{ "base_url": "[operator identity redacted]", "clock_domain_id": null, "completed": 12, "concurrency": 1, "configuration_fingerprint": null, "context_distribution": { "24576": 12 }, "context_seed": 17, "context_tokens": 24576, "controlled_output_policy": "observe", "engine": "ninfer", "failed": 0, "failures": [], "finished_at": "2026-10-05T09:24:02Z", "finished_at_unix_ns": 1791192242437705800, "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": 425.0, "completion_tokens_mean": 311.5833333333333, "completion_tokens_mean_ci95_high": 351.775614654226, "completion_tokens_mean_ci95_low": 271.39105201244064, "completion_tokens_min": 195.0, "completion_tokens_p25": 247.0, "completion_tokens_p50": 299.0, "completion_tokens_p75": 344.0, "completion_tokens_p90": 402.0, "completion_tokens_p95": 425.0, "completion_tokens_p99": 425.0, "completion_tokens_samples": 12, "completion_tokens_stddev": 71.03578910192617, "content_chunks": 99, "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": 272.491067731039, "decode_tok_s_mean": 246.1763068887028, "decode_tok_s_mean_ci95_high": 255.32977130948382, "decode_tok_s_mean_ci95_low": 237.0228424679218, "decode_tok_s_min": 223.07548785844702, "decode_tok_s_p25": 229.51344464649634, "decode_tok_s_p50": 248.17709082516205, "decode_tok_s_p75": 258.5976080942042, "decode_tok_s_p90": 262.1746092070444, "decode_tok_s_p95": 272.491067731039, "decode_tok_s_p99": 272.491067731039, "decode_tok_s_samples": 12, "decode_tok_s_stddev": 16.177821879659906, "e2e_max_ms": 3616.7561000002024, "e2e_mean_ci95_high_ms": 3401.3898963878005, "e2e_mean_ci95_low_ms": 3058.6701036122363, "e2e_mean_ms": 3230.0300000000184, "e2e_min_ms": 2717.003800000384, "e2e_p25_ms": 3010.691299999962, "e2e_p50_ms": 3256.4419000009366, "e2e_p75_ms": 3474.310900001001, "e2e_p90_ms": 3589.485299999069, "e2e_p95_ms": 3616.7561000002024, "e2e_p99_ms": 3616.7561000002024, "e2e_samples": 12, "e2e_stddev_ms": 302.8612723707926, "effective_prefill_tok_s_max": 11603.669565646976, "effective_prefill_tok_s_mean": 11452.675463135476, "effective_prefill_tok_s_mean_ci95_high": 11595.920295900967, "effective_prefill_tok_s_mean_ci95_low": 11309.430630369985, "effective_prefill_tok_s_min": 10663.78767656273, "effective_prefill_tok_s_p25": 11454.166961779063, "effective_prefill_tok_s_p50": 11530.509959069235, "effective_prefill_tok_s_p75": 11538.396566454046, "effective_prefill_tok_s_p90": 11579.786520694724, "effective_prefill_tok_s_p95": 11603.669565646976, "effective_prefill_tok_s_p99": 11603.669565646976, "effective_prefill_tok_s_samples": 12, "effective_prefill_tok_s_stddev": 253.17074313422327, "generation_max_ms": 1673.1046000004426, "generation_mean_ci95_high_ms": 1431.4683841335566, "generation_mean_ci95_low_ms": 1100.8971325328268, "generation_mean_ms": 1266.1827583331917, "generation_min_ms": 769.4195000003674, "generation_p25_ms": 1046.7771999992692, "generation_p50_ms": 1308.288399999583, "generation_p75_ms": 1465.8714999986842, "generation_p90_ms": 1639.6129999993718, "generation_p95_ms": 1673.1046000004426, "generation_p99_ms": 1673.1046000004426, "generation_samples": 12, "generation_stddev_ms": 292.12561392556034, "mean_inter_token_latency_ms_max": 4.482787461769676, "mean_inter_token_latency_ms_mean": 4.078350133860534, "mean_inter_token_latency_ms_mean_ci95_high": 4.230920557415158, "mean_inter_token_latency_ms_mean_ci95_low": 3.9257797103059096, "mean_inter_token_latency_ms_min": 3.66984506437857, "mean_inter_token_latency_ms_p25": 3.8331144638423416, "mean_inter_token_latency_ms_p50": 3.9791617543909394, "mean_inter_token_latency_ms_p75": 4.255191869915729, "mean_inter_token_latency_ms_p90": 4.4451499999996065, "mean_inter_token_latency_ms_p95": 4.482787461769676, "mean_inter_token_latency_ms_p99": 4.482787461769676, "mean_inter_token_latency_ms_samples": 12, "mean_inter_token_latency_ms_stddev": 0.26965278094786954, "mean_time_per_output_token_max_ms": 4.482787461769676, "mean_time_per_output_token_mean_ci95_high_ms": 4.230920557415158, "mean_time_per_output_token_mean_ci95_low_ms": 3.9257797103059096, "mean_time_per_output_token_mean_ms": 4.078350133860534, "mean_time_per_output_token_min_ms": 3.66984506437857, "mean_time_per_output_token_p25_ms": 3.8331144638423416, "mean_time_per_output_token_p50_ms": 3.9791617543909394, "mean_time_per_output_token_p75_ms": 4.255191869915729, "mean_time_per_output_token_p90_ms": 4.4451499999996065, "mean_time_per_output_token_p95_ms": 4.482787461769676, "mean_time_per_output_token_p99_ms": 4.482787461769676, "mean_time_per_output_token_samples": 12, "mean_time_per_output_token_stddev_ms": 0.26965278094786954, "output_token_sources": { "usage": 12 }, "output_tokens": 3739, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 12, "prompt_tokens": 269768, "prompt_tokens_max": 22495.0, "prompt_tokens_mean": 22480.666666666668, "prompt_tokens_mean_ci95_high": 22485.915558805667, "prompt_tokens_mean_ci95_low": 22475.41777452767, "prompt_tokens_min": 22467.0, "prompt_tokens_p25": 22472.0, "prompt_tokens_p50": 22479.0, "prompt_tokens_p75": 22487.0, "prompt_tokens_p90": 22493.0, "prompt_tokens_p95": 22495.0, "prompt_tokens_p99": 22495.0, "prompt_tokens_samples": 12, "prompt_tokens_stddev": 9.276885579794873, "reasoning_chunks": 792, "request_canary_passed": 12, "request_canary_samples": 12, "throughput_performance_eligible": true, "throughput_tok_s": 96.44826342193059, "time_to_first_output_max_ms": 2106.84990000118, "time_to_first_output_mean_ci95_high_ms": 1989.7873062746119, "time_to_first_output_mean_ci95_low_ms": 1937.907177059041, "time_to_first_output_mean_ms": 1963.8472416668264, "time_to_first_output_min_ms": 1937.232000000222, "time_to_first_output_p25_ms": 1943.6514999997598, "time_to_first_output_p50_ms": 1949.6307999997953, "time_to_first_output_p75_ms": 1962.3374000002514, "time_to_first_output_p90_ms": 1964.2872999993415, "time_to_first_output_p95_ms": 2106.84990000118, "time_to_first_output_p99_ms": 2106.84990000118, "time_to_first_output_samples": 12, "time_to_first_output_stddev_ms": 45.84643862479942, "tpot_max_ms": 4.482787461769676, "tpot_mean_ci95_high_ms": 4.230920557415158, "tpot_mean_ci95_low_ms": 3.9257797103059096, "tpot_mean_ms": 4.078350133860534, "tpot_min_ms": 3.66984506437857, "tpot_p25_ms": 3.8331144638423416, "tpot_p50_ms": 3.9791617543909394, "tpot_p75_ms": 4.255191869915729, "tpot_p90_ms": 4.4451499999996065, "tpot_p95_ms": 4.482787461769676, "tpot_p99_ms": 4.482787461769676, "tpot_samples": 12, "tpot_stddev_ms": 0.26965278094786954, "ttft_max_ms": 3466.7708999986644, "ttft_mean_ci95_high_ms": 3273.925431832019, "ttft_mean_ci95_low_ms": 2932.6291181679453, "ttft_mean_ms": 3103.277274999982, "ttft_min_ms": 2577.351500000077, "ttft_p25_ms": 2918.892899999264, "ttft_p50_ms": 3116.5836000000127, "ttft_p75_ms": 3351.5170000009675, "ttft_p90_ms": 3459.469399998852, "ttft_p95_ms": 3466.7708999986644, "ttft_p99_ms": 3466.7708999986644, "ttft_samples": 12, "ttft_stddev_ms": 301.6033447459894 }, "model": "qwen38-swift15-ninfer-dflash2-c4-262k", "performance_eligible": true, "prompt_cache_mode": "unique", "request_canaries": true, "request_timings": [ { "content_chunks": 8, "controlled_output": null, "decode_tok_s": 223.07548785844702, "e2e_ms": 3572.721399999864, "effective_prefill_tok_s": 10663.78767656273, "generation_ms": 1465.8714999986842, "mean_inter_token_latency_ms": 4.482787461769676, "mean_time_per_output_token_ms": 4.482787461769676, "output_token_source": "usage", "output_tokens": 328, "planned_context_tokens": 24576, "prompt_tokens": 22467, "reasoning_chunks": 78, "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": 2106.84990000118, "tpot_ms": 4.482787461769676, "ttft_ms": 3450.3436000013608, "visible_content_truncated": false, "visible_generation_ms": 122.37779999850318 }, { "content_chunks": 10, "controlled_output": null, "decode_tok_s": 229.51344464649634, "e2e_ms": 3616.7561000002024, "effective_prefill_tok_s": 11560.714459357954, "generation_ms": 1673.1046000004426, "mean_inter_token_latency_ms": 4.357043229167819, "mean_time_per_output_token_ms": 4.357043229167819, "output_token_source": "usage", "output_tokens": 385, "planned_context_tokens": 24576, "prompt_tokens": 22470, "reasoning_chunks": 87, "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": 1943.6514999997598, "tpot_ms": 4.357043229167819, "ttft_ms": 3459.469399998852, "visible_content_truncated": false, "visible_generation_ms": 157.28670000135025 }, { "content_chunks": 9, "controlled_output": null, "decode_tok_s": 252.1381379077439, "e2e_ms": 2717.003800000384, "effective_prefill_tok_s": 11538.396566454046, "generation_ms": 769.4195000003674, "mean_inter_token_latency_ms": 3.9660798969091102, "mean_time_per_output_token_ms": 3.9660798969091102, "output_token_source": "usage", "output_tokens": 195, "planned_context_tokens": 24576, "prompt_tokens": 22472, "reasoning_chunks": 37, "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": 1947.5843000000168, "tpot_ms": 3.9660798969091102, "ttft_ms": 2577.351500000077, "visible_content_truncated": false, "visible_generation_ms": 139.65230000030715 }, { "content_chunks": 9, "controlled_output": null, "decode_tok_s": 262.1746092070444, "e2e_ms": 3256.4419000009366, "effective_prefill_tok_s": 11535.538652362036, "generation_ms": 1308.288399999583, "mean_inter_token_latency_ms": 3.8142518950425157, "mean_time_per_output_token_ms": 3.8142518950425157, "output_token_source": "usage", "output_tokens": 344, "planned_context_tokens": 24576, "prompt_tokens": 22473, "reasoning_chunks": 67, "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": 1948.1535000013537, "tpot_ms": 3.8142518950425157, "ttft_ms": 3116.5836000000127, "visible_content_truncated": false, "visible_generation_ms": 139.85830000092392 }, { "content_chunks": 8, "controlled_output": null, "decode_tok_s": 258.5976080942042, "e2e_ms": 3589.485299999069, "effective_prefill_tok_s": 11526.908710895319, "generation_ms": 1639.6129999993718, "mean_inter_token_latency_ms": 3.8670117924513487, "mean_time_per_output_token_ms": 3.8670117924513487, "output_token_source": "usage", "output_tokens": 425, "planned_context_tokens": 24576, "prompt_tokens": 22476, "reasoning_chunks": 87, "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": 1949.8722999996971, "tpot_ms": 3.8670117924513487, "ttft_ms": 3466.7708999986644, "visible_content_truncated": false, "visible_generation_ms": 122.71440000040457 }, { "content_chunks": 8, "controlled_output": null, "decode_tok_s": 260.88446077803604, "e2e_ms": 3474.310900001001, "effective_prefill_tok_s": 11603.669565646976, "generation_ms": 1537.078900000779, "mean_inter_token_latency_ms": 3.8331144638423416, "mean_time_per_output_token_ms": 3.8331144638423416, "output_token_source": "usage", "output_tokens": 402, "planned_context_tokens": 24576, "prompt_tokens": 22479, "reasoning_chunks": 82, "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": 1937.232000000222, "tpot_ms": 3.8331144638423416, "ttft_ms": 3351.5170000009675, "visible_content_truncated": false, "visible_generation_ms": 122.79390000003332 }, { "content_chunks": 9, "controlled_output": null, "decode_tok_s": 272.491067731039, "e2e_ms": 2804.704700000002, "effective_prefill_tok_s": 11531.41405029217, "generation_ms": 855.0739000002068, "mean_inter_token_latency_ms": 3.66984506437857, "mean_time_per_output_token_ms": 3.66984506437857, "output_token_source": "usage", "output_tokens": 234, "planned_context_tokens": 24576, "prompt_tokens": 22482, "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": 1949.6307999997953, "tpot_ms": 3.66984506437857, "ttft_ms": 2665.2228999992076, "visible_content_truncated": false, "visible_generation_ms": 139.48180000079446 }, { "content_chunks": 8, "controlled_output": null, "decode_tok_s": 251.3092107644321, "e2e_ms": 3098.3484000007593, "effective_prefill_tok_s": 11446.899850142867, "generation_ms": 1134.0611000014178, "mean_inter_token_latency_ms": 3.9791617543909394, "mean_time_per_output_token_ms": 3.9791617543909394, "output_token_source": "usage", "output_tokens": 286, "planned_context_tokens": 24576, "prompt_tokens": 22485, "reasoning_chunks": 59, "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": 1964.2872999993415, "tpot_ms": 3.9791617543909394, "ttft_ms": 2975.922499999797, "visible_content_truncated": false, "visible_generation_ms": 122.42590000096243 }, { "content_chunks": 8, "controlled_output": null, "decode_tok_s": 248.17709082516205, "e2e_ms": 3292.0010999987426, "effective_prefill_tok_s": 11530.509959069235, "generation_ms": 1341.783799998666, "mean_inter_token_latency_ms": 4.0293807807767745, "mean_time_per_output_token_ms": 4.0293807807767745, "output_token_source": "usage", "output_tokens": 334, "planned_context_tokens": 24576, "prompt_tokens": 22487, "reasoning_chunks": 70, "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": 1950.2173000000766, "tpot_ms": 4.0293807807767745, "ttft_ms": 3169.5646999996825, "visible_content_truncated": false, "visible_generation_ms": 122.43639999906009 }, { "content_chunks": 8, "controlled_output": null, "decode_tok_s": 224.96428691947145, "e2e_ms": 3286.992100000134, "effective_prefill_tok_s": 11460.31258436858, "generation_ms": 1324.6546999998827, "mean_inter_token_latency_ms": 4.4451499999996065, "mean_time_per_output_token_ms": 4.4451499999996065, "output_token_source": "usage", "output_tokens": 299, "planned_context_tokens": 24576, "prompt_tokens": 22489, "reasoning_chunks": 70, "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": 1962.3374000002514, "tpot_ms": 4.4451499999996065, "ttft_ms": 3164.6536000007472, "visible_content_truncated": false, "visible_generation_ms": 122.33849999938684 }, { "content_chunks": 8, "controlled_output": null, "decode_tok_s": 235.78324873820637, "e2e_ms": 3040.9029999991617, "effective_prefill_tok_s": 11579.786520694724, "generation_ms": 1098.4664999996312, "mean_inter_token_latency_ms": 4.241183397681974, "mean_time_per_output_token_ms": 4.241183397681974, "output_token_source": "usage", "output_tokens": 260, "planned_context_tokens": 24576, "prompt_tokens": 22493, "reasoning_chunks": 57, "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": 1942.4364999995305, "tpot_ms": 4.241183397681974, "ttft_ms": 2918.892899999264, "visible_content_truncated": false, "visible_generation_ms": 122.01009999989765 }, { "content_chunks": 6, "controlled_output": null, "decode_tok_s": 235.0070291941511, "e2e_ms": 3010.691299999962, "effective_prefill_tok_s": 11454.166961779063, "generation_ms": 1046.7771999992692, "mean_inter_token_latency_ms": 4.255191869915729, "mean_time_per_output_token_ms": 4.255191869915729, "output_token_source": "usage", "output_tokens": 247, "planned_context_tokens": 24576, "prompt_tokens": 22495, "reasoning_chunks": 56, "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": 1963.914100000693, "tpot_ms": 4.255191869915729, "ttft_ms": 2923.0347000011534, "visible_content_truncated": false, "visible_generation_ms": 87.65659999880882 } ], "requests": 12, "response_words": 0, "responses_received": 12, "run_id": "benchmark-20261005T092323Z", "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": "2026-10-05T09:23:23Z", "started_at_unix_ns": 1791192203670804700, "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": 38766.89810000062 }