256:a1ce362575e724db75b4bf9a744fc51a0b16e2c4a267aeca076217b0971272f5", "scorer_rebuilt": true, "scorer_recovery": { "replacement_image_id": "sha256:a1ce362575e724db75b4bf9a744fc51a0b16e2c4a267aeca076217b0971272f5", "dockerfile_sha256": "a2a2f40af7d5046ed496855fb3bb0a07dbd80e7f3a1da355454c0ff211eaa4a9", "reason": "Original pinned scorer image disappeared from host; rebuilt retained identical Dockerfile. Same pinned Python base and EvalPlus 0.3.1; transitive dependencies were not fully locked in original build.", "isolation_probe_passed": true, "recovered_at": "2026-10-10T00:31:47.125849+00:00", "original_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a" }, "run_completed_at": "2026-10-10T01:30:46.957023+00:00", "rank": 16, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 6, "validated_heuristic_mean": 97.1, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5236, "reasoning_characters": 1786, "wall_time_s": 98.8465, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "b930a4394461bfce387293a86a0a67fecbe3290421969116e619c92297705ac9" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5236, "reasoning_characters": 1786, "wall_time_s": 48.4827, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "b930a4394461bfce387293a86a0a67fecbe3290421969116e619c92297705ac9" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5236, "reasoning_characters": 1786, "wall_time_s": 48.482, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "b930a4394461bfce387293a86a0a67fecbe3290421969116e619c92297705ac9" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5236, "reasoning_characters": 1786, "wall_time_s": 98.7852, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "b930a4394461bfce387293a86a0a67fecbe3290421969116e619c92297705ac9" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5236, "reasoning_characters": 1786, "wall_time_s": 48.4782, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "b930a4394461bfce387293a86a0a67fecbe3290421969116e619c92297705ac9" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 5236, "reasoning_characters": 1786, "wall_time_s": 48.487, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, "validated_quality": { "overall_hermes": 97.1, "retrieval_facts": "14/14", "word_count": 716 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "b930a4394461bfce387293a86a0a67fecbe3290421969116e619c92297705ac9" } ] }, "long_context_recall_qualified": true }, { "id": "qwen38-gsq-rco-iq2xs", "label": "GSQ-RCO \u00b7 IQ2_XS", "name": "Qwen3.8-27B GSQ-RCO IQ2_XS", "provider": "GSQ-RCO", "quant": "IQ2_XS", "repo_id": "ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF", "revision": "d562806dbafae37109975e970aae91b43e73b440", "filename": "Qwen3.8-27B-GSQ-RCO-IQ2_XS.gguf", "sha256": "F0AE5006DA0CE6225935339E4E989369F94DE95D2263CF969519F8420C9AE02C", "source_url": "https://huggingface.co/ISTA-DASLab/Qwen3.8-27B-GSQ-RCO-GGUF/tree/d562806dbafae37109975e970aae91b43e73b440", "file_bytes": 8422841472, "file_gib": 7.844, "mtp": false, "selected_adaptive": { "mtp": 0, "ubatch": 512 }, "practical_profile": { "ctx": 73728, "ubatch": 1024, "mtp": 0 }, "balanced_score": 93.46, "category_scores": { "business": 86.6675, "research": 97.7275, "knowledge": 100.0, "agent_tools": 90.625, "practical_sysadmin": 95.0, "executive_content": 96.4275, "humaneval_plus": 85.98 }, "role_scores": { "CFO / Analyst": 91.76, "Researcher / Scout": 95.99, "Hermes Coordinator": 95.25, "Librarian / Knowledge": 96.35, "Forge / Coder": 90.47 }, "applied_weighted_score": 94.3, "humaneval_plus_passed": 141, "humaneval_base_passed": 148, "humaneval_evaluated": 164, "humaneval_plus_percent": 85.98, "humaneval_base_percent": 90.24, "adaptive_seconds": 43.3538, "fixed_seconds": 43.3998, "successful_long_context_output": true, "speed_rank_eligible": true, "long_context_empty_finals": 0, "adaptive_first_request_seconds": 95.1739, "adaptive_repetitions_seconds": [ 95.1739, 43.3538, 43.3349 ], "adaptive_generated_tokens": [ 1894, 1894, 1894 ], "adaptive_prompt_tokens": [ 57135, 57135, 57135 ], "adaptive_cached_prompt_tokens": [ 0, 57131, 57131 ], "adaptive_generation_tps_median": 44.25, "adaptive_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 } ], "fixed_quality": [ { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 } ], "free_vram_mib": 5754.0, "vram_gib": 10.302, "peak_temperature_c": 65.0, "allocation_max_tokens": 131072, "context_allocated_tokens": 73728, "context_exercised_prompt_tokens": 57135, "recall_marker_hits_adaptive": [ 14, 14, 14 ], "practical_responses": 72, "practical_empty": 0, "practical_length_finishes": 3, "practical_repeat_identical_tasks": 24, "run_id": "20261009-224646-v2.1-qwen38-gsq-rco-iq2xs", "scorer_image_id": "sha256:b06643a7ed28f7cd6532c8c8b88bb5a6990460fddd2b62da845d3c66fba2d45a", "scorer_rebuilt": false, "scorer_recovery": null, "run_completed_at": "2026-10-09T23:41:19.407719+00:00", "rank": 17, "long_context_recalculation": { "requests": 6, "empty_finals": 0, "valid_completed_finals": 6, "full_marker_recall_completions": 6, "validated_heuristic_mean": 98.6, "rows": [ { "mode": "fixed", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4976, "reasoning_characters": 2431, "wall_time_s": 95.3003, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "91db0ca4adbf480cb0f62e83b14c218f011e5f8ea2b15d0cf85d6f591b4c5de2" }, { "mode": "fixed", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4976, "reasoning_characters": 2431, "wall_time_s": 43.3998, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "91db0ca4adbf480cb0f62e83b14c218f011e5f8ea2b15d0cf85d6f591b4c5de2" }, { "mode": "fixed", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4976, "reasoning_characters": 2431, "wall_time_s": 43.383, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "91db0ca4adbf480cb0f62e83b14c218f011e5f8ea2b15d0cf85d6f591b4c5de2" }, { "mode": "adaptive", "repeat": 1, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4976, "reasoning_characters": 2431, "wall_time_s": 95.1739, "prompt_tokens": 57135, "cache_tokens": 0, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "91db0ca4adbf480cb0f62e83b14c218f011e5f8ea2b15d0cf85d6f591b4c5de2" }, { "mode": "adaptive", "repeat": 2, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4976, "reasoning_characters": 2431, "wall_time_s": 43.3538, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "91db0ca4adbf480cb0f62e83b14c218f011e5f8ea2b15d0cf85d6f591b4c5de2" }, { "mode": "adaptive", "repeat": 3, "status": "ok", "finish_reason": "stop", "empty_final": false, "final_characters": 4976, "reasoning_characters": 2431, "wall_time_s": 43.3349, "prompt_tokens": 57135, "cache_tokens": 57131, "original_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, "validated_quality": { "overall_hermes": 98.6, "retrieval_facts": "14/14", "word_count": 705 }, "valid_completed_final": true, "literal_marker_hits": 14, "literal_marker_total": 14, "completed_full_marker_recall": true, "final_sha256": "91db0ca4adbf480cb0f62e83b14c218f011e5f8ea2b15d0cf85d6f591b4c5de2" } ] }, "long_context_recall_qualified": true } ], "recalculation": { "status": "RECALCULATED_REFERENCE_FOR_REPORT5", "utc": "2026-10-12T04:28:41.711083+00:00", "original_baseline_sha256": "34c731084d5265c67be57f7b58d242d6f2a7e5a540758a78655b3964bdf39bef", "methodology": "Original practical/coding weights and scores unchanged. Empty final => validated dossier quality zero, regardless of reasoning or API stop. Completion speed requires 3/3 adaptive nonempty stop responses. Separate strict reliability gate requires 6/6 completed responses and all 14 original literal markers in each. Marker matching is a limited heuristic, not semantic correctness. No new weighted aggregate.", "raw_sources": { "runs\\20261009-141105-local-5070ti-v2.1-full\\artifacts\\summary.json": "002b951409c8a6978185bf76db6e45904e7226176d10cba8c2eb33fa89244efe", "runs\\20261010-021726-v2.1-qwen38-hauhaucs-uncensored-q2-k-p\\artifacts\\summary.json": "6aa776d4a2371d7c4a1d075b2b59955812eac38c1e4641d2cd48406e2bb40d67", "runs\\20261009-205329-v2.1-qwen38-pi-iq2m-control\\artifacts\\summary.json": "66fc3d27e0d947a23d1c3a9a82daf4a66f24c94bee57ee363515af122f31b0aa", "runs\\20261009-234120-v2.1-qwen38-gsq-rco-iq3xxs-mtp\\artifacts\\summary.json": "8d4bb3e7475dbd19944b4490033094af51218a554cccbf5395d225274ad28953", "runs\\20261009-183243-v2.1-qwen38-gsq-rco-iq3xxs\\artifacts\\summary.json": "35f4a4e580977dc924a06d6c51fee2f63c3ef7c1e3f291e6fd75bb8b00267a16", "runs\\20261009-163846-v2.1-qwen38-unsloth-ud-q2-k-xl\\artifacts\\summary.json": "8d38478989104f276217eba5f0489606838c509f24ac8fa1f923272705a0a4fe", "runs\\20261010-034645-v2.1-qwen38-orcarouter-uncensored-iq3-xxs\\artifacts\\summary.json": "4231ef9b46754438b934df0f3dab41cd3ed3b01dfa4247d93d21565edde1945a", "runs\\20261010-030000-v2.1-qwen38-hauhaucs-uncensored-iq3-xs\\artifacts\\summary.json": "16911beea49e6452b64736e9741a0fc39d50b1b6cfad660dfaf17c7652fc1e42", "runs\\20261009-173259-v2.1-qwen38-gsq-rco-iq3s\\artifacts\\summary.json": "c18120addab60665b10942ade591470a9224b5458b534adf5aaf6ce3291a9df8", "runs\\20261009-201032-v2.1-qwen38-gsq-rco-iq2s-mtp\\artifacts\\summary.json": "289d6fd365b5a6a37bdf297f851da92f50640eb89887670269d0608db8e95ec4", "runs\\20261009-215125-v2.1-qwen38-gsq-rco-iq2s\\artifacts\\summary.json": "efd178231cb1866313ee7046209c121d024b0cd7c4ae396559e30ceec890d50f", "runs\\20261010-013048-v2.1-qwen38-hauhaucs-uncensored-iq2-m\\artifacts\\summary.json": "f5242cac82b7bf5cff012759578278843b3de90a555e32e5c5228c8031444c16", "runs\\20261009-192803-v2.1-qwen38-gsq-rco-iq2xs-mtp\\artifacts\\summary.json": "3ff2c58245f8c9eeec107770b3fbaecd4e8310e0c42c72088a6b1d30f9115f4f", "runs\\20261010-003417-v2.1-qwen38-unsloth-ud-iq2-s\\artifacts\\summary.json": "0b43633430575220b7f92968b47a475072ec63f7c79ca0ed3d3e162bbbb2e3ae", "runs\\20261009-224646-v2.1-qwen38-gsq-rco-iq2xs\\artifacts\\summary.json": "39ee71db83d159f1b683b4d809f3903417cc35af0261209e9265127f828374ef" }, "responses_rechecked": 102 }, "recalculated_rankings": { "balanced": [ "qwen38-gsq-rco-iq3s-mtp", "qwen38-hauhaucs-uncensored-q2-k-p", "qwen38-pi-iq2m-control", "qwen38-pi-rco-transfer-iq3xxs", "qwen38-gsq-rco-iq3xxs-mtp", "qwen38-gsq-rco-iq3xxs", "qwen38-unsloth-ud-q2-k-xl", "qwen38-orcarouter-uncensored-iq3-xxs", "qwen38-hauhaucs-uncensored-iq3-xs", "qwen38-swift-iq2-s", "qwen38-gsq-rco-iq3s", "qwen38-gsq-rco-iq2s-mtp", "qwen38-gsq-rco-iq2s", "qwen38-hauhaucs-uncensored-iq2-m", "qwen38-gsq-rco-iq2xs-mtp", "qwen38-unsloth-ud-iq2-s", "qwen38-gsq-rco-iq2xs" ], "successful_speed": [ "qwen38-gsq-rco-iq2xs-mtp", "qwen38-hauhaucs-uncensored-iq2-m", "qwen38-orcarouter-uncensored-iq3-xxs", "qwen38-gsq-rco-iq2s-mtp", "qwen38-hauhaucs-uncensored-iq3-xs", "qwen38-gsq-rco-iq3xxs-mtp", "qwen38-gsq-rco-iq3s-mtp", "qwen38-swift-iq2-s", "qwen38-gsq-rco-iq2s", "qwen38-pi-iq2m-control", "qwen38-unsloth-ud-q2-k-xl", "qwen38-gsq-rco-iq2xs", "qwen38-pi-rco-transfer-iq3xxs", "qwen38-unsloth-ud-iq2-s", "qwen38-gsq-rco-iq3s", "qwen38-gsq-rco-iq3xxs" ], "coding": [ "qwen38-gsq-rco-iq3s-mtp", "qwen38-hauhaucs-uncensored-iq2-m", "qwen38-gsq-rco-iq3s", "qwen38-gsq-rco-iq2xs-mtp", "qwen38-gsq-rco-iq3xxs-mtp", "qwen38-orcarouter-uncensored-iq3-xxs", "qwen38-pi-rco-transfer-iq3xxs", "qwen38-gsq-rco-iq3xxs", "qwen38-hauhaucs-uncensored-iq3-xs", "qwen38-gsq-rco-iq2xs", "qwen38-hauhaucs-uncensored-q2-k-p", "qwen38-gsq-rco-iq2s", "qwen38-gsq-rco-iq2s-mtp", "qwen38-unsloth-ud-q2-k-xl", "qwen38-pi-iq2m-control", "qwen38-swift-iq2-s", "qwen38-unsloth-ud-iq2-s" ], "long_context_recall_qualified_practical_coding": [ "qwen38-gsq-rco-iq3s-mtp", "qwen38-pi-iq2m-control", "qwen38-pi-rco-transfer-iq3xxs", "qwen38-gsq-rco-iq3xxs-mtp", "qwen38-gsq-rco-iq3xxs", "qwen38-unsloth-ud-q2-k-xl", "qwen38-orcarouter-uncensored-iq3-xxs", "qwen38-hauhaucs-uncensored-iq3-xs", "qwen38-swift-iq2-s", "qwen38-gsq-rco-iq2s", "qwen38-gsq-rco-iq2xs-mtp", "qwen38-unsloth-ud-iq2-s", "qwen38-gsq-rco-iq2xs" ], "full_recall_successful_speed": [ "qwen38-gsq-rco-iq2xs-mtp", "qwen38-orcarouter-uncensored-iq3-xxs", "qwen38-hauhaucs-uncensored-iq3-xs", "qwen38-gsq-rco-iq3xxs-mtp", "qwen38-gsq-rco-iq3s-mtp", "qwen38-swift-iq2-s", "qwen38-gsq-rco-iq2s", "qwen38-pi-iq2m-control", "qwen38-unsloth-ud-q2-k-xl", "qwen38-gsq-rco-iq2xs", "qwen38-pi-rco-transfer-iq3xxs", "qwen38-unsloth-ud-iq2-s", "qwen38-gsq-rco-iq3xxs" ] }, "reliability_qualified_role_recommendations": { "CFO / Analyst": { "score": 99.64, "model_ids": [ "qwen38-pi-iq2m-control" ] }, "Researcher / Scout": { "score": 99.29, "model_ids": [ "qwen38-pi-iq2m-control" ] }, "Hermes Coordinator": { "score": 98.13, "model_ids": [ "qwen38-hauhaucs-uncensored-iq3-xs" ] }, "Librarian / Knowledge": { "score": 98.93, "model_ids": [ "qwen38-pi-iq2m-control" ] }, "Forge / Coder": { "score": 94.12, "model_ids": [ "qwen38-gsq-rco-iq3s-mtp" ] } } }, "sources": [ { "title": "Pinned Strata source", "url": "https://github.com/Niko1221/Strata/tree/7695859aa13afb248bdb4d36f9b2e95707db079d" }, { "title": "Community benchmark methodology", "url": "https://github.com/Niko1221/Strata/blob/v0.1.43/docs/COMMUNITY_BENCHMARKS.md" }, { "title": "Quantized model revision", "url": "https://huggingface.co/ISTA-DASLab/Qwen3.8-Flash-Next-GSQ-RCO-GGUF/tree/ed59f92082b1e93c0e96d60a8b11aab089b52f09" }, { "title": "Original Qwen model", "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next/tree/de4b8e4d43b917e7706784d8bb445c9af86a3540" }, { "title": "Unbiased Bob\u2019s independent security discussion", "url": "https://www.youtube.com/watch?v=i18vpg0T-mk" }, { "title": "Bob\u2019s reference fork (not tested engine)", "url": "https://github.com/robertkarl/UnbiasedStrata" }, { "title": "Official community benchmark submission \u2014 PR #2053 (awaiting maintainer review)", "url": "https://github.com/Niko1221/Strata/pull/2053" } ], "upstream_submission": { "status": "SUBMITTED_OPEN_AWAITING_MAINTAINER_REVIEW", "url": "https://github.com/Niko1221/Strata/pull/2053", "number": 2053, "commit": "fec32477a902a9225e7f1dbca5a0dc7222bd4aeb", "maintainer_acceptance": "Not observed; submission is open, not merged" } }