{ "base_url": "http://127.0.0.1:39275/v1", "completed": 8, "concurrency": 4, "context_distribution": { "4096": 8 }, "context_seed": 0, "context_tokens": 4096, "engine": "sglang", "failed": 0, "failures": [], "finished_at": "2026-09-05T03:44:10Z", "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": 120, "content_chunks_s": null, "decode_tok_s_max": 114.3070330677872, "decode_tok_s_mean": 95.84106226056701, "decode_tok_s_min": 77.50580774529361, "decode_tok_s_p25": 85.34295676311379, "decode_tok_s_p50": 91.6398101018525, "decode_tok_s_p75": 102.65115164197843, "decode_tok_s_p90": 114.3070330677872, "decode_tok_s_p95": 114.3070330677872, "decode_tok_s_p99": 114.3070330677872, "decode_tok_s_samples": 8, "e2e_max_ms": 894.4666999159381, "e2e_mean_ms": 694.2706874833675, "e2e_min_ms": 583.2450999878347, "e2e_p25_ms": 599.2480999557301, "e2e_p50_ms": 618.5476999962702, "e2e_p75_ms": 730.9853000333533, "e2e_p90_ms": 894.4666999159381, "e2e_p95_ms": 894.4666999159381, "e2e_p99_ms": 894.4666999159381, "e2e_samples": 8, "effective_prefill_tok_s_max": 26052.219811135772, "effective_prefill_tok_s_mean": 22157.490376945065, "effective_prefill_tok_s_min": 19744.021734607577, "effective_prefill_tok_s_p25": 20917.098702104107, "effective_prefill_tok_s_p50": 21013.119688118124, "effective_prefill_tok_s_p75": 23014.116778311753, "effective_prefill_tok_s_p90": 26052.219811135772, "effective_prefill_tok_s_p95": 26052.219811135772, "effective_prefill_tok_s_p99": 26052.219811135772, "effective_prefill_tok_s_samples": 8, "generation_max_ms": 722.5265000015497, "generation_mean_ms": 530.1550875010435, "generation_min_ms": 411.1733000027016, "generation_p25_ms": 442.2575000207871, "generation_p50_ms": 457.86139997653663, "generation_p75_ms": 574.1540000308305, "generation_p90_ms": 722.5265000015497, "generation_p95_ms": 722.5265000015497, "generation_p99_ms": 722.5265000015497, "generation_samples": 8, "mean_inter_token_latency_ms_max": 12.902258928599101, "mean_inter_token_latency_ms_mean": 10.578113838919792, "mean_inter_token_latency_ms_min": 8.748368085163863, "mean_inter_token_latency_ms_p25": 9.48500425426012, "mean_inter_token_latency_ms_p50": 10.051306818654252, "mean_inter_token_latency_ms_p75": 11.066524139284317, "mean_inter_token_latency_ms_p90": 12.902258928599101, "mean_inter_token_latency_ms_p95": 12.902258928599101, "mean_inter_token_latency_ms_p99": 12.902258928599101, "mean_inter_token_latency_ms_samples": 8, "output_token_sources": { "usage": 8 }, "output_tokens": 406, "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": 242.1264041800427, "time_to_first_output_max_ms": 182.99210001714528, "time_to_first_output_mean_ms": 164.11559998232406, "time_to_first_output_min_ms": 138.68299999739975, "time_to_first_output_p25_ms": 156.83130000252277, "time_to_first_output_p50_ms": 160.6863000197336, "time_to_first_output_p75_ms": 172.0717999851331, "time_to_first_output_p90_ms": 182.99210001714528, "time_to_first_output_p95_ms": 182.99210001714528, "time_to_first_output_p99_ms": 182.99210001714528, "time_to_first_output_samples": 8, "tpot_max_ms": 12.902258928599101, "tpot_mean_ms": 10.578113838919792, "tpot_min_ms": 8.748368085163863, "tpot_p25_ms": 9.48500425426012, "tpot_p50_ms": 10.051306818654252, "tpot_p75_ms": 11.066524139284317, "tpot_p90_ms": 12.902258928599101, "tpot_p95_ms": 12.902258928599101, "tpot_p99_ms": 12.902258928599101, "tpot_samples": 8, "ttft_max_ms": 182.99240001942962, "ttft_mean_ms": 164.11591248470359, "ttft_min_ms": 138.68319999892265, "ttft_p25_ms": 156.83160000480711, "ttft_p50_ms": 160.68700002506375, "ttft_p75_ms": 172.0721999881789, "ttft_p90_ms": 182.99240001942962, "ttft_p95_ms": 182.99240001942962, "ttft_p99_ms": 182.99240001942962, "ttft_samples": 8 }, "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k12-chunk1k-lazy96", "request_timings": [ { "content_chunks": 14, "decode_tok_s": 105.4295784370215, "e2e_ms": 618.5246999375522, "effective_prefill_tok_s": 20917.098702104107, "generation_ms": 445.79519995022565, "mean_inter_token_latency_ms": 9.48500425426012, "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": 172.72949998732656, "tpot_ms": 9.48500425426012, "ttft_ms": 172.729599988088, "visible_generation_ms": 445.7950999494642 }, { "content_chunks": 14, "decode_tok_s": 102.65115164197843, "e2e_ms": 618.5476999962702, "effective_prefill_tok_s": 22484.804240039713, "generation_ms": 457.86139997653663, "mean_inter_token_latency_ms": 9.741731914394396, "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": 160.6863000197336, "tpot_ms": 9.741731914394396, "ttft_ms": 160.68700002506375, "visible_generation_ms": 457.8606999712065 }, { "content_chunks": 13, "decode_tok_s": 114.3070330677872, "e2e_ms": 583.2450999878347, "effective_prefill_tok_s": 20997.04890814277, "generation_ms": 411.1733000027016, "mean_inter_token_latency_ms": 8.748368085163863, "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": 172.0717999851331, "tpot_ms": 8.748368085163863, "ttft_ms": 172.0721999881789, "visible_generation_ms": 411.1728999996558 }, { "content_chunks": 18, "decode_tok_s": 77.50580774529361, "e2e_ms": 894.4666999159381, "effective_prefill_tok_s": 21013.119688118124, "generation_ms": 722.5265000015497, "mean_inter_token_latency_ms": 12.902258928599101, "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": 171.94019991438836, "tpot_ms": 12.902258928599101, "ttft_ms": 171.94039991591126, "visible_generation_ms": 722.5263000000268 }, { "content_chunks": 15, "decode_tok_s": 91.6398101018525, "e2e_ms": 728.6064999643713, "effective_prefill_tok_s": 19744.021734607577, "generation_ms": 545.614399947226, "mean_inter_token_latency_ms": 10.912287998944521, "output_token_source": "usage", "output_tokens": 51, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 4, "time_to_first_output_ms": 182.99210001714528, "tpot_ms": 10.912287998944521, "ttft_ms": 182.99240001942962, "visible_generation_ms": 545.6140999449417 }, { "content_chunks": 12, "decode_tok_s": 99.48955076608515, "e2e_ms": 599.2480999557301, "effective_prefill_tok_s": 23014.116778311753, "generation_ms": 442.2575000207871, "mean_inter_token_latency_ms": 10.051306818654252, "output_token_source": "usage", "output_tokens": 45, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 5, "time_to_first_output_ms": 156.99059993494302, "tpot_ms": 10.051306818654252, "ttft_ms": 156.99089993722737, "visible_generation_ms": 442.2572000185028 }, { "content_chunks": 15, "decode_tok_s": 85.34295676311379, "e2e_ms": 730.9853000333533, "effective_prefill_tok_s": 23037.4931531007, "generation_ms": 574.1540000308305, "mean_inter_token_latency_ms": 11.717428572057765, "output_token_source": "usage", "output_tokens": 50, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 6, "time_to_first_output_ms": 156.83130000252277, "tpot_ms": 11.717428572057765, "ttft_ms": 156.83160000480711, "visible_generation_ms": 574.1537000285462 }, { "content_chunks": 19, "decode_tok_s": 90.3626095614039, "e2e_ms": 780.5414000758901, "effective_prefill_tok_s": 26052.219811135772, "generation_ms": 641.8584000784904, "mean_inter_token_latency_ms": 11.066524139284317, "output_token_source": "usage", "output_tokens": 59, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 7, "time_to_first_output_ms": 138.68299999739975, "tpot_ms": 11.066524139284317, "ttft_ms": 138.68319999892265, "visible_generation_ms": 641.8582000769675 } ], "requests": 8, "run_id": "benchmark-20260905T034408Z", "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": "2026-09-05T03:44:08Z", "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": 1676.8100999761373 }