{ "base_url": "[operator identity redacted]", "clock_domain_id": null, "completed": 12, "concurrency": 4, "configuration_fingerprint": null, "context_distribution": { "8192": 12 }, "context_seed": 17, "context_tokens": 8192, "controlled_output_policy": "observe", "engine": "ninfer", "failed": 0, "failures": [], "finished_at": "2026-10-05T09:22:43Z", "finished_at_unix_ns": 1791192163152789100, "gpu": "rtx5090", "manifest": null, "max_context_tokens": 262144, "max_tokens": 8192, "measurement_protocol": "capacity-v4-reasoning", "metric_population": { "aggregate_throughput_eligible": true, "attempted_requests": 12, "excluded_requests": 0, "failure_policy": "failed or contract-invalid responses retain sanitized timing and metadata but are excluded from performance distributions", "p99_interpretation": "descriptive-nearest-rank-only", "p99_interpretation_by_metric": { "decode_tok_s": "descriptive-nearest-rank-only", "e2e": "descriptive-nearest-rank-only", "effective_prefill_tok_s": "descriptive-nearest-rank-only", "tpot": "descriptive-nearest-rank-only", "ttft": "descriptive-nearest-rank-only" }, "p99_minimum_recommended_samples": 100, "p99_samples_by_metric": { "decode_tok_s": 12, "e2e": 12, "effective_prefill_tok_s": 12, "tpot": 12, "ttft": 12 }, "performance_eligible_requests": 12, "successful_requests": 12 }, "metrics": { "completion_tokens_max": 1075.0, "completion_tokens_mean": 322.5833333333333, "completion_tokens_mean_ci95_high": 461.9630551276989, "completion_tokens_mean_ci95_low": 183.20361153896775, "completion_tokens_min": 142.0, "completion_tokens_p25": 182.0, "completion_tokens_p50": 254.0, "completion_tokens_p75": 317.0, "completion_tokens_p90": 365.0, "completion_tokens_p95": 1075.0, "completion_tokens_p99": 1075.0, "completion_tokens_samples": 12, "completion_tokens_stddev": 246.33955070679218, "content_chunks": 103, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 190.0468359081069, "decode_tok_s_mean": 106.4300564235535, "decode_tok_s_mean_ci95_high": 126.11127221091745, "decode_tok_s_mean_ci95_low": 86.74884063618954, "decode_tok_s_min": 64.83959893835588, "decode_tok_s_p25": 79.73071405486952, "decode_tok_s_p50": 103.56026898131675, "decode_tok_s_p75": 122.74156099197825, "decode_tok_s_p90": 134.96793551288846, "decode_tok_s_p95": 190.0468359081069, "decode_tok_s_p99": 190.0468359081069, "decode_tok_s_samples": 12, "decode_tok_s_stddev": 34.78455683514394, "e2e_max_ms": 9406.110500000068, "e2e_mean_ci95_high_ms": 5510.054237379341, "e2e_mean_ci95_low_ms": 3089.0101126205936, "e2e_mean_ms": 4299.5321749999675, "e2e_min_ms": 2150.733099999343, "e2e_p25_ms": 2545.1195999994525, "e2e_p50_ms": 3588.9893999992637, "e2e_p75_ms": 5131.659900000159, "e2e_p90_ms": 6589.487100000042, "e2e_p95_ms": 9406.110500000068, "e2e_p99_ms": 9406.110500000068, "e2e_samples": 12, "e2e_stddev_ms": 2139.4752201266706, "effective_prefill_tok_s_max": 11724.968890129623, "effective_prefill_tok_s_mean": 8692.077381055138, "effective_prefill_tok_s_mean_ci95_high": 10763.093737360297, "effective_prefill_tok_s_mean_ci95_low": 6621.06102474998, "effective_prefill_tok_s_min": 2278.071754101909, "effective_prefill_tok_s_p25": 4094.884513270737, "effective_prefill_tok_s_p50": 10554.901263504711, "effective_prefill_tok_s_p75": 11283.96717971, "effective_prefill_tok_s_p90": 11346.76996554428, "effective_prefill_tok_s_p95": 11724.968890129623, "effective_prefill_tok_s_p99": 11724.968890129623, "effective_prefill_tok_s_samples": 12, "effective_prefill_tok_s_stddev": 3660.311788190513, "generation_max_ms": 8750.09239999963, "generation_mean_ci95_high_ms": 4269.8414634472865, "generation_mean_ci95_low_ms": 1963.8980198863312, "generation_mean_ms": 3116.8697416668087, "generation_min_ms": 1410.1786999999604, "generation_p25_ms": 1865.86989999887, "generation_p50_ms": 2124.96850000025, "generation_p75_ms": 2956.1518999998952, "generation_p90_ms": 4873.564999999871, "generation_p95_ms": 8750.09239999963, "generation_p99_ms": 8750.09239999963, "generation_samples": 12, "generation_stddev_ms": 2037.7608181775051, "mean_inter_token_latency_ms_max": 15.422674050632505, "mean_inter_token_latency_ms_mean": 10.260685990046754, "mean_inter_token_latency_ms_mean_ci95_high": 12.003475995677245, "mean_inter_token_latency_ms_mean_ci95_low": 8.517895984416263, "mean_inter_token_latency_ms_min": 5.261860820895374, "mean_inter_token_latency_ms_p25": 8.09125989010951, "mean_inter_token_latency_ms_p50": 9.191477339895911, "mean_inter_token_latency_ms_p75": 11.824607599999581, "mean_inter_token_latency_ms_p90": 15.070698581562057, "mean_inter_token_latency_ms_p95": 15.422674050632505, "mean_inter_token_latency_ms_p99": 15.422674050632505, "mean_inter_token_latency_ms_samples": 12, "mean_inter_token_latency_ms_stddev": 3.0802049353829197, "mean_time_per_output_token_max_ms": 15.422674050632505, "mean_time_per_output_token_mean_ci95_high_ms": 12.003475995677245, "mean_time_per_output_token_mean_ci95_low_ms": 8.517895984416263, "mean_time_per_output_token_mean_ms": 10.260685990046754, "mean_time_per_output_token_min_ms": 5.261860820895374, "mean_time_per_output_token_p25_ms": 8.09125989010951, "mean_time_per_output_token_p50_ms": 9.191477339895911, "mean_time_per_output_token_p75_ms": 11.824607599999581, "mean_time_per_output_token_p90_ms": 15.070698581562057, "mean_time_per_output_token_p95_ms": 15.422674050632505, "mean_time_per_output_token_p99_ms": 15.422674050632505, "mean_time_per_output_token_samples": 12, "mean_time_per_output_token_stddev_ms": 3.0802049353829197, "output_token_sources": { "usage": 12 }, "output_tokens": 3871, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 12, "prompt_tokens": 89085, "prompt_tokens_max": 7433.0, "prompt_tokens_mean": 7423.75, "prompt_tokens_mean_ci95_high": 7427.21377312225, "prompt_tokens_mean_ci95_low": 7420.28622687775, "prompt_tokens_min": 7415.0, "prompt_tokens_p25": 7418.0, "prompt_tokens_p50": 7422.0, "prompt_tokens_p75": 7427.0, "prompt_tokens_p90": 7433.0, "prompt_tokens_p95": 7433.0, "prompt_tokens_p99": 7433.0, "prompt_tokens_samples": 12, "prompt_tokens_stddev": 6.1218684016624145, "reasoning_chunks": 999, "request_canary_passed": 12, "request_canary_samples": 12, "throughput_performance_eligible": true, "throughput_tok_s": 266.22515792942846, "time_to_first_output_max_ms": 3257.140600000639, "time_to_first_output_mean_ci95_high_ms": 1690.7813641057558, "time_to_first_output_mean_ci95_low_ms": 674.5435025605623, "time_to_first_output_mean_ms": 1182.662433333159, "time_to_first_output_min_ms": 632.8374999993684, "time_to_first_output_p25_ms": 656.0181000004377, "time_to_first_output_p50_ms": 679.2497000005824, "time_to_first_output_p75_ms": 1110.654500000237, "time_to_first_output_p90_ms": 2626.146199998402, "time_to_first_output_p95_ms": 3257.140600000639, "time_to_first_output_p99_ms": 3257.140600000639, "time_to_first_output_samples": 12, "time_to_first_output_stddev_ms": 898.048779985419, "tpot_max_ms": 15.422674050632505, "tpot_mean_ci95_high_ms": 12.003475995677245, "tpot_mean_ci95_low_ms": 8.517895984416263, "tpot_mean_ms": 10.260685990046754, "tpot_min_ms": 5.261860820895374, "tpot_p25_ms": 8.09125989010951, "tpot_p50_ms": 9.191477339895911, "tpot_p75_ms": 11.824607599999581, "tpot_p90_ms": 15.070698581562057, "tpot_p95_ms": 15.422674050632505, "tpot_p99_ms": 15.422674050632505, "tpot_samples": 12, "tpot_stddev_ms": 3.0802049353829197, "ttft_max_ms": 9240.16809999921, "ttft_mean_ci95_high_ms": 5243.762514818229, "ttft_mean_ci95_low_ms": 2915.6391518485525, "ttft_mean_ms": 4079.7008333333906, "ttft_min_ms": 2031.1125999996875, "ttft_p25_ms": 2436.043800000334, "ttft_p50_ms": 3453.0902999995305, "ttft_p75_ms": 4937.5627000008535, "ttft_p90_ms": 5925.403499999447, "ttft_p95_ms": 9240.16809999921, "ttft_p99_ms": 9240.16809999921, "ttft_samples": 12, "ttft_stddev_ms": 2057.361199465101 }, "model": "qwen38-swift15-ninfer-dflash2-c4-262k", "performance_eligible": true, "prompt_cache_mode": "unique", "request_canaries": true, "request_timings": [ { "content_chunks": 13, "controlled_output": null, "decode_tok_s": 64.83959893835588, "e2e_ms": 5984.219500000108, "effective_prefill_tok_s": 6676.243602306944, "generation_ms": 4873.564999999871, "mean_inter_token_latency_ms": 15.422674050632505, "mean_time_per_output_token_ms": 15.422674050632505, "output_token_source": "usage", "output_tokens": 317, "planned_context_tokens": 8192, "prompt_tokens": 7415, "reasoning_chunks": 86, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00000", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 0, "time_to_first_output_ms": 1110.654500000237, "tpot_ms": 15.422674050632505, "ttft_ms": 5453.364599999986, "visible_content_truncated": false, "visible_generation_ms": 530.8549000001221 }, { "content_chunks": 7, "controlled_output": null, "decode_tok_s": 79.73071405486952, "e2e_ms": 6589.487100000042, "effective_prefill_tok_s": 2824.29058976401, "generation_ms": 3963.34090000164, "mean_inter_token_latency_ms": 12.542218037979874, "mean_time_per_output_token_ms": 12.542218037979874, "output_token_source": "usage", "output_tokens": 317, "planned_context_tokens": 8192, "prompt_tokens": 7417, "reasoning_chunks": 84, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00001", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 1, "time_to_first_output_ms": 2626.146199998402, "tpot_ms": 12.542218037979874, "ttft_ms": 5925.403499999447, "visible_content_truncated": false, "visible_generation_ms": 664.0836000005947 }, { "content_chunks": 6, "controlled_output": null, "decode_tok_s": 66.35392477581829, "e2e_ms": 3936.4970000005997, "effective_prefill_tok_s": 4094.884513270737, "generation_ms": 2124.96850000025, "mean_inter_token_latency_ms": 15.070698581562057, "mean_time_per_output_token_ms": 15.070698581562057, "output_token_source": "usage", "output_tokens": 142, "planned_context_tokens": 8192, "prompt_tokens": 7418, "reasoning_chunks": 40, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00002", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 2, "time_to_first_output_ms": 1811.5285000003496, "tpot_ms": 15.070698581562057, "ttft_ms": 3839.071200000035, "visible_content_truncated": false, "visible_generation_ms": 97.42580000056478 }, { "content_chunks": 10, "controlled_output": null, "decode_tok_s": 134.96793551288846, "e2e_ms": 5131.659900000159, "effective_prefill_tok_s": 2278.071754101909, "generation_ms": 1874.5192999995197, "mean_inter_token_latency_ms": 7.409167193673991, "mean_time_per_output_token_ms": 7.409167193673991, "output_token_source": "usage", "output_tokens": 254, "planned_context_tokens": 8192, "prompt_tokens": 7420, "reasoning_chunks": 59, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00003", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 3, "time_to_first_output_ms": 3257.140600000639, "tpot_ms": 7.409167193673991, "ttft_ms": 4937.5627000008535, "visible_content_truncated": false, "visible_generation_ms": 194.0971999993053 }, { "content_chunks": 8, "controlled_output": null, "decode_tok_s": 84.56940253983865, "e2e_ms": 3588.9893999992637, "effective_prefill_tok_s": 11724.968890129623, "generation_ms": 2956.1518999998952, "mean_inter_token_latency_ms": 11.824607599999581, "mean_time_per_output_token_ms": 11.824607599999581, "output_token_source": "usage", "output_tokens": 251, "planned_context_tokens": 8192, "prompt_tokens": 7420, "reasoning_chunks": 58, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00004", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 4, "time_to_first_output_ms": 632.8374999993684, "tpot_ms": 11.824607599999581, "ttft_ms": 3453.0902999995305, "visible_content_truncated": false, "visible_generation_ms": 135.89909999973315 }, { "content_chunks": 11, "controlled_output": null, "decode_tok_s": 122.74156099197825, "e2e_ms": 9406.110500000068, "effective_prefill_tok_s": 11313.712228359322, "generation_ms": 8750.09239999963, "mean_inter_token_latency_ms": 8.147199627560177, "mean_time_per_output_token_ms": 8.147199627560177, "output_token_source": "usage", "output_tokens": 1075, "planned_context_tokens": 8192, "prompt_tokens": 7422, "reasoning_chunks": 280, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00005", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 5, "time_to_first_output_ms": 656.0181000004377, "tpot_ms": 8.147199627560177, "ttft_ms": 9240.16809999921, "visible_content_truncated": false, "visible_generation_ms": 165.94240000085847 }, { "content_chunks": 9, "controlled_output": null, "decode_tok_s": 86.9088398293583, "e2e_ms": 2743.8999000005424, "effective_prefill_tok_s": 11227.087041239962, "generation_ms": 2082.642000001215, "mean_inter_token_latency_ms": 11.506309392271906, "mean_time_per_output_token_ms": 11.506309392271906, "output_token_source": "usage", "output_tokens": 182, "planned_context_tokens": 8192, "prompt_tokens": 7424, "reasoning_chunks": 42, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00006", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 6, "time_to_first_output_ms": 661.2578999993275, "tpot_ms": 11.506309392271906, "ttft_ms": 2588.590600000316, "visible_content_truncated": false, "visible_generation_ms": 155.30930000022636 }, { "content_chunks": 9, "controlled_output": null, "decode_tok_s": 111.0550098612172, "e2e_ms": 3530.9870000000956, "effective_prefill_tok_s": 10554.901263504711, "generation_ms": 2827.4276000011014, "mean_inter_token_latency_ms": 9.004546496818794, "mean_time_per_output_token_ms": 9.004546496818794, "output_token_source": "usage", "output_tokens": 315, "planned_context_tokens": 8192, "prompt_tokens": 7426, "reasoning_chunks": 80, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00007", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 7, "time_to_first_output_ms": 703.5593999989942, "tpot_ms": 9.004546496818794, "ttft_ms": 3356.045599999561, "visible_content_truncated": false, "visible_generation_ms": 174.9414000005345 }, { "content_chunks": 8, "controlled_output": null, "decode_tok_s": 123.59014709469004, "e2e_ms": 3603.4091000001354, "effective_prefill_tok_s": 11283.96717971, "generation_ms": 2945.218599999862, "mean_inter_token_latency_ms": 8.09125989010951, "mean_time_per_output_token_ms": 8.09125989010951, "output_token_source": "usage", "output_tokens": 365, "planned_context_tokens": 8192, "prompt_tokens": 7427, "reasoning_chunks": 85, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00008", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 8, "time_to_first_output_ms": 658.1905000002735, "tpot_ms": 8.09125989010951, "ttft_ms": 3468.435700000555, "visible_content_truncated": false, "visible_generation_ms": 134.97339999958058 }, { "content_chunks": 9, "controlled_output": null, "decode_tok_s": 103.56026898131675, "e2e_ms": 2383.273999999801, "effective_prefill_tok_s": 11346.76996554428, "generation_ms": 1728.4620999998879, "mean_inter_token_latency_ms": 9.656212849161385, "mean_time_per_output_token_ms": 9.656212849161385, "output_token_source": "usage", "output_tokens": 180, "planned_context_tokens": 8192, "prompt_tokens": 7430, "reasoning_chunks": 51, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00009", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 9, "time_to_first_output_ms": 654.8118999999133, "tpot_ms": 9.656212849161385, "ttft_ms": 2227.5213000011718, "visible_content_truncated": false, "visible_generation_ms": 155.7526999986294 }, { "content_chunks": 6, "controlled_output": null, "decode_tok_s": 108.79643859420366, "e2e_ms": 2545.1195999994525, "effective_prefill_tok_s": 10942.956618153275, "generation_ms": 1865.86989999887, "mean_inter_token_latency_ms": 9.191477339895911, "mean_time_per_output_token_ms": 9.191477339895911, "output_token_source": "usage", "output_tokens": 204, "planned_context_tokens": 8192, "prompt_tokens": 7433, "reasoning_chunks": 60, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00010", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 10, "time_to_first_output_ms": 679.2497000005824, "tpot_ms": 9.191477339895911, "ttft_ms": 2436.043800000334, "visible_content_truncated": false, "visible_generation_ms": 109.07579999911832 }, { "content_chunks": 7, "controlled_output": null, "decode_tok_s": 190.0468359081069, "e2e_ms": 2150.733099999343, "effective_prefill_tok_s": 10037.074926576897, "generation_ms": 1410.1786999999604, "mean_inter_token_latency_ms": 5.261860820895374, "mean_time_per_output_token_ms": 5.261860820895374, "output_token_source": "usage", "output_tokens": 269, "planned_context_tokens": 8192, "prompt_tokens": 7433, "reasoning_chunks": 74, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00011", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 11, "time_to_first_output_ms": 740.5543999993824, "tpot_ms": 5.261860820895374, "ttft_ms": 2031.1125999996875, "visible_content_truncated": false, "visible_generation_ms": 119.6204999996553 } ], "requests": 12, "response_words": 0, "responses_received": 12, "run_id": "benchmark-20261005T092228Z", "sampling": { "temperature": { "effective_request": 0.9, "requested": 0.9, "sent": true }, "top_p": { "effective_request": 0.95, "requested": 0.95, "sent": true } }, "schema": "anvil-serving.benchmark/v1", "serve_flags": { "no_thinking": false, "reasoning_effort": null, "shared_prefix_burst": false, "thinking_mode": "default" }, "shared_prefix_tokens": 0, "source_recipe": null, "started_at": "2026-10-05T09:22:28Z", "started_at_unix_ns": 1791192148612462600, "tier": null, "timestamp_precision": "nanosecond-unix-fields; ISO labels are second precision", "timing_methodology": { "artifact_timestamps": "time.time_ns Unix timestamps; independent artifacts may align them only when the operator supplied the same explicit clock_domain_id", "clock": "client time.perf_counter", "controlled_output": "--response-words adherence is recorded; observe preserves historical behavior while strict rejects non-exact or truncated completions; visible response capture is capped at 8192 characters", "decode": "usage completion tokens after the first token divided by client-observed generation time; completion tokens may include reasoning tokens", "distributions": "request-level mean, sample standard deviation, normal-approximation 95% confidence interval for the mean, min, p25, p50, p75, p90, p95, p99, and max using the nearest-rank percentile method", "effective_prefill": "usage.prompt_tokens divided by client-observed time to first output; includes queueing, scheduling, prefill, and first-token work", "generation": "client-observed E2E minus time to first output", "mean_inter_token_latency": "legacy key for a TPOT proxy: client-observed generation time divided by completion tokens after the first token; not raw per-token ITL", "p99": "numeric nearest-rank p99 is always retained for compatibility; it is descriptive only below 100 eligible samples", "request_canaries": "when enabled, each response must begin with its own synthetic marker and contain no other ANVIL_REQ marker; bounded/truncated captures fail closed and are excluded from performance metrics", "time_to_first_output": "request start through the first non-empty streamed reasoning or content delta", "tpot": "client-observed mean time per output token after the first output; identical to the legacy mean_inter_token_latency key", "ttft": "request start through the first non-empty streamed content delta" }, "wall_clock_ms": 14540.323799999896 }