{ "base_url": "http://127.0.0.1:30001/v1", "clock_domain_id": null, "completed": 32, "concurrency": 4, "configuration_fingerprint": null, "context_distribution": { "32768": 32 }, "context_seed": 1912, "context_tokens": 32768, "controlled_output_policy": "strict", "engine": "vllm-v84", "failed": 0, "failures": [], "finished_at": "2026-09-19T11:28:10Z", "finished_at_unix_ns": 1789817290740653633, "gpu": "2x-rtx-pro-6000-max-q", "manifest": null, "max_context_tokens": 327680, "max_tokens": 4096, "measurement_protocol": "capacity-v4-reasoning", "metric_population": { "aggregate_throughput_eligible": true, "attempted_requests": 32, "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": 32, "e2e": 32, "effective_prefill_tok_s": 32, "tpot": 32, "ttft": 32 }, "performance_eligible_requests": 32, "successful_requests": 32 }, "metrics": { "completion_tokens_max": 186.0, "completion_tokens_mean": 146.59375, "completion_tokens_mean_ci95_high": 157.2824255507987, "completion_tokens_mean_ci95_low": 135.9050744492013, "completion_tokens_min": 90.0, "completion_tokens_p25": 122.0, "completion_tokens_p50": 124.0, "completion_tokens_p75": 181.0, "completion_tokens_p90": 185.0, "completion_tokens_p95": 185.0, "completion_tokens_p99": 186.0, "completion_tokens_samples": 32, "completion_tokens_stddev": 30.84912230152089, "content_chunks": 864, "content_chunks_s": null, "controlled_output_exact_adherence": 32, "controlled_output_exact_adherence_rate": 1.0, "controlled_output_samples": 32, "controlled_output_unobservable": 0, "decode_tok_s_max": 59.22180410778109, "decode_tok_s_mean": 9.343915207278132, "decode_tok_s_mean_ci95_high": 12.585992177014028, "decode_tok_s_mean_ci95_low": 6.101838237542235, "decode_tok_s_min": 4.581818885397002, "decode_tok_s_p25": 6.0903923560983895, "decode_tok_s_p50": 8.257159080865938, "decode_tok_s_p75": 8.741537591174163, "decode_tok_s_p90": 10.424317937312233, "decode_tok_s_p95": 15.131258819055347, "decode_tok_s_p99": 59.22180410778109, "decode_tok_s_samples": 32, "decode_tok_s_stddev": 9.35712085889382, "e2e_max_ms": 53373.11863002833, "e2e_mean_ci95_high_ms": 31375.02962985185, "e2e_mean_ci95_low_ms": 25404.358786059092, "e2e_mean_ms": 28389.694207955472, "e2e_min_ms": 8267.784975003451, "e2e_p25_ms": 25635.39082603529, "e2e_p50_ms": 27175.21648807451, "e2e_p75_ms": 33615.291971946135, "e2e_p90_ms": 36818.89876991045, "e2e_p95_ms": 44785.34463001415, "e2e_p99_ms": 53373.11863002833, "e2e_samples": 32, "e2e_stddev_ms": 8616.126207915642, "effective_prefill_tok_s_max": 4090.316524128506, "effective_prefill_tok_s_mean": 3422.9975392032406, "effective_prefill_tok_s_mean_ci95_high": 3779.64037419724, "effective_prefill_tok_s_mean_ci95_low": 3066.3547042092414, "effective_prefill_tok_s_min": 1064.0736037506952, "effective_prefill_tok_s_p25": 2492.464750585708, "effective_prefill_tok_s_p50": 4063.0416354677272, "effective_prefill_tok_s_p75": 4071.234954066806, "effective_prefill_tok_s_p90": 4077.091579420828, "effective_prefill_tok_s_p95": 4085.1600617399226, "effective_prefill_tok_s_p99": 4090.316524128506, "effective_prefill_tok_s_samples": 32, "effective_prefill_tok_s_stddev": 1029.324763615722, "generation_max_ms": 36356.805211049505, "generation_mean_ci95_high_ms": 22213.402440850416, "generation_mean_ci95_low_ms": 17232.525229324154, "generation_mean_ms": 19722.963835087285, "generation_min_ms": 2043.166394927539, "generation_p25_ms": 14051.248422940262, "generation_p50_ms": 19910.48383002635, "generation_p75_ms": 21102.436059038155, "generation_p90_ms": 28861.983885988593, "generation_p95_ms": 31270.132742007263, "generation_p99_ms": 36356.805211049505, "generation_samples": 32, "generation_stddev_ms": 7187.779699036249, "mean_inter_token_latency_ms_max": 218.2539347391408, "mean_inter_token_latency_ms_mean": 134.65891799139618, "mean_inter_token_latency_ms_mean_ci95_high": 148.81436249996386, "mean_inter_token_latency_ms_mean_ci95_low": 120.50347348282851, "mean_inter_token_latency_ms_min": 16.88567268535156, "mean_inter_token_latency_ms_p25": 114.25484069585092, "mean_inter_token_latency_ms_p50": 120.206442431429, "mean_inter_token_latency_ms_p75": 163.91707403288984, "mean_inter_token_latency_ms_p90": 174.0526563058846, "mean_inter_token_latency_ms_p95": 197.5913326687473, "mean_inter_token_latency_ms_p99": 218.2539347391408, "mean_inter_token_latency_ms_samples": 32, "mean_inter_token_latency_ms_stddev": 40.85473797027787, "mean_time_per_output_token_max_ms": 218.2539347391408, "mean_time_per_output_token_mean_ci95_high_ms": 148.81436249996386, "mean_time_per_output_token_mean_ci95_low_ms": 120.50347348282851, "mean_time_per_output_token_mean_ms": 134.65891799139618, "mean_time_per_output_token_min_ms": 16.88567268535156, "mean_time_per_output_token_p25_ms": 114.25484069585092, "mean_time_per_output_token_p50_ms": 120.206442431429, "mean_time_per_output_token_p75_ms": 163.91707403288984, "mean_time_per_output_token_p90_ms": 174.0526563058846, "mean_time_per_output_token_p95_ms": 197.5913326687473, "mean_time_per_output_token_p99_ms": 218.2539347391408, "mean_time_per_output_token_samples": 32, "mean_time_per_output_token_stddev_ms": 40.85473797027787, "output_token_sources": { "usage": 32 }, "output_tokens": 4691, "prefix_cache_hit_avg": null, "prompt_token_samples": 32, "prompt_tokens": 809193, "prompt_tokens_max": 25295.0, "prompt_tokens_mean": 25287.28125, "prompt_tokens_mean_ci95_high": 25288.50144926238, "prompt_tokens_mean_ci95_low": 25286.06105073762, "prompt_tokens_min": 25282.0, "prompt_tokens_p25": 25285.0, "prompt_tokens_p50": 25286.0, "prompt_tokens_p75": 25289.0, "prompt_tokens_p90": 25292.0, "prompt_tokens_p95": 25294.0, "prompt_tokens_p99": 25295.0, "prompt_tokens_samples": 32, "prompt_tokens_stddev": 3.521678256447554, "reasoning_chunks": 3427, "request_canary_passed": 32, "request_canary_samples": 32, "throughput_performance_eligible": true, "throughput_tok_s": 20.6038541796123, "time_to_first_output_max_ms": 23762.453941977583, "time_to_first_output_mean_ci95_high_ms": 10249.269739960913, "time_to_first_output_mean_ci95_low_ms": 7084.191005775462, "time_to_first_output_mean_ms": 8666.730372868187, "time_to_first_output_min_ms": 6181.673191022128, "time_to_first_output_p25_ms": 6210.783960064873, "time_to_first_output_p50_ms": 6224.618580075912, "time_to_first_output_p75_ms": 6978.7208759225905, "time_to_first_output_p90_ms": 15729.079336044379, "time_to_first_output_p95_ms": 18817.583614960313, "time_to_first_output_p99_ms": 23762.453941977583, "time_to_first_output_samples": 32, "time_to_first_output_stddev_ms": 4567.446195779318, "tpot_max_ms": 218.2539347391408, "tpot_mean_ci95_high_ms": 148.81436249996386, "tpot_mean_ci95_low_ms": 120.50347348282851, "tpot_mean_ms": 134.65891799139618, "tpot_min_ms": 16.88567268535156, "tpot_p25_ms": 114.25484069585092, "tpot_p50_ms": 120.206442431429, "tpot_p75_ms": 163.91707403288984, "tpot_p90_ms": 174.0526563058846, "tpot_p95_ms": 197.5913326687473, "tpot_p99_ms": 218.2539347391408, "tpot_samples": 32, "tpot_stddev_ms": 40.85473797027787, "ttft_max_ms": 52858.83679694962, "ttft_mean_ci95_high_ms": 27925.649677223697, "ttft_mean_ci95_low_ms": 22217.61757538247, "ttft_mean_ms": 25071.633626303083, "ttft_min_ms": 7904.5311390655115, "ttft_p25_ms": 19747.123728040606, "ttft_p50_ms": 25423.657369916327, "ttft_p75_ms": 27522.02979789581, "ttft_p90_ms": 33415.30487895943, "ttft_p95_ms": 40724.103079061024, "ttft_p99_ms": 52858.83679694962, "ttft_samples": 32, "ttft_stddev_ms": 8237.118788658036 }, "model": "glm53-flash-exl3-4bpw-v84-fp8-tp2-c4-327k-r10-vision8-apc-nospec", "performance_eligible": true, "prompt_cache_mode": "unique", "request_canaries": true, "request_timings": [ { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 9.859253164602677, "e2e_ms": 25635.39082603529, "effective_prefill_tok_s": 1921.2296825486355, "generation_ms": 12475.589980953373, "mean_inter_token_latency_ms": 101.42756082075913, "mean_time_per_output_token_ms": 101.42756082075913, "output_token_source": "usage", "output_tokens": 124, "planned_context_tokens": 32768, "prompt_tokens": 25283, "reasoning_chunks": 92, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00000", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 0, "time_to_first_output_ms": 13159.800845081918, "tpot_ms": 101.42756082075913, "ttft_ms": 25120.986972004175, "visible_content_truncated": false, "visible_generation_ms": 514.4038540311158 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 5.9161885088986885, "e2e_ms": 38247.569973929785, "effective_prefill_tok_s": 3623.6800360342895, "generation_ms": 31270.132742007263, "mean_inter_token_latency_ms": 169.0277445513906, "mean_time_per_output_token_ms": 169.0277445513906, "output_token_source": "usage", "output_tokens": 186, "planned_context_tokens": 32768, "prompt_tokens": 25284, "reasoning_chunks": 137, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00001", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 1, "time_to_first_output_ms": 6977.437231922522, "tpot_ms": 169.0277445513906, "ttft_ms": 33415.30487895943, "visible_content_truncated": false, "visible_generation_ms": 4832.265094970353 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 10.424317937312233, "e2e_ms": 31000.634895986877, "effective_prefill_tok_s": 1343.690056989994, "generation_ms": 12183.051281026565, "mean_inter_token_latency_ms": 95.92953764587845, "mean_time_per_output_token_ms": 95.92953764587845, "output_token_source": "usage", "output_tokens": 128, "planned_context_tokens": 32768, "prompt_tokens": 25285, "reasoning_chunks": 95, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00002", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 2, "time_to_first_output_ms": 18817.583614960313, "tpot_ms": 95.92953764587845, "ttft_ms": 25423.657369916327, "visible_content_truncated": false, "visible_generation_ms": 5576.97752607055 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 8.752364397951625, "e2e_ms": 44785.34463001415, "effective_prefill_tok_s": 1064.0736037506952, "generation_ms": 21022.89068803657, "mean_inter_token_latency_ms": 114.25484069585092, "mean_time_per_output_token_ms": 114.25484069585092, "output_token_source": "usage", "output_tokens": 185, "planned_context_tokens": 32768, "prompt_tokens": 25285, "reasoning_chunks": 135, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00003", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 3, "time_to_first_output_ms": 23762.453941977583, "tpot_ms": 114.25484069585092, "ttft_ms": 40724.103079061024, "visible_content_truncated": false, "visible_generation_ms": 4061.2415509531274 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 8.53869611271652, "e2e_ms": 27145.06123494357, "effective_prefill_tok_s": 4090.316524128506, "generation_ms": 20963.38804392144, "mean_inter_token_latency_ms": 117.11389968671196, "mean_time_per_output_token_ms": 117.11389968671196, "output_token_source": "usage", "output_tokens": 180, "planned_context_tokens": 32768, "prompt_tokens": 25285, "reasoning_chunks": 131, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00004", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 4, "time_to_first_output_ms": 6181.673191022128, "tpot_ms": 117.11389968671196, "ttft_ms": 26629.683968028985, "visible_content_truncated": false, "visible_generation_ms": 515.3772669145837 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 8.897554788687247, "e2e_ms": 27175.21648807451, "effective_prefill_tok_s": 3640.6365905209113, "generation_ms": 20230.27722502593, "mean_inter_token_latency_ms": 112.39042902792183, "mean_time_per_output_token_ms": 112.39042902792183, "output_token_source": "usage", "output_tokens": 181, "planned_context_tokens": 32768, "prompt_tokens": 25284, "reasoning_chunks": 132, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00005", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 5, "time_to_first_output_ms": 6944.939263048582, "tpot_ms": 112.39042902792183, "ttft_ms": 21568.250657990575, "visible_content_truncated": false, "visible_generation_ms": 5606.965830083936 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 8.719372468904744, "e2e_ms": 27304.004744975828, "effective_prefill_tok_s": 4076.871720751929, "generation_ms": 21102.436059038155, "mean_inter_token_latency_ms": 114.68715249477258, "mean_time_per_output_token_ms": 114.68715249477258, "output_token_source": "usage", "output_tokens": 185, "planned_context_tokens": 32768, "prompt_tokens": 25283, "reasoning_chunks": 136, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00006", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 6, "time_to_first_output_ms": 6201.568685937673, "tpot_ms": 114.68715249477258, "ttft_ms": 24454.34963691514, "visible_content_truncated": false, "visible_generation_ms": 2849.655108060688 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 8.319021674486738, "e2e_ms": 27123.695655027404, "effective_prefill_tok_s": 4072.634935382197, "generation_ms": 20915.920983068645, "mean_inter_token_latency_ms": 120.206442431429, "mean_time_per_output_token_ms": 120.206442431429, "output_token_source": "usage", "output_tokens": 175, "planned_context_tokens": 32768, "prompt_tokens": 25282, "reasoning_chunks": 126, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00007", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 7, "time_to_first_output_ms": 6207.774671958759, "tpot_ms": 120.206442431429, "ttft_ms": 20744.985989993438, "visible_content_truncated": false, "visible_generation_ms": 6378.709665033966 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 6.077312939666739, "e2e_ms": 26123.588906950317, "effective_prefill_tok_s": 4069.0603517732434, "generation_ms": 19910.115079022944, "mean_inter_token_latency_ms": 164.54640561175987, "mean_time_per_output_token_ms": 164.54640561175987, "output_token_source": "usage", "output_tokens": 122, "planned_context_tokens": 32768, "prompt_tokens": 25283, "reasoning_chunks": 90, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00008", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 8, "time_to_first_output_ms": 6213.473827927373, "tpot_ms": 164.54640561175987, "ttft_ms": 25607.96258994378, "visible_content_truncated": false, "visible_generation_ms": 515.6263170065358 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 6.717237004943977, "e2e_ms": 33924.323638901114, "effective_prefill_tok_s": 3623.0135077092395, "generation_ms": 26945.602762978524, "mean_inter_token_latency_ms": 148.87073349711892, "mean_time_per_output_token_ms": 148.87073349711892, "output_token_source": "usage", "output_tokens": 182, "planned_context_tokens": 32768, "prompt_tokens": 25284, "reasoning_chunks": 133, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00009", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 9, "time_to_first_output_ms": 6978.7208759225905, "tpot_ms": 148.87073349711892, "ttft_ms": 27522.02979789581, "visible_content_truncated": false, "visible_generation_ms": 6402.293841005303 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 8.611334477757417, "e2e_ms": 20262.032383005135, "effective_prefill_tok_s": 4071.1446674979647, "generation_ms": 14051.248422940262, "mean_inter_token_latency_ms": 116.12602002429968, "mean_time_per_output_token_ms": 116.12602002429968, "output_token_source": "usage", "output_tokens": 122, "planned_context_tokens": 32768, "prompt_tokens": 25285, "reasoning_chunks": 90, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00010", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 10, "time_to_first_output_ms": 6210.783960064873, "tpot_ms": 116.12602002429968, "ttft_ms": 19747.123728040606, "visible_content_truncated": false, "visible_generation_ms": 514.908654964529 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 8.586343639778551, "e2e_ms": 27295.952432090417, "effective_prefill_tok_s": 4067.7501262244587, "generation_ms": 21079.98556701932, "mean_inter_token_latency_ms": 116.46400865756532, "mean_time_per_output_token_ms": 116.46400865756532, "output_token_source": "usage", "output_tokens": 182, "planned_context_tokens": 32768, "prompt_tokens": 25285, "reasoning_chunks": 133, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00011", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 11, "time_to_first_output_ms": 6215.9668650710955, "tpot_ms": 116.46400865756532, "ttft_ms": 26783.136683050543, "visible_content_truncated": false, "visible_generation_ms": 512.8157490398735 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 8.588950754027412, "e2e_ms": 27303.369342000224, "effective_prefill_tok_s": 4058.889726985948, "generation_ms": 21073.586888960563, "mean_inter_token_latency_ms": 116.42865684508598, "mean_time_per_output_token_ms": 116.42865684508598, "output_token_source": "usage", "output_tokens": 182, "planned_context_tokens": 32768, "prompt_tokens": 25286, "reasoning_chunks": 133, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00012", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 12, "time_to_first_output_ms": 6229.782453039661, "tpot_ms": 116.42865684508598, "ttft_ms": 26787.0652040001, "visible_content_truncated": false, "visible_generation_ms": 516.3041380001232 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 8.257159080865938, "e2e_ms": 28035.354368039407, "effective_prefill_tok_s": 4054.6244646069276, "generation_ms": 21799.265127046965, "mean_inter_token_latency_ms": 121.10702848359425, "mean_time_per_output_token_ms": 121.10702848359425, "output_token_source": "usage", "output_tokens": 181, "planned_context_tokens": 32768, "prompt_tokens": 25285, "reasoning_chunks": 132, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00013", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 13, "time_to_first_output_ms": 6236.089240992442, "tpot_ms": 121.10702848359425, "ttft_ms": 26762.338475091383, "visible_content_truncated": false, "visible_generation_ms": 1273.015892948024 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 6.3507776565058816, "e2e_ms": 33615.291971946135, "effective_prefill_tok_s": 4067.187565799095, "generation_ms": 27398.21946399752, "mean_inter_token_latency_ms": 157.4610314022846, "mean_time_per_output_token_ms": 157.4610314022846, "output_token_source": "usage", "output_tokens": 175, "planned_context_tokens": 32768, "prompt_tokens": 25286, "reasoning_chunks": 126, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00014", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 14, "time_to_first_output_ms": 6217.072507948615, "tpot_ms": 157.4610314022846, "ttft_ms": 20760.304201976396, "visible_content_truncated": false, "visible_generation_ms": 12854.987769969739 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 11.762727206591538, "e2e_ms": 13788.515344960615, "effective_prefill_tok_s": 4063.807993581065, "generation_ms": 7566.272552008741, "mean_inter_token_latency_ms": 85.01429833717685, "mean_time_per_output_token_ms": 85.01429833717685, "output_token_source": "usage", "output_tokens": 90, "planned_context_tokens": 32768, "prompt_tokens": 25286, "reasoning_chunks": 59, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00015", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 15, "time_to_first_output_ms": 6222.242792951874, "tpot_ms": 85.01429833717685, "ttft_ms": 13274.368942016736, "visible_content_truncated": false, "visible_generation_ms": 514.1464029438794 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 5.196398476716738, "e2e_ms": 36818.89876991045, "effective_prefill_tok_s": 4064.922100143363, "generation_ms": 30598.115350934677, "mean_inter_token_latency_ms": 192.44097704990364, "mean_time_per_output_token_ms": 192.44097704990364, "output_token_source": "usage", "output_tokens": 160, "planned_context_tokens": 32768, "prompt_tokens": 25287, "reasoning_chunks": 111, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00016", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 16, "time_to_first_output_ms": 6220.7834189757705, "tpot_ms": 192.44097704990364, "ttft_ms": 36316.23864895664, "visible_content_truncated": false, "visible_generation_ms": 502.66012095380574 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 5.745387753477169, "e2e_ms": 32984.758117003366, "effective_prefill_tok_s": 2120.6122065410655, "generation_ms": 21060.371413012035, "mean_inter_token_latency_ms": 174.0526563058846, "mean_time_per_output_token_ms": 174.0526563058846, "output_token_source": "usage", "output_tokens": 122, "planned_context_tokens": 32768, "prompt_tokens": 25287, "reasoning_chunks": 90, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00017", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 17, "time_to_first_output_ms": 11924.386703991331, "tpot_ms": 174.0526563058846, "ttft_ms": 29718.243035953492, "visible_content_truncated": false, "visible_generation_ms": 3266.5150810498744 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 5.060950733484113, "e2e_ms": 53373.11863002833, "effective_prefill_tok_s": 1486.1033278686266, "generation_ms": 36356.805211049505, "mean_inter_token_latency_ms": 197.5913326687473, "mean_time_per_output_token_ms": 197.5913326687473, "output_token_source": "usage", "output_tokens": 185, "planned_context_tokens": 32768, "prompt_tokens": 25288, "reasoning_chunks": 136, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00018", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 18, "time_to_first_output_ms": 17016.313418978825, "tpot_ms": 197.5913326687473, "ttft_ms": 52858.83679694962, "visible_content_truncated": false, "visible_generation_ms": 514.2818330787122 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 6.460550124000102, "e2e_ms": 34767.70571293309, "effective_prefill_tok_s": 1607.6592570839744, "generation_ms": 19038.626376888715, "mean_inter_token_latency_ms": 154.78558029990825, "mean_time_per_output_token_ms": 154.78558029990825, "output_token_source": "usage", "output_tokens": 124, "planned_context_tokens": 32768, "prompt_tokens": 25287, "reasoning_chunks": 92, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00019", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 19, "time_to_first_output_ms": 15729.079336044379, "tpot_ms": 154.78558029990825, "ttft_ms": 21069.495638017543, "visible_content_truncated": false, "visible_generation_ms": 13698.21007491555 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 4.581818885397002, "e2e_ms": 30878.024737001397, "effective_prefill_tok_s": 3801.656034376635, "generation_ms": 24226.186756044626, "mean_inter_token_latency_ms": 218.2539347391408, "mean_time_per_output_token_ms": 218.2539347391408, "output_token_source": "usage", "output_tokens": 112, "planned_context_tokens": 32768, "prompt_tokens": 25288, "reasoning_chunks": 80, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00020", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 20, "time_to_first_output_ms": 6651.8379809567705, "tpot_ms": 218.2539347391408, "ttft_ms": 29588.961918954737, "visible_content_truncated": false, "visible_generation_ms": 1289.0628180466592 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 6.0903923560983895, "e2e_ms": 34242.105957935564, "effective_prefill_tok_s": 1759.2655219637682, "generation_ms": 19867.357129929587, "mean_inter_token_latency_ms": 164.19303413164948, "mean_time_per_output_token_ms": 164.19303413164948, "output_token_source": "usage", "output_tokens": 122, "planned_context_tokens": 32768, "prompt_tokens": 25289, "reasoning_chunks": 90, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00021", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 21, "time_to_first_output_ms": 14374.748828005977, "tpot_ms": 164.19303413164948, "ttft_ms": 27866.80489196442, "visible_content_truncated": false, "visible_generation_ms": 6375.3010659711435 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 6.088501826445347, "e2e_ms": 26063.981221988797, "effective_prefill_tok_s": 4085.1600617399226, "generation_ms": 19873.52610693779, "mean_inter_token_latency_ms": 164.24401741270898, "mean_time_per_output_token_ms": 164.24401741270898, "output_token_source": "usage", "output_tokens": 122, "planned_context_tokens": 32768, "prompt_tokens": 25289, "reasoning_chunks": 90, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00022", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 22, "time_to_first_output_ms": 6190.455115051009, "tpot_ms": 164.24401741270898, "ttft_ms": 19699.39791597426, "visible_content_truncated": false, "visible_generation_ms": 6364.583306014538 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 6.177649978274647, "e2e_ms": 26122.358183958568, "effective_prefill_tok_s": 4071.234954066806, "generation_ms": 19910.48383002635, "mean_inter_token_latency_ms": 161.873852276637, "mean_time_per_output_token_ms": 161.873852276637, "output_token_source": "usage", "output_tokens": 124, "planned_context_tokens": 32768, "prompt_tokens": 25290, "reasoning_chunks": 92, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00023", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 23, "time_to_first_output_ms": 6211.874353932217, "tpot_ms": 161.873852276637, "ttft_ms": 25607.38544794731, "visible_content_truncated": false, "visible_generation_ms": 514.9727360112593 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 6.236577523951263, "e2e_ms": 35071.43940299284, "effective_prefill_tok_s": 4072.6598219034663, "generation_ms": 28861.983885988593, "mean_inter_token_latency_ms": 160.34435492215886, "mean_time_per_output_token_ms": 160.34435492215886, "output_token_source": "usage", "output_tokens": 181, "planned_context_tokens": 32768, "prompt_tokens": 25289, "reasoning_chunks": 132, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00024", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 24, "time_to_first_output_ms": 6209.455517004244, "tpot_ms": 160.34435492215886, "ttft_ms": 28707.237408030778, "visible_content_truncated": false, "visible_generation_ms": 6364.201994962059 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 8.616100034367657, "e2e_ms": 20241.225824924186, "effective_prefill_tok_s": 4080.513659642622, "generation_ms": 14043.476691003889, "mean_inter_token_latency_ms": 116.06179083474288, "mean_time_per_output_token_ms": 116.06179083474288, "output_token_source": "usage", "output_tokens": 122, "planned_context_tokens": 32768, "prompt_tokens": 25290, "reasoning_chunks": 90, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00025", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 25, "time_to_first_output_ms": 6197.749133920297, "tpot_ms": 116.06179083474288, "ttft_ms": 19724.960865918547, "visible_content_truncated": false, "visible_generation_ms": 516.264959005639 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 8.741537591174163, "e2e_ms": 20281.65327804163, "effective_prefill_tok_s": 4072.033245027469, "generation_ms": 14070.751136983745, "mean_inter_token_latency_ms": 114.39635070718492, "mean_time_per_output_token_ms": 114.39635070718492, "output_token_source": "usage", "output_tokens": 124, "planned_context_tokens": 32768, "prompt_tokens": 25291, "reasoning_chunks": 92, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00026", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 26, "time_to_first_output_ms": 6210.902141057886, "tpot_ms": 114.39635070718492, "ttft_ms": 13915.085226064548, "visible_content_truncated": false, "visible_generation_ms": 6366.568051977083 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 6.100645743587094, "e2e_ms": 26058.85909695644, "effective_prefill_tok_s": 4063.0416354677272, "generation_ms": 19833.96595797967, "mean_inter_token_latency_ms": 163.91707403288984, "mean_time_per_output_token_ms": 163.91707403288984, "output_token_source": "usage", "output_tokens": 122, "planned_context_tokens": 32768, "prompt_tokens": 25292, "reasoning_chunks": 90, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00027", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 27, "time_to_first_output_ms": 6224.893138976768, "tpot_ms": 163.91707403288984, "ttft_ms": 20906.738216988742, "visible_content_truncated": false, "visible_generation_ms": 5152.120879967697 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 6.184245282656403, "e2e_ms": 26092.691641068086, "effective_prefill_tok_s": 4077.091579420828, "generation_ms": 19889.249921077862, "mean_inter_token_latency_ms": 161.7012188705517, "mean_time_per_output_token_ms": 161.7012188705517, "output_token_source": "usage", "output_tokens": 124, "planned_context_tokens": 32768, "prompt_tokens": 25292, "reasoning_chunks": 92, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00028", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 28, "time_to_first_output_ms": 6203.441719990224, "tpot_ms": 161.7012188705517, "ttft_ms": 25577.825715998188, "visible_content_truncated": false, "visible_generation_ms": 514.8659250698984 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 8.992855582735748, "e2e_ms": 20163.206988945603, "effective_prefill_tok_s": 3709.1887368016596, "generation_ms": 13343.926063971594, "mean_inter_token_latency_ms": 111.19938386642995, "mean_time_per_output_token_ms": 111.19938386642995, "output_token_source": "usage", "output_tokens": 121, "planned_context_tokens": 32768, "prompt_tokens": 25294, "reasoning_chunks": 88, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00029", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 29, "time_to_first_output_ms": 6819.280924974009, "tpot_ms": 111.19938386642995, "ttft_ms": 19668.48014004063, "visible_content_truncated": false, "visible_generation_ms": 494.72684890497476 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 15.131258819055347, "e2e_ms": 18277.055279002525, "effective_prefill_tok_s": 2492.464750585708, "generation_ms": 8128.867628984153, "mean_inter_token_latency_ms": 66.08835470718824, "mean_time_per_output_token_ms": 66.08835470718824, "output_token_source": "usage", "output_tokens": 124, "planned_context_tokens": 32768, "prompt_tokens": 25294, "reasoning_chunks": 92, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00030", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 30, "time_to_first_output_ms": 10148.187650018372, "tpot_ms": 66.08835470718824, "ttft_ms": 17836.400859989226, "visible_content_truncated": false, "visible_generation_ms": 440.65441901329905 }, { "content_chunks": 27, "controlled_output": { "capture_complete": true, "exact_adherence": true, "observed_code_words": 16, "observed_extra_words": 0, "passed": true, "policy": "strict", "requested_words": 16 }, "decode_tok_s": 59.22180410778109, "e2e_ms": 8267.784975003451, "effective_prefill_tok_s": 4063.702807584961, "generation_ms": 2043.166394927539, "mean_inter_token_latency_ms": 16.88567268535156, "mean_time_per_output_token_ms": 16.88567268535156, "output_token_source": "usage", "output_tokens": 122, "planned_context_tokens": 32768, "prompt_tokens": 25295, "reasoning_chunks": 90, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_1912_00031", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 31, "time_to_first_output_ms": 6224.618580075912, "tpot_ms": 16.88567268535156, "ttft_ms": 7904.5311390655115, "visible_content_truncated": false, "visible_generation_ms": 363.2538359379396 } ], "requests": 32, "response_words": 16, "responses_received": 32, "run_id": "benchmark-20260919T112423Z", "sampling": { "temperature": { "effective_request": 0.0, "requested": 0.0, "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": "max", "shared_prefix_burst": false, "thinking_mode": "default" }, "shared_prefix_tokens": 0, "source_recipe": null, "started_at": "2026-09-19T11:24:23Z", "started_at_unix_ns": 1789817063064802504, "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": 227675.84933899343 }