{ "base_url": "http://127.0.0.1:39276/v1", "completed": 8, "concurrency": 4, "context_distribution": { "4096": 8 }, "context_seed": 0, "context_tokens": 4096, "engine": "sglang", "failed": 0, "failures": [], "finished_at": "2026-09-05T03:46:16Z", "gpu": "Primary Node / RTX PRO 6000 Blackwell Max-Q / TP=1", "manifest": null, "max_context_tokens": 262144, "max_tokens": 512, "measurement_protocol": "capacity-v3", "metrics": { "content_chunks": 113, "content_chunks_s": null, "decode_tok_s_max": 121.71347713435081, "decode_tok_s_mean": 105.33166302586653, "decode_tok_s_min": 80.37775246589307, "decode_tok_s_p25": 95.34160898944239, "decode_tok_s_p50": 108.80989751225708, "decode_tok_s_p75": 111.96962429470061, "decode_tok_s_p90": 121.71347713435081, "decode_tok_s_p95": 121.71347713435081, "decode_tok_s_p99": 121.71347713435081, "decode_tok_s_samples": 8, "e2e_max_ms": 868.1024999823421, "e2e_mean_ms": 649.1315250168554, "e2e_min_ms": 551.9587000599131, "e2e_p25_ms": 557.6773000648245, "e2e_p50_ms": 591.8942999560386, "e2e_p75_ms": 666.039300034754, "e2e_p90_ms": 868.1024999823421, "e2e_p95_ms": 868.1024999823421, "e2e_p99_ms": 868.1024999823421, "e2e_samples": 8, "effective_prefill_tok_s_max": 27329.38987142653, "effective_prefill_tok_s_mean": 21302.332820669555, "effective_prefill_tok_s_min": 17473.2339882373, "effective_prefill_tok_s_p25": 19907.76199227129, "effective_prefill_tok_s_p50": 21010.345839788926, "effective_prefill_tok_s_p75": 21080.293580311434, "effective_prefill_tok_s_p90": 27329.38987142653, "effective_prefill_tok_s_p95": 27329.38987142653, "effective_prefill_tok_s_p99": 27329.38987142653, "effective_prefill_tok_s_samples": 8, "generation_max_ms": 696.7102000489831, "generation_mean_ms": 477.0971375401132, "generation_min_ms": 386.15280005615205, "generation_p25_ms": 419.7567000519484, "generation_p50_ms": 431.94600008428097, "generation_p75_ms": 484.76350004784763, "generation_p90_ms": 696.7102000489831, "generation_p95_ms": 696.7102000489831, "generation_p99_ms": 696.7102000489831, "generation_samples": 8, "mean_inter_token_latency_ms_max": 12.44125357230327, "mean_inter_token_latency_ms_mean": 9.638829585194099, "mean_inter_token_latency_ms_min": 8.21601702247132, "mean_inter_token_latency_ms_p25": 8.813881819051774, "mean_inter_token_latency_ms_p50": 8.934710637983331, "mean_inter_token_latency_ms_p75": 10.094839584780857, "mean_inter_token_latency_ms_p90": 12.44125357230327, "mean_inter_token_latency_ms_p95": 12.44125357230327, "mean_inter_token_latency_ms_p99": 12.44125357230327, "mean_inter_token_latency_ms_samples": 8, "output_token_sources": { "usage": 8 }, "output_tokens": 402, "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": 282.75512371908275, "time_to_first_output_max_ms": 206.7733999574557, "time_to_first_output_mean_ms": 172.03438747674227, "time_to_first_output_min_ms": 132.20200000796467, "time_to_first_output_p25_ms": 159.88289995584637, "time_to_first_output_p50_ms": 171.52450000867248, "time_to_first_output_p75_ms": 181.05010001454502, "time_to_first_output_p90_ms": 206.7733999574557, "time_to_first_output_p95_ms": 206.7733999574557, "time_to_first_output_p99_ms": 206.7733999574557, "time_to_first_output_samples": 8, "tpot_max_ms": 12.44125357230327, "tpot_mean_ms": 9.638829585194099, "tpot_min_ms": 8.21601702247132, "tpot_p25_ms": 8.813881819051774, "tpot_p50_ms": 8.934710637983331, "tpot_p75_ms": 10.094839584780857, "tpot_p90_ms": 12.44125357230327, "tpot_p95_ms": 12.44125357230327, "tpot_p99_ms": 12.44125357230327, "tpot_samples": 8, "ttft_max_ms": 206.77369995974004, "ttft_mean_ms": 172.03473747940734, "ttft_min_ms": 132.20240001101047, "ttft_p25_ms": 159.88359996117651, "ttft_p50_ms": 171.52500001247972, "ttft_p75_ms": 181.05030001606792, "ttft_p90_ms": 206.77369995974004, "ttft_p95_ms": 206.77369995974004, "ttft_p99_ms": 206.77369995974004, "ttft_samples": 8 }, "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k12-chunk1k-mamba40", "request_timings": [ { "content_chunks": 14, "decode_tok_s": 108.80989751225708, "e2e_ms": 591.8289000401273, "effective_prefill_tok_s": 22597.78876288693, "generation_ms": 431.94600008428097, "mean_inter_token_latency_ms": 9.190340427325127, "output_token_source": "usage", "output_tokens": 48, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 0, "time_to_first_output_ms": 159.88289995584637, "tpot_ms": 9.190340427325127, "ttft_ms": 159.88359996117651, "visible_generation_ms": 431.9453000789508 }, { "content_chunks": 13, "decode_tok_s": 111.92304267233793, "e2e_ms": 591.8942999560386, "effective_prefill_tok_s": 21010.345839788926, "generation_ms": 419.93139998521656, "mean_inter_token_latency_ms": 8.934710637983331, "output_token_source": "usage", "output_tokens": 48, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 1, "time_to_first_output_ms": 171.96289997082204, "tpot_ms": 8.934710637983331, "ttft_ms": 171.96319997310638, "visible_generation_ms": 419.9310999829322 }, { "content_chunks": 13, "decode_tok_s": 121.71347713435081, "e2e_ms": 557.6773000648245, "effective_prefill_tok_s": 21064.04624305754, "generation_ms": 386.15280005615205, "mean_inter_token_latency_ms": 8.21601702247132, "output_token_source": "usage", "output_tokens": 48, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 2, "time_to_first_output_ms": 171.52450000867248, "tpot_ms": 8.21601702247132, "ttft_ms": 171.52500001247972, "visible_generation_ms": 386.1523000523448 }, { "content_chunks": 18, "decode_tok_s": 80.37775246589307, "e2e_ms": 868.1024999823421, "effective_prefill_tok_s": 21080.293580311434, "generation_ms": 696.7102000489831, "mean_inter_token_latency_ms": 12.44125357230327, "output_token_source": "usage", "output_tokens": 57, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 3, "time_to_first_output_ms": 171.39229993335903, "tpot_ms": 12.44125357230327, "ttft_ms": 171.39249993488193, "visible_generation_ms": 696.7100000474602 }, { "content_chunks": 13, "decode_tok_s": 95.34160898944239, "e2e_ms": 699.7375999344513, "effective_prefill_tok_s": 17473.2339882373, "generation_ms": 492.9641999769956, "mean_inter_token_latency_ms": 10.488599999510544, "output_token_source": "usage", "output_tokens": 48, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 4, "time_to_first_output_ms": 206.7733999574557, "tpot_ms": 10.488599999510544, "ttft_ms": 206.77369995974004, "visible_generation_ms": 492.96389997471124 }, { "content_chunks": 14, "decode_tok_s": 99.06051419654217, "e2e_ms": 666.039300034754, "effective_prefill_tok_s": 19907.76199227129, "generation_ms": 484.55230006948113, "mean_inter_token_latency_ms": 10.094839584780857, "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": 181.48699996527284, "tpot_ms": 10.094839584780857, "ttft_ms": 181.48719996679574, "visible_generation_ms": 484.55210006795824 }, { "content_chunks": 14, "decode_tok_s": 113.4573869414082, "e2e_ms": 665.8136000623927, "effective_prefill_tok_s": 19955.80228737649, "generation_ms": 484.76350004784763, "mean_inter_token_latency_ms": 8.813881819051774, "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": 181.05010001454502, "tpot_ms": 8.813881819051774, "ttft_ms": 181.05030001606792, "visible_generation_ms": 484.76330004632473 }, { "content_chunks": 14, "decode_tok_s": 111.96962429470061, "e2e_ms": 551.9587000599131, "effective_prefill_tok_s": 27329.38987142653, "generation_ms": 419.7567000519484, "mean_inter_token_latency_ms": 8.930993618126562, "output_token_source": "usage", "output_tokens": 48, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 7, "time_to_first_output_ms": 132.20200000796467, "tpot_ms": 8.930993618126562, "ttft_ms": 132.20240001101047, "visible_generation_ms": 419.75630004890263 } ], "requests": 8, "run_id": "benchmark-20260905T034615Z", "schema": "anvil-serving.benchmark/v1", "serve_flags": { "no_thinking": false, "reasoning_effort": null, "shared_prefix_burst": false, "thinking_mode": "disabled" }, "source_recipe": null, "started_at": "2026-09-05T03:46:15Z", "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": 1421.7249000212178 }