{ "base_url": "http://127.0.0.1:30000/v1", "clock_domain_id": null, "completed": 32, "concurrency": 4, "configuration_fingerprint": "sha256:aaed9d0d1186574e41f0de85bc287d313578f906210a54e3386929fe01549908", "context_distribution": { "32768": 32 }, "context_seed": 1912, "context_tokens": 32768, "controlled_output_policy": "strict", "engine": "vllm-v84", "failed": 0, "failures": [], "finished_at": "2026-09-19T11:45:52Z", "finished_at_unix_ns": 1789818352760382262, "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": 213.0, "completion_tokens_mean": 139.0625, "completion_tokens_mean_ci95_high": 149.7718353815499, "completion_tokens_mean_ci95_low": 128.3531646184501, "completion_tokens_min": 95.0, "completion_tokens_p25": 122.0, "completion_tokens_p50": 122.0, "completion_tokens_p75": 181.0, "completion_tokens_p90": 182.0, "completion_tokens_p95": 186.0, "completion_tokens_p99": 213.0, "completion_tokens_samples": 32, "completion_tokens_stddev": 30.908749674673235, "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": 62.62876258164369, "decode_tok_s_mean": 8.926222326610922, "decode_tok_s_mean_ci95_high": 12.466199709576467, "decode_tok_s_mean_ci95_low": 5.386244943645378, "decode_tok_s_min": 5.005408256024358, "decode_tok_s_p25": 6.197260724489422, "decode_tok_s_p50": 6.279049545474271, "decode_tok_s_p75": 8.098056967146018, "decode_tok_s_p90": 8.925571204390417, "decode_tok_s_p95": 21.567625444156892, "decode_tok_s_p99": 62.62876258164369, "decode_tok_s_samples": 32, "decode_tok_s_stddev": 10.216906174456915, "e2e_max_ms": 50690.20374608226, "e2e_mean_ci95_high_ms": 30228.82122951217, "e2e_mean_ci95_low_ms": 25039.222170414512, "e2e_mean_ms": 27634.02169996334, "e2e_min_ms": 11428.57501306571, "e2e_p25_ms": 25492.959321010858, "e2e_p50_ms": 25641.06835005805, "e2e_p75_ms": 28050.74174690526, "e2e_p90_ms": 37002.18298193067, "e2e_p95_ms": 42158.96884794347, "e2e_p99_ms": 50690.20374608226, "e2e_samples": 32, "e2e_stddev_ms": 7488.980992504657, "effective_prefill_tok_s_max": 4228.0564612131475, "effective_prefill_tok_s_mean": 3587.1328610339365, "effective_prefill_tok_s_mean_ci95_high": 3927.827405312716, "effective_prefill_tok_s_mean_ci95_low": 3246.438316755157, "effective_prefill_tok_s_min": 889.8702716958536, "effective_prefill_tok_s_p25": 2956.924572335684, "effective_prefill_tok_s_p50": 4173.184386193486, "effective_prefill_tok_s_p75": 4188.886499881146, "effective_prefill_tok_s_p90": 4204.182545476011, "effective_prefill_tok_s_p95": 4224.836614012895, "effective_prefill_tok_s_p99": 4228.0564612131475, "effective_prefill_tok_s_samples": 32, "effective_prefill_tok_s_stddev": 983.2956023379004, "generation_max_ms": 35961.10263001174, "generation_mean_ci95_high_ms": 21353.862631503747, "generation_mean_ci95_low_ms": 17595.490457361135, "generation_mean_ms": 19474.67654443244, "generation_min_ms": 2874.078819062561, "generation_p25_ms": 19010.319559951313, "generation_p50_ms": 19489.605233073235, "generation_p75_ms": 20614.42562995944, "generation_p90_ms": 23892.416129936464, "generation_p95_ms": 26179.119369015098, "generation_p99_ms": 35961.10263001174, "generation_samples": 32, "generation_stddev_ms": 5423.613164406268, "mean_inter_token_latency_ms_max": 199.7839035000652, "mean_inter_token_latency_ms_mean": 145.96901703839387, "mean_inter_token_latency_ms_mean_ci95_high": 159.72891201875228, "mean_inter_token_latency_ms_mean_ci95_low": 132.20912205803546, "mean_inter_token_latency_ms_min": 15.96710455034756, "mean_inter_token_latency_ms_p25": 114.4165332442046, "mean_inter_token_latency_ms_p50": 158.86968079617444, "mean_inter_token_latency_ms_p75": 161.3464342720791, "mean_inter_token_latency_ms_p90": 186.45235676075956, "mean_inter_token_latency_ms_p95": 197.45798454492945, "mean_inter_token_latency_ms_p99": 199.7839035000652, "mean_inter_token_latency_ms_samples": 32, "mean_inter_token_latency_ms_stddev": 39.713122649086436, "mean_time_per_output_token_max_ms": 199.7839035000652, "mean_time_per_output_token_mean_ci95_high_ms": 159.72891201875228, "mean_time_per_output_token_mean_ci95_low_ms": 132.20912205803546, "mean_time_per_output_token_mean_ms": 145.96901703839387, "mean_time_per_output_token_min_ms": 15.96710455034756, "mean_time_per_output_token_p25_ms": 114.4165332442046, "mean_time_per_output_token_p50_ms": 158.86968079617444, "mean_time_per_output_token_p75_ms": 161.3464342720791, "mean_time_per_output_token_p90_ms": 186.45235676075956, "mean_time_per_output_token_p95_ms": 197.45798454492945, "mean_time_per_output_token_p99_ms": 199.7839035000652, "mean_time_per_output_token_samples": 32, "mean_time_per_output_token_stddev_ms": 39.713122649086436, "output_token_sources": { "usage": 32 }, "output_tokens": 4450, "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": 3272, "request_canary_passed": 32, "request_canary_samples": 32, "throughput_performance_eligible": true, "throughput_tok_s": 20.023399778930106, "time_to_first_output_max_ms": 28414.25408201758, "time_to_first_output_mean_ci95_high_ms": 9762.715764304618, "time_to_first_output_mean_ci95_low_ms": 6555.974546757183, "time_to_first_output_mean_ms": 8159.3451555309, "time_to_first_output_min_ms": 5980.052591999993, "time_to_first_output_p25_ms": 6034.313054988161, "time_to_first_output_p50_ms": 6056.0680370545015, "time_to_first_output_p75_ms": 7970.719125005417, "time_to_first_output_p90_ms": 11740.686884033494, "time_to_first_output_p95_ms": 17840.444164001383, "time_to_first_output_p99_ms": 28414.25408201758, "time_to_first_output_samples": 32, "time_to_first_output_stddev_ms": 4627.568286608564, "tpot_max_ms": 199.7839035000652, "tpot_mean_ci95_high_ms": 159.72891201875228, "tpot_mean_ci95_low_ms": 132.20912205803546, "tpot_mean_ms": 145.96901703839387, "tpot_min_ms": 15.96710455034756, "tpot_p25_ms": 114.4165332442046, "tpot_p50_ms": 158.86968079617444, "tpot_p75_ms": 161.3464342720791, "tpot_p90_ms": 186.45235676075956, "tpot_p95_ms": 197.45798454492945, "tpot_p99_ms": 199.7839035000652, "tpot_samples": 32, "tpot_stddev_ms": 39.713122649086436, "ttft_max_ms": 41645.41569794528, "ttft_mean_ci95_high_ms": 26151.802156269936, "ttft_mean_ci95_low_ms": 22055.67556635991, "ttft_mean_ms": 24103.738861314923, "ttft_min_ms": 11038.879632018507, "ttft_p25_ms": 20105.78355391044, "ttft_p50_ms": 24306.48852896411, "ttft_p75_ms": 26114.31557603646, "ttft_p90_ms": 31252.784471027553, "ttft_p95_ms": 33348.89446105808, "ttft_p99_ms": 41645.41569794528, "ttft_samples": 32, "ttft_stddev_ms": 5911.01813943655 }, "model": "glm53-flash-exl3-4bpw-v84-fp8-tp2-c4-327k-r9-vision8-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": 5.216296378399627, "e2e_ms": 41420.39267101791, "effective_prefill_tok_s": 1417.1732366964427, "generation_ms": 23579.948507016525, "mean_inter_token_latency_ms": 191.70689843102866, "mean_time_per_output_token_ms": 191.70689843102866, "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": 17840.444164001383, "tpot_ms": 191.70689843102866, "ttft_ms": 30003.862192039378, "visible_content_truncated": false, "visible_generation_ms": 11416.53047897853 }, { "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": 5.363300402167178, "e2e_ms": 34301.4220520854, "effective_prefill_tok_s": 2153.536692506846, "generation_ms": 22560.735168051906, "mean_inter_token_latency_ms": 186.45235676075956, "mean_time_per_output_token_ms": 186.45235676075956, "output_token_source": "usage", "output_tokens": 122, "planned_context_tokens": 32768, "prompt_tokens": 25284, "reasoning_chunks": 90, "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": 11740.686884033494, "tpot_ms": 186.45235676075956, "ttft_ms": 29721.56049509067, "visible_content_truncated": false, "visible_generation_ms": 4579.861556994729 }, { "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": 5.431867185226939, "e2e_ms": 50690.20374608226, "effective_prefill_tok_s": 889.8702716958536, "generation_ms": 22275.949664064683, "mean_inter_token_latency_ms": 184.09875755425358, "mean_time_per_output_token_ms": 184.09875755425358, "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_00002", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 2, "time_to_first_output_ms": 28414.25408201758, "tpot_ms": 184.09875755425358, "ttft_ms": 33348.89446105808, "visible_content_truncated": false, "visible_generation_ms": 17341.30928502418 }, { "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": 5.064368515178786, "e2e_ms": 30002.06959794741, "effective_prefill_tok_s": 4138.532591478017, "generation_ms": 23892.416129936464, "mean_inter_token_latency_ms": 197.45798454492945, "mean_time_per_output_token_ms": 197.45798454492945, "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_00003", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 3, "time_to_first_output_ms": 6109.653468010947, "tpot_ms": 197.45798454492945, "ttft_ms": 29492.319167940877, "visible_content_truncated": false, "visible_generation_ms": 509.75043000653386 }, { "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": 5.005408256024358, "e2e_ms": 42158.96884794347, "effective_prefill_tok_s": 4079.629845324052, "generation_ms": 35961.10263001174, "mean_inter_token_latency_ms": 199.7839035000652, "mean_time_per_output_token_ms": 199.7839035000652, "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_00004", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 4, "time_to_first_output_ms": 6197.8662179317325, "tpot_ms": 199.7839035000652, "ttft_ms": 41645.41569794528, "visible_content_truncated": false, "visible_generation_ms": 513.553149998188 }, { "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": 8.925571204390417, "e2e_ms": 26159.371722955257, "effective_prefill_tok_s": 2006.2182323125692, "generation_ms": 13556.555342976935, "mean_inter_token_latency_ms": 112.0376474626193, "mean_time_per_output_token_ms": 112.0376474626193, "output_token_source": "usage", "output_tokens": 122, "planned_context_tokens": 32768, "prompt_tokens": 25284, "reasoning_chunks": 90, "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": 12602.816379978321, "tpot_ms": 112.0376474626193, "ttft_ms": 25644.700999953784, "visible_content_truncated": false, "visible_generation_ms": 514.6707230014727 }, { "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": 7.296456723663749, "e2e_ms": 37002.18298193067, "effective_prefill_tok_s": 2170.6970793686323, "generation_ms": 25354.772460996173, "mean_inter_token_latency_ms": 137.05282411349282, "mean_time_per_output_token_ms": 137.05282411349282, "output_token_source": "usage", "output_tokens": 186, "planned_context_tokens": 32768, "prompt_tokens": 25283, "reasoning_chunks": 137, "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": 11647.4105209345, "tpot_ms": 137.05282411349282, "ttft_ms": 31252.784471027553, "visible_content_truncated": false, "visible_generation_ms": 5749.39851090312 }, { "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": 8.098056967146018, "e2e_ms": 34149.838494020514, "effective_prefill_tok_s": 3171.8593521488338, "generation_ms": 26179.119369015098, "mean_inter_token_latency_ms": 123.48641211799574, "mean_time_per_output_token_ms": 123.48641211799574, "output_token_source": "usage", "output_tokens": 213, "planned_context_tokens": 32768, "prompt_tokens": 25282, "reasoning_chunks": 163, "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": 7970.719125005417, "tpot_ms": 123.48641211799574, "ttft_ms": 32052.905441028997, "visible_content_truncated": false, "visible_generation_ms": 2096.933052991517 }, { "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": 6.25779803767229, "e2e_ms": 30454.07045504544, "effective_prefill_tok_s": 2274.0200847055344, "generation_ms": 19335.87489905767, "mean_inter_token_latency_ms": 159.80061900047662, "mean_time_per_output_token_ms": 159.80061900047662, "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": 11118.195555987768, "tpot_ms": 159.80061900047662, "ttft_ms": 24245.836218004115, "visible_content_truncated": false, "visible_generation_ms": 6208.234237041324 }, { "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": 5.365506859489175, "e2e_ms": 24990.372151951306, "effective_prefill_tok_s": 4228.0564612131475, "generation_ms": 19010.319559951313, "mean_inter_token_latency_ms": 186.375681960307, "mean_time_per_output_token_ms": 186.375681960307, "output_token_source": "usage", "output_tokens": 103, "planned_context_tokens": 32768, "prompt_tokens": 25284, "reasoning_chunks": 72, "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": 5980.052591999993, "tpot_ms": 186.375681960307, "ttft_ms": 18771.266203955747, "visible_content_truncated": false, "visible_generation_ms": 6219.105947995558 }, { "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": 6.23150690600078, "e2e_ms": 25402.30103500653, "effective_prefill_tok_s": 4224.836614012895, "generation_ms": 19417.454208945855, "mean_inter_token_latency_ms": 160.47482817310623, "mean_time_per_output_token_ms": 160.47482817310623, "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": 5984.846826060675, "tpot_ms": 160.47482817310623, "ttft_ms": 24889.89291805774, "visible_content_truncated": false, "visible_generation_ms": 512.4081169487908 }, { "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": 6.311056510845738, "e2e_ms": 25492.959321010858, "effective_prefill_tok_s": 4211.812201916337, "generation_ms": 19489.605233073235, "mean_inter_token_latency_ms": 158.45207506563605, "mean_time_per_output_token_ms": 158.45207506563605, "output_token_source": "usage", "output_tokens": 124, "planned_context_tokens": 32768, "prompt_tokens": 25285, "reasoning_chunks": 92, "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": 6003.354087937623, "tpot_ms": 158.45207506563605, "ttft_ms": 24306.48852896411, "visible_content_truncated": false, "visible_generation_ms": 1186.470792046748 }, { "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": 8.780259185924072, "e2e_ms": 26628.912328975275, "effective_prefill_tok_s": 4204.182545476011, "generation_ms": 20614.42562995944, "mean_inter_token_latency_ms": 113.89185430916818, "mean_time_per_output_token_ms": 113.89185430916818, "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": 6014.486699015833, "tpot_ms": 113.89185430916818, "ttft_ms": 26114.31557603646, "visible_content_truncated": false, "visible_generation_ms": 514.5967529388145 }, { "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": 8.94586668375876, "e2e_ms": 26711.456581950188, "effective_prefill_tok_s": 4192.147379748084, "generation_ms": 20679.941534996033, "mean_inter_token_latency_ms": 111.7834677567353, "mean_time_per_output_token_ms": 111.7834677567353, "output_token_source": "usage", "output_tokens": 186, "planned_context_tokens": 32768, "prompt_tokens": 25285, "reasoning_chunks": 136, "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": 6031.515046954155, "tpot_ms": 111.7834677567353, "ttft_ms": 21384.819295955822, "visible_content_truncated": false, "visible_generation_ms": 5326.637285994366 }, { "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": 8.782717169464146, "e2e_ms": 26645.679697045125, "effective_prefill_tok_s": 4188.488028491084, "generation_ms": 20608.656354015693, "mean_inter_token_latency_ms": 113.85997985644029, "mean_time_per_output_token_ms": 113.85997985644029, "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_00014", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 14, "time_to_first_output_ms": 6037.023343029432, "tpot_ms": 113.85997985644029, "ttft_ms": 26130.117880995385, "visible_content_truncated": false, "visible_generation_ms": 515.5618160497397 }, { "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": 8.739995625157187, "e2e_ms": 26629.28903894499, "effective_prefill_tok_s": 4190.369271461275, "generation_ms": 20594.97598395683, "mean_inter_token_latency_ms": 114.4165332442046, "mean_time_per_output_token_ms": 114.4165332442046, "output_token_source": "usage", "output_tokens": 181, "planned_context_tokens": 32768, "prompt_tokens": 25286, "reasoning_chunks": 132, "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": 6034.313054988161, "tpot_ms": 114.4165332442046, "ttft_ms": 24980.0215610303, "visible_content_truncated": false, "visible_generation_ms": 1649.267477914691 }, { "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": 6.210289243711332, "e2e_ms": 25503.94814100582, "effective_prefill_tok_s": 4200.391203770989, "generation_ms": 19483.79459499847, "mean_inter_token_latency_ms": 161.02309582643366, "mean_time_per_output_token_ms": 161.02309582643366, "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_00016", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 16, "time_to_first_output_ms": 6020.15354600735, "tpot_ms": 161.02309582643366, "ttft_ms": 19265.891690040007, "visible_content_truncated": false, "visible_generation_ms": 6238.056450965814 }, { "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": 6.20981553982448, "e2e_ms": 25521.96867193561, "effective_prefill_tok_s": 4188.886499881146, "generation_ms": 19485.280878958292, "mean_inter_token_latency_ms": 161.03537916494457, "mean_time_per_output_token_ms": 161.03537916494457, "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": 6036.687792977318, "tpot_ms": 161.03537916494457, "ttft_ms": 20179.35489094816, "visible_content_truncated": false, "visible_generation_ms": 5342.613780987449 }, { "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": 6.656418799621106, "e2e_ms": 25721.427125041373, "effective_prefill_tok_s": 4185.940625395091, "generation_ms": 19680.252091027796, "mean_inter_token_latency_ms": 150.2309319925786, "mean_time_per_output_token_ms": 150.2309319925786, "output_token_source": "usage", "output_tokens": 132, "planned_context_tokens": 32768, "prompt_tokens": 25288, "reasoning_chunks": 99, "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": 6041.175034013577, "tpot_ms": 150.2309319925786, "ttft_ms": 25207.20035303384, "visible_content_truncated": false, "visible_generation_ms": 514.2267720075324 }, { "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": 6.294467232441748, "e2e_ms": 25574.20223101508, "effective_prefill_tok_s": 4191.28621684418, "generation_ms": 19540.970737929456, "mean_inter_token_latency_ms": 158.86968079617444, "mean_time_per_output_token_ms": 158.86968079617444, "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": 6033.231493085623, "tpot_ms": 158.86968079617444, "ttft_ms": 20223.603987018578, "visible_content_truncated": false, "visible_generation_ms": 5350.598243996501 }, { "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": 6.279049545474271, "e2e_ms": 24829.62946803309, "effective_prefill_tok_s": 4188.851922087992, "generation_ms": 18792.653115000576, "mean_inter_token_latency_ms": 159.25977216102183, "mean_time_per_output_token_ms": 159.25977216102183, "output_token_source": "usage", "output_tokens": 119, "planned_context_tokens": 32768, "prompt_tokens": 25288, "reasoning_chunks": 87, "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": 6036.976353032514, "tpot_ms": 159.25977216102183, "ttft_ms": 19241.801223019138, "visible_content_truncated": false, "visible_generation_ms": 5587.828245013952 }, { "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": 6.197260724489422, "e2e_ms": 25573.30222800374, "effective_prefill_tok_s": 4181.004273747142, "generation_ms": 19524.75543296896, "mean_inter_token_latency_ms": 161.3616151485038, "mean_time_per_output_token_ms": 161.3616151485038, "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": 6048.546795034781, "tpot_ms": 161.3616151485038, "ttft_ms": 22068.217006977648, "visible_content_truncated": false, "visible_generation_ms": 3505.0852210260928 }, { "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": 6.197843816701249, "e2e_ms": 25570.311668911017, "effective_prefill_tok_s": 4181.8018921317525, "generation_ms": 19522.918546921574, "mean_inter_token_latency_ms": 161.3464342720791, "mean_time_per_output_token_ms": 161.3464342720791, "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": 6047.393121989444, "tpot_ms": 161.3464342720791, "ttft_ms": 25060.498698963784, "visible_content_truncated": false, "visible_generation_ms": 509.8129699472338 }, { "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": 6.637290215552148, "e2e_ms": 20218.474221997894, "effective_prefill_tok_s": 4175.976862423153, "generation_ms": 14162.406184943393, "mean_inter_token_latency_ms": 150.6638955845042, "mean_time_per_output_token_ms": 150.6638955845042, "output_token_source": "usage", "output_tokens": 95, "planned_context_tokens": 32768, "prompt_tokens": 25290, "reasoning_chunks": 64, "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": 6056.0680370545015, "tpot_ms": 150.6638955845042, "ttft_ms": 18810.3819739772, "visible_content_truncated": false, "visible_generation_ms": 1408.0922480206937 }, { "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": 6.263748994166952, "e2e_ms": 26182.13598697912, "effective_prefill_tok_s": 3771.672731556325, "generation_ms": 19477.153397048824, "mean_inter_token_latency_ms": 159.64879833646577, "mean_time_per_output_token_ms": 159.64879833646577, "output_token_source": "usage", "output_tokens": 123, "planned_context_tokens": 32768, "prompt_tokens": 25289, "reasoning_chunks": 91, "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": 6704.982589930296, "tpot_ms": 159.64879833646577, "ttft_ms": 21605.005137971602, "visible_content_truncated": false, "visible_generation_ms": 4577.130849007517 }, { "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": 6.193933400454515, "e2e_ms": 25590.510117006488, "effective_prefill_tok_s": 4176.529868486338, "generation_ms": 19535.24395194836, "mean_inter_token_latency_ms": 161.44829712354016, "mean_time_per_output_token_ms": 161.44829712354016, "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": 6055.2661650581285, "tpot_ms": 161.44829712354016, "ttft_ms": 21543.037784053013, "visible_content_truncated": false, "visible_generation_ms": 4047.4723329534754 }, { "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": 6.354978561523728, "e2e_ms": 25641.06835005805, "effective_prefill_tok_s": 4023.280112332219, "generation_ms": 19354.904003092088, "mean_inter_token_latency_ms": 157.35694311456982, "mean_time_per_output_token_ms": 157.35694311456982, "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": 6286.164346965961, "tpot_ms": 157.35694311456982, "ttft_ms": 25128.31591104623, "visible_content_truncated": false, "visible_generation_ms": 512.7524390118197 }, { "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": 7.081932778924135, "e2e_ms": 28050.74174690526, "effective_prefill_tok_s": 2306.6097332823533, "generation_ms": 17085.731222992763, "mean_inter_token_latency_ms": 141.20439027266747, "mean_time_per_output_token_ms": 141.20439027266747, "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": 10965.010523912497, "tpot_ms": 141.20439027266747, "ttft_ms": 24325.115763000213, "visible_content_truncated": false, "visible_generation_ms": 3725.625983905047 }, { "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": 6.2147657844145465, "e2e_ms": 25602.49009996187, "effective_prefill_tok_s": 4124.101457255621, "generation_ms": 19469.760276959278, "mean_inter_token_latency_ms": 160.90710972693617, "mean_time_per_output_token_ms": 160.90710972693617, "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_00028", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 28, "time_to_first_output_ms": 6132.729823002592, "tpot_ms": 160.90710972693617, "ttft_ms": 20105.78355391044, "visible_content_truncated": false, "visible_generation_ms": 5496.706546051428 }, { "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": 8.834899177940104, "e2e_ms": 19756.764377001673, "effective_prefill_tok_s": 4173.184386193486, "generation_ms": 13695.685436017811, "mean_inter_token_latency_ms": 113.18748294229596, "mean_time_per_output_token_ms": 113.18748294229596, "output_token_source": "usage", "output_tokens": 122, "planned_context_tokens": 32768, "prompt_tokens": 25294, "reasoning_chunks": 90, "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": 6061.078940983862, "tpot_ms": 113.18748294229596, "ttft_ms": 19254.94041002821, "visible_content_truncated": false, "visible_generation_ms": 501.8239669734612 }, { "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": 21.567625444156892, "e2e_ms": 14683.65422799252, "effective_prefill_tok_s": 4020.3793068068826, "generation_ms": 8392.20805594232, "mean_inter_token_latency_ms": 46.36579036432221, "mean_time_per_output_token_ms": 46.36579036432221, "output_token_source": "usage", "output_tokens": 182, "planned_context_tokens": 32768, "prompt_tokens": 25294, "reasoning_chunks": 133, "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": 6291.4461720502, "tpot_ms": 46.36579036432221, "ttft_ms": 14276.41443698667, "visible_content_truncated": false, "visible_generation_ms": 407.23979100584984 }, { "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": 62.62876258164369, "e2e_ms": 11428.57501306571, "effective_prefill_tok_s": 2956.924572335684, "generation_ms": 2874.078819062561, "mean_inter_token_latency_ms": 15.96710455034756, "mean_time_per_output_token_ms": 15.96710455034756, "output_token_source": "usage", "output_tokens": 181, "planned_context_tokens": 32768, "prompt_tokens": 25295, "reasoning_chunks": 132, "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": 8554.49619400315, "tpot_ms": 15.96710455034756, "ttft_ms": 11038.879632018507, "visible_content_truncated": false, "visible_generation_ms": 389.69538104720414 } ], "requests": 32, "response_words": 16, "responses_received": 32, "run_id": "benchmark-20260919T114210Z", "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": "2026-09-19T11:42:10Z", "started_at_unix_ns": 1789818130520398544, "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": 222239.98167796526 }