{ "base_url": "[operator identity redacted]", "clock_domain_id": null, "completed": 12, "concurrency": 1, "configuration_fingerprint": null, "context_distribution": { "8192": 12 }, "context_seed": 17, "context_tokens": 8192, "controlled_output_policy": "observe", "engine": "ninfer", "failed": 0, "failures": [], "finished_at": "2026-10-05T09:22:06Z", "finished_at_unix_ns": 1791192126744518100, "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": 396.0, "completion_tokens_mean": 311.25, "completion_tokens_mean_ci95_high": 355.0112257113255, "completion_tokens_mean_ci95_low": 267.4887742886745, "completion_tokens_min": 186.0, "completion_tokens_p25": 188.0, "completion_tokens_p50": 344.0, "completion_tokens_p75": 357.0, "completion_tokens_p90": 370.0, "completion_tokens_p95": 396.0, "completion_tokens_p99": 396.0, "completion_tokens_samples": 12, "completion_tokens_stddev": 77.34353707500536, "content_chunks": 97, "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": 288.2265367486065, "decode_tok_s_mean": 261.8792507061624, "decode_tok_s_mean_ci95_high": 272.4822628192165, "decode_tok_s_mean_ci95_low": 251.2762385931083, "decode_tok_s_min": 214.25633883145548, "decode_tok_s_p25": 248.54777565168436, "decode_tok_s_p50": 266.6946945809353, "decode_tok_s_p75": 269.5834081569028, "decode_tok_s_p90": 277.09834889500286, "decode_tok_s_p95": 288.2265367486065, "decode_tok_s_p99": 288.2265367486065, "decode_tok_s_samples": 12, "decode_tok_s_stddev": 18.739750707222367, "e2e_max_ms": 2421.978399999716, "e2e_mean_ci95_high_ms": 1969.905781649099, "e2e_mean_ci95_low_ms": 1618.173185017421, "e2e_mean_ms": 1794.03948333326, "e2e_min_ms": 1322.1831999999267, "e2e_p25_ms": 1442.1449000001303, "e2e_p50_ms": 1837.6459000010072, "e2e_p75_ms": 1901.1549000006198, "e2e_p90_ms": 2039.2243999995117, "e2e_p95_ms": 2421.978399999716, "e2e_p99_ms": 2421.978399999716, "e2e_samples": 12, "e2e_stddev_ms": 310.8258816552022, "effective_prefill_tok_s_max": 13125.663081722823, "effective_prefill_tok_s_mean": 12399.844886082561, "effective_prefill_tok_s_mean_ci95_high": 13433.969219281873, "effective_prefill_tok_s_mean_ci95_low": 11365.72055288325, "effective_prefill_tok_s_min": 6618.810686746546, "effective_prefill_tok_s_p25": 12780.503634537177, "effective_prefill_tok_s_p50": 12837.030454651413, "effective_prefill_tok_s_p75": 13089.180574298363, "effective_prefill_tok_s_p90": 13110.57022501909, "effective_prefill_tok_s_p95": 13125.663081722823, "effective_prefill_tok_s_p99": 13125.663081722823, "effective_prefill_tok_s_samples": 12, "effective_prefill_tok_s_stddev": 1827.7100882086686, "generation_max_ms": 1473.116299999674, "generation_mean_ci95_high_ms": 1313.908020849218, "generation_mean_ci95_low_ms": 1034.2023624845149, "generation_mean_ms": 1174.0551916668664, "generation_min_ms": 744.3237000006775, "generation_p25_ms": 863.451700000951, "generation_p50_ms": 1270.307100001446, "generation_p75_ms": 1316.2650000012945, "generation_p90_ms": 1368.778599999132, "generation_p95_ms": 1473.116299999674, "generation_p99_ms": 1473.116299999674, "generation_samples": 12, "generation_stddev_ms": 247.1757201286575, "mean_inter_token_latency_ms_max": 4.6673064864916265, "mean_inter_token_latency_ms_mean": 3.8385619739724883, "mean_inter_token_latency_ms_mean_ci95_high": 4.011960373381864, "mean_inter_token_latency_ms_mean_ci95_low": 3.665163574563112, "mean_inter_token_latency_ms_min": 3.4694931677030416, "mean_inter_token_latency_ms_p25": 3.697373595509254, "mean_inter_token_latency_ms_p50": 3.746557725947615, "mean_inter_token_latency_ms_p75": 3.7950043731750736, "mean_inter_token_latency_ms_p90": 4.071846524058787, "mean_inter_token_latency_ms_p95": 4.6673064864916265, "mean_inter_token_latency_ms_p99": 4.6673064864916265, "mean_inter_token_latency_ms_samples": 12, "mean_inter_token_latency_ms_stddev": 0.3064641201307766, "mean_time_per_output_token_max_ms": 4.6673064864916265, "mean_time_per_output_token_mean_ci95_high_ms": 4.011960373381864, "mean_time_per_output_token_mean_ci95_low_ms": 3.665163574563112, "mean_time_per_output_token_mean_ms": 3.8385619739724883, "mean_time_per_output_token_min_ms": 3.4694931677030416, "mean_time_per_output_token_p25_ms": 3.697373595509254, "mean_time_per_output_token_p50_ms": 3.746557725947615, "mean_time_per_output_token_p75_ms": 3.7950043731750736, "mean_time_per_output_token_p90_ms": 4.071846524058787, "mean_time_per_output_token_p95_ms": 4.6673064864916265, "mean_time_per_output_token_p99_ms": 4.6673064864916265, "mean_time_per_output_token_samples": 12, "mean_time_per_output_token_stddev_ms": 0.3064641201307766, "output_token_sources": { "usage": 12 }, "output_tokens": 3735, "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": 756, "request_canary_passed": 12, "request_canary_samples": 12, "throughput_performance_eligible": true, "throughput_tok_s": 173.46026705189746, "time_to_first_output_max_ms": 1120.291900000666, "time_to_first_output_mean_ci95_high_ms": 709.2158293822566, "time_to_first_output_mean_ci95_low_ms": 530.7527539505303, "time_to_first_output_mean_ms": 619.9842916663935, "time_to_first_output_min_ms": 566.1080999998376, "time_to_first_output_p25_ms": 567.2131999999692, "time_to_first_output_p50_ms": 575.2148999999918, "time_to_first_output_p75_ms": 580.2803000005952, "time_to_first_output_p90_ms": 587.8564999984519, "time_to_first_output_p95_ms": 1120.291900000666, "time_to_first_output_p99_ms": 1120.291900000666, "time_to_first_output_samples": 12, "time_to_first_output_stddev_ms": 157.707711185075, "tpot_max_ms": 4.6673064864916265, "tpot_mean_ci95_high_ms": 4.011960373381864, "tpot_mean_ci95_low_ms": 3.665163574563112, "tpot_mean_ms": 3.8385619739724883, "tpot_min_ms": 3.4694931677030416, "tpot_p25_ms": 3.697373595509254, "tpot_p50_ms": 3.746557725947615, "tpot_p75_ms": 3.7950043731750736, "tpot_p90_ms": 4.071846524058787, "tpot_p95_ms": 4.6673064864916265, "tpot_p99_ms": 4.6673064864916265, "tpot_samples": 12, "tpot_stddev_ms": 0.3064641201307766, "ttft_max_ms": 2320.046700000603, "ttft_mean_ci95_high_ms": 1861.0954243211013, "ttft_mean_ci95_low_ms": 1486.481792345549, "ttft_mean_ms": 1673.788608333325, "ttft_min_ms": 1186.7464000006294, "ttft_p25_ms": 1239.986599999611, "ttft_p50_ms": 1733.2922000005055, "ttft_p75_ms": 1815.9362000005785, "ttft_p90_ms": 1936.7531999996572, "ttft_p95_ms": 2320.046700000603, "ttft_p99_ms": 2320.046700000603, "ttft_samples": 12, "ttft_stddev_ms": 331.04583866814596 }, "model": "qwen38-swift15-ninfer-dflash2-c4-262k", "performance_eligible": true, "prompt_cache_mode": "unique", "request_canaries": true, "request_timings": [ { "content_chunks": 7, "controlled_output": null, "decode_tok_s": 263.5043076810355, "e2e_ms": 2421.978399999716, "effective_prefill_tok_s": 6618.810686746546, "generation_ms": 1301.6864999990503, "mean_inter_token_latency_ms": 3.7950043731750736, "mean_time_per_output_token_ms": 3.7950043731750736, "output_token_source": "usage", "output_tokens": 344, "planned_context_tokens": 8192, "prompt_tokens": 7415, "reasoning_chunks": 72, "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": 1120.291900000666, "tpot_ms": 3.7950043731750736, "ttft_ms": 2320.046700000603, "visible_content_truncated": false, "visible_generation_ms": 101.93169999911333 }, { "content_chunks": 15, "controlled_output": null, "decode_tok_s": 214.25633883145548, "e2e_ms": 1442.1449000001303, "effective_prefill_tok_s": 12816.808630221538, "generation_ms": 863.451700000951, "mean_inter_token_latency_ms": 4.6673064864916265, "mean_time_per_output_token_ms": 4.6673064864916265, "output_token_source": "usage", "output_tokens": 186, "planned_context_tokens": 8192, "prompt_tokens": 7417, "reasoning_chunks": 38, "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": 578.6931999991793, "tpot_ms": 4.6673064864916265, "ttft_ms": 1205.366199999844, "visible_content_truncated": false, "visible_generation_ms": 236.77870000028634 }, { "content_chunks": 9, "controlled_output": null, "decode_tok_s": 248.54777565168436, "e2e_ms": 1322.1831999999267, "effective_prefill_tok_s": 12837.030454651413, "generation_ms": 744.3237000006775, "mean_inter_token_latency_ms": 4.023371351355014, "mean_time_per_output_token_ms": 4.023371351355014, "output_token_source": "usage", "output_tokens": 186, "planned_context_tokens": 8192, "prompt_tokens": 7418, "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": 577.8594999992492, "tpot_ms": 4.023371351355014, "ttft_ms": 1186.7464000006294, "visible_content_truncated": false, "visible_generation_ms": 135.43679999929736 }, { "content_chunks": 8, "controlled_output": null, "decode_tok_s": 288.2265367486065, "e2e_ms": 1705.0332999988314, "effective_prefill_tok_s": 12622.128019371292, "generation_ms": 1117.1768000003794, "mean_inter_token_latency_ms": 3.4694931677030416, "mean_time_per_output_token_ms": 3.4694931677030416, "output_token_source": "usage", "output_tokens": 323, "planned_context_tokens": 8192, "prompt_tokens": 7420, "reasoning_chunks": 60, "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": 587.8564999984519, "tpot_ms": 3.4694931677030416, "ttft_ms": 1586.1402999998973, "visible_content_truncated": false, "visible_generation_ms": 118.8929999989341 }, { "content_chunks": 8, "controlled_output": null, "decode_tok_s": 263.53779687025525, "e2e_ms": 1832.764099999622, "effective_prefill_tok_s": 12780.503634537177, "generation_ms": 1252.192300000388, "mean_inter_token_latency_ms": 3.7945221212132973, "mean_time_per_output_token_ms": 3.7945221212132973, "output_token_source": "usage", "output_tokens": 331, "planned_context_tokens": 8192, "prompt_tokens": 7420, "reasoning_chunks": 68, "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": 580.5717999992339, "tpot_ms": 3.7945221212132973, "ttft_ms": 1714.1309999988152, "visible_content_truncated": false, "visible_generation_ms": 118.63310000080673 }, { "content_chunks": 7, "controlled_output": null, "decode_tok_s": 268.1390464555225, "e2e_ms": 2039.2243999995117, "effective_prefill_tok_s": 13110.57022501909, "generation_ms": 1473.116299999674, "mean_inter_token_latency_ms": 3.7294083544295544, "mean_time_per_output_token_ms": 3.7294083544295544, "output_token_source": "usage", "output_tokens": 396, "planned_context_tokens": 8192, "prompt_tokens": 7422, "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": 566.1080999998376, "tpot_ms": 3.7294083544295544, "ttft_ms": 1936.7531999996572, "visible_content_truncated": false, "visible_generation_ms": 102.47119999985443 }, { "content_chunks": 7, "controlled_output": null, "decode_tok_s": 245.5888241591206, "e2e_ms": 1341.7155999995884, "effective_prefill_tok_s": 12793.817057019489, "generation_ms": 761.4352999989933, "mean_inter_token_latency_ms": 4.071846524058787, "mean_time_per_output_token_ms": 4.071846524058787, "output_token_source": "usage", "output_tokens": 188, "planned_context_tokens": 8192, "prompt_tokens": 7424, "reasoning_chunks": 40, "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": 580.2803000005952, "tpot_ms": 4.071846524058787, "ttft_ms": 1239.986599999611, "visible_content_truncated": false, "visible_generation_ms": 101.72899999997753 }, { "content_chunks": 7, "controlled_output": null, "decode_tok_s": 277.09834889500286, "e2e_ms": 1837.6459000010072, "effective_prefill_tok_s": 13089.180574298363, "generation_ms": 1270.307100001446, "mean_inter_token_latency_ms": 3.608826988640472, "mean_time_per_output_token_ms": 3.608826988640472, "output_token_source": "usage", "output_tokens": 353, "planned_context_tokens": 8192, "prompt_tokens": 7426, "reasoning_chunks": 69, "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": 567.3387999995612, "tpot_ms": 3.608826988640472, "ttft_ms": 1735.5012000007264, "visible_content_truncated": false, "visible_generation_ms": 102.14470000028086 }, { "content_chunks": 8, "controlled_output": null, "decode_tok_s": 270.4622549407983, "e2e_ms": 1891.4799000012863, "effective_prefill_tok_s": 12911.696133045418, "generation_ms": 1316.2650000012945, "mean_inter_token_latency_ms": 3.697373595509254, "mean_time_per_output_token_ms": 3.697373595509254, "output_token_source": "usage", "output_tokens": 357, "planned_context_tokens": 8192, "prompt_tokens": 7427, "reasoning_chunks": 71, "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": 575.2148999999918, "tpot_ms": 3.697373595509254, "ttft_ms": 1772.751700000299, "visible_content_truncated": false, "visible_generation_ms": 118.72820000098727 }, { "content_chunks": 7, "controlled_output": null, "decode_tok_s": 269.5834081569028, "e2e_ms": 1940.8666999988782, "effective_prefill_tok_s": 12987.510140489368, "generation_ms": 1368.778599999132, "mean_inter_token_latency_ms": 3.7094271002686505, "mean_time_per_output_token_ms": 3.7094271002686505, "output_token_source": "usage", "output_tokens": 370, "planned_context_tokens": 8192, "prompt_tokens": 7430, "reasoning_chunks": 76, "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": 572.0880999997462, "tpot_ms": 3.7094271002686505, "ttft_ms": 1838.811599998735, "visible_content_truncated": false, "visible_generation_ms": 102.05510000014328 }, { "content_chunks": 8, "controlled_output": null, "decode_tok_s": 266.9116755026297, "e2e_ms": 1852.2825000000012, "effective_prefill_tok_s": 13104.419995868226, "generation_ms": 1285.069300000032, "mean_inter_token_latency_ms": 3.746557725947615, "mean_time_per_output_token_ms": 3.746557725947615, "output_token_source": "usage", "output_tokens": 344, "planned_context_tokens": 8192, "prompt_tokens": 7433, "reasoning_chunks": 69, "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": 567.2131999999692, "tpot_ms": 3.746557725947615, "ttft_ms": 1733.2922000005055, "visible_content_truncated": false, "visible_generation_ms": 118.99029999949562 }, { "content_chunks": 6, "controlled_output": null, "decode_tok_s": 266.6946945809353, "e2e_ms": 1901.1549000006198, "effective_prefill_tok_s": 13125.663081722823, "generation_ms": 1334.8597000003792, "mean_inter_token_latency_ms": 3.7496058988774696, "mean_time_per_output_token_ms": 3.7496058988774696, "output_token_source": "usage", "output_tokens": 357, "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": 566.2952000002406, "tpot_ms": 3.7496058988774696, "ttft_ms": 1815.9362000005785, "visible_content_truncated": false, "visible_generation_ms": 85.21870000004128 } ], "requests": 12, "response_words": 0, "responses_received": 12, "run_id": "benchmark-20261005T092145Z", "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": "2026-10-05T09:21:45Z", "started_at_unix_ns": 1791192105212205900, "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": 21532.30860000076 }