{ "base_url": "http://127.0.0.1:39269/v1", "completed": 8, "concurrency": 4, "context_distribution": { "4096": 8 }, "context_seed": 0, "context_tokens": 4096, "engine": "sglang-5f55db35-dflash2-k16", "failed": 0, "failures": [], "finished_at": "2026-09-05T03:18:19Z", "gpu": "RTX PRO 6000 Blackwell Max-Q TP1", "manifest": null, "max_context_tokens": 262144, "max_tokens": 512, "measurement_protocol": "capacity-v3", "metrics": { "content_chunks": 117, "content_chunks_s": null, "decode_tok_s_max": 120.23496583963718, "decode_tok_s_mean": 90.06528808182483, "decode_tok_s_min": 64.05613161734671, "decode_tok_s_p25": 68.89159733990327, "decode_tok_s_p50": 71.45929643545314, "decode_tok_s_p75": 111.8637393858998, "decode_tok_s_p90": 120.23496583963718, "decode_tok_s_p95": 120.23496583963718, "decode_tok_s_p99": 120.23496583963718, "decode_tok_s_samples": 8, "e2e_max_ms": 1084.4914000481367, "e2e_mean_ms": 842.4737125169486, "e2e_min_ms": 651.8642000155523, "e2e_p25_ms": 717.3146000131965, "e2e_p50_ms": 778.5506000509486, "e2e_p75_ms": 875.6477000424638, "e2e_p90_ms": 1084.4914000481367, "e2e_p95_ms": 1084.4914000481367, "e2e_p99_ms": 1084.4914000481367, "e2e_samples": 8, "effective_prefill_tok_s_max": 20195.538429699478, "effective_prefill_tok_s_mean": 14245.102386243705, "effective_prefill_tok_s_min": 10071.894312678462, "effective_prefill_tok_s_p25": 10092.500151194661, "effective_prefill_tok_s_p50": 11887.762151755867, "effective_prefill_tok_s_p75": 17641.825527093788, "effective_prefill_tok_s_p90": 20195.538429699478, "effective_prefill_tok_s_p95": 20195.538429699478, "effective_prefill_tok_s_p99": 20195.538429699478, "effective_prefill_tok_s_samples": 8, "generation_max_ms": 780.5654000258073, "generation_mean_ms": 568.8733125134604, "generation_min_ms": 414.6633999189362, "generation_p25_ms": 449.12060000933707, "generation_p50_ms": 491.6696000145748, "generation_p75_ms": 693.8133999938145, "generation_p90_ms": 780.5654000258073, "generation_p95_ms": 780.5654000258073, "generation_p99_ms": 780.5654000258073, "generation_samples": 8, "mean_inter_token_latency_ms_max": 15.611308000516146, "mean_inter_token_latency_ms_mean": 11.80221985710462, "mean_inter_token_latency_ms_min": 8.317048148321057, "mean_inter_token_latency_ms_p25": 8.638820831644503, "mean_inter_token_latency_ms_p50": 10.242138637377964, "mean_inter_token_latency_ms_p75": 14.15945714273091, "mean_inter_token_latency_ms_p90": 15.611308000516146, "mean_inter_token_latency_ms_p95": 15.611308000516146, "mean_inter_token_latency_ms_p99": 15.611308000516146, "mean_inter_token_latency_ms_samples": 8, "output_token_sources": { "usage": 8 }, "output_tokens": 397, "prefix_cache_hit_avg": null, "prompt_token_samples": 8, "prompt_tokens": 28904, "prompt_tokens_max": 3613.0, "prompt_tokens_mean": 3613.0, "prompt_tokens_min": 3613.0, "prompt_tokens_p25": 3613.0, "prompt_tokens_p50": 3613.0, "prompt_tokens_p75": 3613.0, "prompt_tokens_p90": 3613.0, "prompt_tokens_p95": 3613.0, "prompt_tokens_p99": 3613.0, "prompt_tokens_samples": 8, "reasoning_chunks": 0, "throughput_tok_s": 224.19644353001542, "time_to_first_output_max_ms": 358.7210000259802, "time_to_first_output_mean_ms": 273.6004000034882, "time_to_first_output_min_ms": 178.90089994762093, "time_to_first_output_p25_ms": 201.21009997092187, "time_to_first_output_p50_ms": 225.64499999862164, "time_to_first_output_p75_ms": 357.6141999801621, "time_to_first_output_p90_ms": 358.7210000259802, "time_to_first_output_p95_ms": 358.7210000259802, "time_to_first_output_p99_ms": 358.7210000259802, "time_to_first_output_samples": 8, "tpot_max_ms": 15.611308000516146, "tpot_mean_ms": 11.80221985710462, "tpot_min_ms": 8.317048148321057, "tpot_p25_ms": 8.638820831644503, "tpot_p50_ms": 10.242138637377964, "tpot_p75_ms": 14.15945714273091, "tpot_p90_ms": 15.611308000516146, "tpot_p95_ms": 15.611308000516146, "tpot_p99_ms": 15.611308000516146, "tpot_samples": 8, "ttft_max_ms": 358.7212000275031, "ttft_mean_ms": 273.6007000057725, "ttft_min_ms": 178.90109994914383, "ttft_p25_ms": 201.21049997396767, "ttft_p50_ms": 225.64520000014454, "ttft_p75_ms": 357.6145999832079, "ttft_p90_ms": 358.7212000275031, "ttft_p95_ms": 358.7212000275031, "ttft_p99_ms": 358.7212000275031, "ttft_samples": 8 }, "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k16", "request_timings": [ { "content_chunks": 13, "decode_tok_s": 115.75653894070146, "e2e_ms": 773.3843999449164, "effective_prefill_tok_s": 10071.894312678462, "generation_ms": 414.6633999189362, "mean_inter_token_latency_ms": 8.638820831644503, "output_token_source": "usage", "output_tokens": 49, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 0, "time_to_first_output_ms": 358.7210000259802, "tpot_ms": 8.638820831644503, "ttft_ms": 358.7212000275031, "visible_generation_ms": 414.6631999174133 }, { "content_chunks": 17, "decode_tok_s": 64.05613161734671, "e2e_ms": 1084.4914000481367, "effective_prefill_tok_s": 11887.762151755867, "generation_ms": 780.5654000258073, "mean_inter_token_latency_ms": 15.611308000516146, "output_token_source": "usage", "output_tokens": 51, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 1, "time_to_first_output_ms": 303.9260000223294, "tpot_ms": 15.611308000516146, "ttft_ms": 303.9266000268981, "visible_generation_ms": 780.5648000212386 }, { "content_chunks": 14, "decode_tok_s": 120.23496583963718, "e2e_ms": 807.1092000463977, "effective_prefill_tok_s": 10092.500151194661, "generation_ms": 449.12060000933707, "mean_inter_token_latency_ms": 8.317048148321057, "output_token_source": "usage", "output_tokens": 55, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 2, "time_to_first_output_ms": 357.9886000370607, "tpot_ms": 8.317048148321057, "ttft_ms": 357.9888000385836, "visible_generation_ms": 449.12040000781417 }, { "content_chunks": 17, "decode_tok_s": 70.62417647228612, "e2e_ms": 1051.4275999739766, "effective_prefill_tok_s": 10103.066377678582, "generation_ms": 693.8133999938145, "mean_inter_token_latency_ms": 14.15945714273091, "output_token_source": "usage", "output_tokens": 50, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 3, "time_to_first_output_ms": 357.6141999801621, "tpot_ms": 14.15945714273091, "ttft_ms": 357.6145999832079, "visible_generation_ms": 693.8129999907687 }, { "content_chunks": 12, "decode_tok_s": 71.45929643545314, "e2e_ms": 778.5506000509486, "effective_prefill_tok_s": 17641.825527093788, "generation_ms": 573.7532000057399, "mean_inter_token_latency_ms": 13.993980487944876, "output_token_source": "usage", "output_tokens": 42, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 4, "time_to_first_output_ms": 204.79740004520863, "tpot_ms": 13.993980487944876, "ttft_ms": 204.79760004673153, "visible_generation_ms": 573.753000004217 }, { "content_chunks": 16, "decode_tok_s": 68.89159733990327, "e2e_ms": 875.6477000424638, "effective_prefill_tok_s": 20195.538429699478, "generation_ms": 696.7468000948429, "mean_inter_token_latency_ms": 14.515558335309228, "output_token_source": "usage", "output_tokens": 49, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 5, "time_to_first_output_ms": 178.90089994762093, "tpot_ms": 14.515558335309228, "ttft_ms": 178.90109994914383, "visible_generation_ms": 696.74660009332 }, { "content_chunks": 15, "decode_tok_s": 111.8637393858998, "e2e_ms": 717.3146000131965, "effective_prefill_tok_s": 16011.877063626804, "generation_ms": 491.6696000145748, "mean_inter_token_latency_ms": 8.93944727299227, "output_token_source": "usage", "output_tokens": 56, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 6, "time_to_first_output_ms": 225.64499999862164, "tpot_ms": 8.93944727299227, "ttft_ms": 225.64520000014454, "visible_generation_ms": 491.6694000130519 }, { "content_chunks": 13, "decode_tok_s": 97.63585862337094, "e2e_ms": 651.8642000155523, "effective_prefill_tok_s": 17956.355076222004, "generation_ms": 450.65410004463047, "mean_inter_token_latency_ms": 10.242138637377964, "output_token_source": "usage", "output_tokens": 45, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 7, "time_to_first_output_ms": 201.21009997092187, "tpot_ms": 10.242138637377964, "ttft_ms": 201.21049997396767, "visible_generation_ms": 450.65370004158467 } ], "requests": 8, "run_id": "benchmark-20260905T031817Z", "schema": "anvil-serving.benchmark/v1", "serve_flags": { "no_thinking": true, "reasoning_effort": null, "shared_prefix_burst": false, "thinking_mode": "disabled" }, "source_recipe": null, "started_at": "2026-09-05T03:18:17Z", "tier": null, "timing_methodology": { "clock": "client time.perf_counter", "decode": "usage completion tokens after the first token divided by client-observed generation time; completion tokens may include reasoning tokens", "distributions": "request-level 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": "client-observed generation time divided by completion tokens after the first token; not a raw per-token timestamp trace", "time_to_first_output": "request start through the first non-empty streamed reasoning or content delta", "tpot": "alias of mean_inter_token_latency for capacity-v3: client-observed generation time divided by completion tokens after the first token", "ttft": "request start through the first non-empty streamed content delta" }, "wall_clock_ms": 1770.7684999331832 }