{ "caveats": [ "Finite lab cohort; configured 262144 context differs from measured inputs. Aggregate uses first-output to final-completion window. Whole-host memory sampled during speed only; missing counters remain unassessed. Broader coding quality and delivery acceptance unassessed." ], "engine": { "name": "NVIDIA NGC vLLM", "recipe": "with multi-token prediction", "version": "0.29.0+5013de39.dev" }, "format": "olir/0.1", "history": { "corrections": [], "supersedes": [] }, "id": "7fa388c66cc846fd8bed4b9a", "measured_at": "2026-10-07T04:37:19.350179+00:00", "metrics": [ { "basis": null, "n": null, "name": "wall_seconds", "scope": "machine_total", "source_field": "wall_seconds", "statistic": "single", "unit": "s", "value": 79.81211304664612 }, { "basis": null, "n": 8, "name": "ttft_median_s", "scope": "per_person", "source_field": "ttft_median_s", "statistic": "median", "unit": "s", "value": 54.980975570506416 }, { "basis": null, "n": null, "name": "needle_total", "scope": "machine_total", "source_field": "needle_total", "statistic": "single", "unit": "count", "value": 24 }, { "basis": null, "n": 8, "name": "ttft_max_s", "scope": "per_person", "source_field": "ttft_max_s", "statistic": "max", "unit": "s", "value": 65.05511922802543 }, { "basis": null, "n": null, "name": "requests_ok", "scope": "machine_total", "source_field": "requests_ok", "statistic": "single", "unit": "count", "value": 8 }, { "basis": null, "n": null, "name": "needle_correct", "scope": "machine_total", "source_field": "needle_correct", "statistic": "single", "unit": "count", "value": 24 }, { "basis": null, "n": null, "name": "preemptions", "scope": "machine_total", "source_field": "preemptions", "statistic": "single", "unit": "count", "value": 0 }, { "basis": null, "n": null, "name": "decode_overlap", "scope": "machine_total", "source_field": "decode_overlap", "statistic": "single", "unit": "count", "value": 8 }, { "basis": null, "n": null, "name": "aggregate_concurrency", "scope": "machine_total", "source_field": "aggregate_concurrency", "statistic": "single", "unit": "count", "value": 8 }, { "basis": null, "n": null, "name": "requests_total", "scope": "machine_total", "source_field": "requests_total", "statistic": "single", "unit": "count", "value": 8 }, { "basis": null, "n": null, "name": "mtp_acceptance_rate", "scope": "machine_total", "source_field": "mtp_acceptance_rate", "statistic": "single", "unit": "ratio", "value": 0.8618331053351573 }, { "basis": null, "n": null, "name": "correctness_probes_passed", "scope": "machine_total", "source_field": "correctness_probes_passed", "statistic": "single", "unit": "count", "value": 8 }, { "basis": null, "n": null, "name": "aggregate_decode_tps", "scope": "machine_total", "source_field": "aggregate_tps", "statistic": "single", "unit": "tok/s", "value": 149.84376022177028 }, { "basis": null, "n": null, "name": "aggregate_e2e_tps", "scope": "machine_total", "source_field": "aggregate_end_to_end_tps", "statistic": "single", "unit": "tok/s", "value": 51.320530727035084 }, { "basis": null, "n": null, "name": "correctness_probes_total", "scope": "machine_total", "source_field": "correctness_probes_total", "statistic": "single", "unit": "count", "value": 8 }, { "basis": null, "n": 8, "name": "decode_tps", "scope": "per_person", "source_field": "per_request_decode_tps", "statistic": "mean", "unit": "tok/s", "value": 24.112197205943012 }, { "basis": null, "n": null, "name": "forced_output_tokens", "scope": "machine_total", "source_field": "forced_output_tokens", "statistic": "single", "unit": "tokens", "value": 512 }, { "basis": null, "n": null, "name": "peak_host_unavailable_gib", "scope": "machine_total", "source_field": "peak_host_unavailable_gib", "statistic": "max", "unit": "GiB", "value": 93.4132308959961 }, { "basis": null, "n": null, "name": "ttft_s", "scope": "per_person", "source_field": "team_load_requests[0].ttft_s", "statistic": "single", "unit": "s", "value": 54.981445715995505 }, { "basis": null, "n": null, "name": "total_s", "scope": "per_person", "source_field": "team_load_requests[0].total_s", "statistic": "single", "unit": "s", "value": 79.12366025097435 }, { "basis": null, "n": null, "name": "decode_tps", "scope": "per_person", "source_field": "team_load_requests[0].decode_tps", "statistic": "single", "unit": "tok/s", "value": 21.16624385304957 }, { "basis": null, "n": null, "name": "user_number", "scope": "per_person", "source_field": "team_load_requests[0].user_number", "statistic": "single", "unit": "count", "value": 0 }, { "basis": null, "n": null, "name": "prompt_tokens", "scope": "per_person", "source_field": "team_load_requests[0].prompt_tokens", "statistic": "single", "unit": "tokens", "value": 40101 }, { "basis": null, "n": null, "name": "output_tokens", "scope": "per_person", "source_field": "team_load_requests[0].output_tokens", "statistic": "single", "unit": "tokens", "value": 512 }, { "basis": null, "n": null, "name": "needle_correct", "scope": "per_person", "source_field": "team_load_requests[0].needle_correct", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "needle_total", "scope": "per_person", "source_field": "team_load_requests[0].needle_total", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "ttft_s", "scope": "per_person", "source_field": "team_load_requests[1].ttft_s", "statistic": "single", "unit": "s", "value": 52.53015207097633 }, { "basis": null, "n": null, "name": "total_s", "scope": "per_person", "source_field": "team_load_requests[1].total_s", "statistic": "single", "unit": "s", "value": 79.22427673795028 }, { "basis": null, "n": null, "name": "decode_tps", "scope": "per_person", "source_field": "team_load_requests[1].decode_tps", "statistic": "single", "unit": "tok/s", "value": 19.142789148363075 }, { "basis": null, "n": null, "name": "user_number", "scope": "per_person", "source_field": "team_load_requests[1].user_number", "statistic": "single", "unit": "count", "value": 1 }, { "basis": null, "n": null, "name": "prompt_tokens", "scope": "per_person", "source_field": "team_load_requests[1].prompt_tokens", "statistic": "single", "unit": "tokens", "value": 40103 }, { "basis": null, "n": null, "name": "output_tokens", "scope": "per_person", "source_field": "team_load_requests[1].output_tokens", "statistic": "single", "unit": "tokens", "value": 512 }, { "basis": null, "n": null, "name": "needle_correct", "scope": "per_person", "source_field": "team_load_requests[1].needle_correct", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "needle_total", "scope": "per_person", "source_field": "team_load_requests[1].needle_total", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "ttft_s", "scope": "per_person", "source_field": "team_load_requests[2].ttft_s", "statistic": "single", "unit": "s", "value": 54.9799332160037 }, { "basis": null, "n": null, "name": "total_s", "scope": "per_person", "source_field": "team_load_requests[2].total_s", "statistic": "single", "unit": "s", "value": 79.34082178003155 }, { "basis": null, "n": null, "name": "decode_tps", "scope": "per_person", "source_field": "team_load_requests[2].decode_tps", "statistic": "single", "unit": "tok/s", "value": 20.97624635722692 }, { "basis": null, "n": null, "name": "user_number", "scope": "per_person", "source_field": "team_load_requests[2].user_number", "statistic": "single", "unit": "count", "value": 2 }, { "basis": null, "n": null, "name": "prompt_tokens", "scope": "per_person", "source_field": "team_load_requests[2].prompt_tokens", "statistic": "single", "unit": "tokens", "value": 40105 }, { "basis": null, "n": null, "name": "output_tokens", "scope": "per_person", "source_field": "team_load_requests[2].output_tokens", "statistic": "single", "unit": "tokens", "value": 512 }, { "basis": null, "n": null, "name": "needle_correct", "scope": "per_person", "source_field": "team_load_requests[2].needle_correct", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "needle_total", "scope": "per_person", "source_field": "team_load_requests[2].needle_total", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "ttft_s", "scope": "per_person", "source_field": "team_load_requests[3].ttft_s", "statistic": "single", "unit": "s", "value": 65.05511922802543 }, { "basis": null, "n": null, "name": "total_s", "scope": "per_person", "source_field": "team_load_requests[3].total_s", "statistic": "single", "unit": "s", "value": 79.81169429101283 }, { "basis": null, "n": null, "name": "decode_tps", "scope": "per_person", "source_field": "team_load_requests[3].decode_tps", "statistic": "single", "unit": "tok/s", "value": 34.62863149605059 }, { "basis": null, "n": null, "name": "user_number", "scope": "per_person", "source_field": "team_load_requests[3].user_number", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "prompt_tokens", "scope": "per_person", "source_field": "team_load_requests[3].prompt_tokens", "statistic": "single", "unit": "tokens", "value": 40096 }, { "basis": null, "n": null, "name": "output_tokens", "scope": "per_person", "source_field": "team_load_requests[3].output_tokens", "statistic": "single", "unit": "tokens", "value": 512 }, { "basis": null, "n": null, "name": "needle_correct", "scope": "per_person", "source_field": "team_load_requests[3].needle_correct", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "needle_total", "scope": "per_person", "source_field": "team_load_requests[3].needle_total", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "ttft_s", "scope": "per_person", "source_field": "team_load_requests[4].ttft_s", "statistic": "single", "unit": "s", "value": 54.14935153600527 }, { "basis": null, "n": null, "name": "total_s", "scope": "per_person", "source_field": "team_load_requests[4].total_s", "statistic": "single", "unit": "s", "value": 79.07180817902554 }, { "basis": null, "n": null, "name": "decode_tps", "scope": "per_person", "source_field": "team_load_requests[4].decode_tps", "statistic": "single", "unit": "tok/s", "value": 20.503596708758227 }, { "basis": null, "n": null, "name": "user_number", "scope": "per_person", "source_field": "team_load_requests[4].user_number", "statistic": "single", "unit": "count", "value": 4 }, { "basis": null, "n": null, "name": "prompt_tokens", "scope": "per_person", "source_field": "team_load_requests[4].prompt_tokens", "statistic": "single", "unit": "tokens", "value": 40105 }, { "basis": null, "n": null, "name": "output_tokens", "scope": "per_person", "source_field": "team_load_requests[4].output_tokens", "statistic": "single", "unit": "tokens", "value": 512 }, { "basis": null, "n": null, "name": "needle_correct", "scope": "per_person", "source_field": "team_load_requests[4].needle_correct", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "needle_total", "scope": "per_person", "source_field": "team_load_requests[4].needle_total", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "ttft_s", "scope": "per_person", "source_field": "team_load_requests[5].ttft_s", "statistic": "single", "unit": "s", "value": 64.94253399403533 }, { "basis": null, "n": null, "name": "total_s", "scope": "per_person", "source_field": "team_load_requests[5].total_s", "statistic": "single", "unit": "s", "value": 79.79206638300093 }, { "basis": null, "n": null, "name": "decode_tps", "scope": "per_person", "source_field": "team_load_requests[5].decode_tps", "statistic": "single", "unit": "tok/s", "value": 34.41185800434458 }, { "basis": null, "n": null, "name": "user_number", "scope": "per_person", "source_field": "team_load_requests[5].user_number", "statistic": "single", "unit": "count", "value": 5 }, { "basis": null, "n": null, "name": "prompt_tokens", "scope": "per_person", "source_field": "team_load_requests[5].prompt_tokens", "statistic": "single", "unit": "tokens", "value": 40102 }, { "basis": null, "n": null, "name": "output_tokens", "scope": "per_person", "source_field": "team_load_requests[5].output_tokens", "statistic": "single", "unit": "tokens", "value": 512 }, { "basis": null, "n": null, "name": "needle_correct", "scope": "per_person", "source_field": "team_load_requests[5].needle_correct", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "needle_total", "scope": "per_person", "source_field": "team_load_requests[5].needle_total", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "ttft_s", "scope": "per_person", "source_field": "team_load_requests[6].ttft_s", "statistic": "single", "unit": "s", "value": 54.98134611500427 }, { "basis": null, "n": null, "name": "total_s", "scope": "per_person", "source_field": "team_load_requests[6].total_s", "statistic": "single", "unit": "s", "value": 79.17577939503826 }, { "basis": null, "n": null, "name": "decode_tps", "scope": "per_person", "source_field": "team_load_requests[6].decode_tps", "statistic": "single", "unit": "tok/s", "value": 21.120560836681936 }, { "basis": null, "n": null, "name": "user_number", "scope": "per_person", "source_field": "team_load_requests[6].user_number", "statistic": "single", "unit": "count", "value": 6 }, { "basis": null, "n": null, "name": "prompt_tokens", "scope": "per_person", "source_field": "team_load_requests[6].prompt_tokens", "statistic": "single", "unit": "tokens", "value": 40104 }, { "basis": null, "n": null, "name": "output_tokens", "scope": "per_person", "source_field": "team_load_requests[6].output_tokens", "statistic": "single", "unit": "tokens", "value": 512 }, { "basis": null, "n": null, "name": "needle_correct", "scope": "per_person", "source_field": "team_load_requests[6].needle_correct", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "needle_total", "scope": "per_person", "source_field": "team_load_requests[6].needle_total", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "ttft_s", "scope": "per_person", "source_field": "team_load_requests[7].ttft_s", "statistic": "single", "unit": "s", "value": 54.98060502600856 }, { "basis": null, "n": null, "name": "total_s", "scope": "per_person", "source_field": "team_load_requests[7].total_s", "statistic": "single", "unit": "s", "value": 79.37474802898942 }, { "basis": null, "n": null, "name": "decode_tps", "scope": "per_person", "source_field": "team_load_requests[7].decode_tps", "statistic": "single", "unit": "tok/s", "value": 20.94765124306921 }, { "basis": null, "n": null, "name": "user_number", "scope": "per_person", "source_field": "team_load_requests[7].user_number", "statistic": "single", "unit": "count", "value": 7 }, { "basis": null, "n": null, "name": "prompt_tokens", "scope": "per_person", "source_field": "team_load_requests[7].prompt_tokens", "statistic": "single", "unit": "tokens", "value": 40100 }, { "basis": null, "n": null, "name": "output_tokens", "scope": "per_person", "source_field": "team_load_requests[7].output_tokens", "statistic": "single", "unit": "tokens", "value": 512 }, { "basis": null, "n": null, "name": "needle_correct", "scope": "per_person", "source_field": "team_load_requests[7].needle_correct", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": null, "name": "needle_total", "scope": "per_person", "source_field": "team_load_requests[7].needle_total", "statistic": "single", "unit": "count", "value": 3 }, { "basis": null, "n": 8, "name": "decode_tps", "scope": "per_person", "source_field": "per_person_decode_median_tps", "statistic": "median", "unit": "tok/s", "value": 21.04840359695443 } ], "model": { "artifact_sha256": [ "07141c2db92e47bc08777132cd1a0323faf300eab3a7d7c111bc2bf075fda050", "6dea9c759a0f941cf9e1cc1501216b0e107966a66c90425050e212efbd053f02", "9758875fc55e49561165f4a44342b654c123c3e25c6811b34abb83553fb1a164" ], "publisher_repo": "Qwen/Qwen3.6-35B-A3B", "publisher_revision": "995ad96eacd98c81ed38be0c5b274b04031597b0", "variant": "NVFP4 variant", "weights_dtype": "nvfp4", "weights_format": "safetensors" }, "outcome": { "gates": [ { "met": true, "name": "requests", "observed": "8/8" }, { "met": true, "name": "hidden-facts", "observed": "24/24" }, { "met": true, "name": "correctness-probes", "observed": "8/8" } ], "reasons": [], "status": "pass" }, "publication_group": "pre-standard", "published": null, "publisher": { "name": "Skilak Keep" }, "quality": [ { "passed": 24, "skipped": null, "suite": "hidden-facts", "total": 24, "version": "1" }, { "passed": 8, "skipped": null, "suite": "correctness-probes", "total": 8, "version": "1" } ], "record_date": "2026-10-07", "reproduce": { "driver": null, "fixture_sha256": "e2097959b1ca0142f2b93edc3ec411e474c0a6622800dd247af7a71bd9c39e83", "harness": { "name": "keep-team-load-v1", "version": "1" }, "harness_sha256": null, "os": null, "workload_name": null, "workload_version": null }, "requests": [ { "cached_prompt_tokens": null, "checks_passed": true, "completed": true, "decode_tps": 21.16624385304957, "effective_prefill_tps": null, "needle_correct": 3, "needle_total": 3, "output_tokens": 512, "prompt_tokens": 40101, "total_s": 79.12366025097435, "ttft_s": 54.981445715995505, "user_number": 0 }, { "cached_prompt_tokens": null, "checks_passed": true, "completed": true, "decode_tps": 19.142789148363075, "effective_prefill_tps": null, "needle_correct": 3, "needle_total": 3, "output_tokens": 512, "prompt_tokens": 40103, "total_s": 79.22427673795028, "ttft_s": 52.53015207097633, "user_number": 1 }, { "cached_prompt_tokens": null, "checks_passed": true, "completed": true, "decode_tps": 20.97624635722692, "effective_prefill_tps": null, "needle_correct": 3, "needle_total": 3, "output_tokens": 512, "prompt_tokens": 40105, "total_s": 79.34082178003155, "ttft_s": 54.9799332160037, "user_number": 2 }, { "cached_prompt_tokens": null, "checks_passed": true, "completed": true, "decode_tps": 34.62863149605059, "effective_prefill_tps": null, "needle_correct": 3, "needle_total": 3, "output_tokens": 512, "prompt_tokens": 40096, "total_s": 79.81169429101283, "ttft_s": 65.05511922802543, "user_number": 3 }, { "cached_prompt_tokens": null, "checks_passed": true, "completed": true, "decode_tps": 20.503596708758227, "effective_prefill_tps": null, "needle_correct": 3, "needle_total": 3, "output_tokens": 512, "prompt_tokens": 40105, "total_s": 79.07180817902554, "ttft_s": 54.14935153600527, "user_number": 4 }, { "cached_prompt_tokens": null, "checks_passed": true, "completed": true, "decode_tps": 34.41185800434458, "effective_prefill_tps": null, "needle_correct": 3, "needle_total": 3, "output_tokens": 512, "prompt_tokens": 40102, "total_s": 79.79206638300093, "ttft_s": 64.94253399403533, "user_number": 5 }, { "cached_prompt_tokens": null, "checks_passed": true, "completed": true, "decode_tps": 21.120560836681936, "effective_prefill_tps": null, "needle_correct": 3, "needle_total": 3, "output_tokens": 512, "prompt_tokens": 40104, "total_s": 79.17577939503826, "ttft_s": 54.98134611500427, "user_number": 6 }, { "cached_prompt_tokens": null, "checks_passed": true, "completed": true, "decode_tps": 20.94765124306921, "effective_prefill_tps": null, "needle_correct": 3, "needle_total": 3, "output_tokens": 512, "prompt_tokens": 40100, "total_s": 79.37474802898942, "ttft_s": 54.98060502600856, "user_number": 7 } ], "scenario": { "id": "team-load", "version": "1" }, "settings": { "concurrency": 8, "configured_context_tokens": 262144, "context_tokens": 262144, "kv_dtype": "fp8", "max_output_tokens": 512, "speculative_decoding": "with multi-token prediction", "thinking": false }, "setup_complete": false, "setup_id": "cf0bbc6439f3c0ed87c21751d34d5655", "standard": "pre-standard", "system": { "driver": null, "hardware": "NVIDIA GB10, 128 GB unified memory", "os": null }, "workload": { "cache": null, "content_tokens": null, "document": null, "input_tokens": { "max": 40105, "median": 40102.5, "min": 40096 }, "output_tokens": 512, "repetitions": null, "users": 8, "warmup": null } }