{ "base_url": "http://127.0.0.1:39272/v1", "completed": 5, "concurrency": 1, "context_distribution": { "4096": 5 }, "context_seed": 0, "context_tokens": 4096, "engine": "sglang", "failed": 0, "failures": [], "finished_at": "2026-09-05T03:25:35Z", "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": 67, "content_chunks_s": null, "decode_tok_s_max": 120.83523370101383, "decode_tok_s_mean": 111.72003641265482, "decode_tok_s_min": 101.18879232412495, "decode_tok_s_p25": 106.58935385490052, "decode_tok_s_p50": 111.19175800853768, "decode_tok_s_p75": 118.79504417469711, "decode_tok_s_p90": 120.83523370101383, "decode_tok_s_p95": 120.83523370101383, "decode_tok_s_p99": 120.83523370101383, "decode_tok_s_samples": 5, "e2e_max_ms": 579.068300081417, "e2e_mean_ms": 527.7396599994972, "e2e_min_ms": 498.49749996792525, "e2e_p25_ms": 502.7095000259578, "e2e_p50_ms": 527.22059993539, "e2e_p75_ms": 531.2023999867961, "e2e_p90_ms": 579.068300081417, "e2e_p95_ms": 579.068300081417, "e2e_p99_ms": 579.068300081417, "e2e_samples": 5, "effective_prefill_tok_s_max": 34565.09967997865, "effective_prefill_tok_s_mean": 32692.01582534096, "effective_prefill_tok_s_min": 29026.491781876535, "effective_prefill_tok_s_p25": 32983.95719183238, "effective_prefill_tok_s_p50": 33140.28255457062, "effective_prefill_tok_s_p75": 33744.24791844663, "effective_prefill_tok_s_p90": 34565.09967997865, "effective_prefill_tok_s_p95": 34565.09967997865, "effective_prefill_tok_s_p99": 34565.09967997865, "effective_prefill_tok_s_samples": 5, "generation_max_ms": 454.59580002352595, "generation_mean_ms": 416.8137599946931, "generation_min_ms": 388.9594000065699, "generation_p25_ms": 395.63939999789, "generation_p50_ms": 422.18100000172853, "generation_p75_ms": 422.6931999437511, "generation_p90_ms": 454.59580002352595, "generation_p95_ms": 454.59580002352595, "generation_p99_ms": 454.59580002352595, "generation_samples": 5, "mean_inter_token_latency_ms_max": 9.882517391815782, "mean_inter_token_latency_ms_mean": 8.990276244107907, "mean_inter_token_latency_ms_min": 8.275731915033402, "mean_inter_token_latency_ms_p25": 8.417859574423192, "mean_inter_token_latency_ms_p50": 8.993472339228747, "mean_inter_token_latency_ms_p75": 9.381800000038412, "mean_inter_token_latency_ms_p90": 9.882517391815782, "mean_inter_token_latency_ms_p95": 9.882517391815782, "mean_inter_token_latency_ms_p99": 9.882517391815782, "mean_inter_token_latency_ms_samples": 5, "output_token_sources": { "usage": 5 }, "output_tokens": 237, "prefix_cache_hit_avg": null, "prompt_token_samples": 5, "prompt_tokens": 18065, "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": 5, "reasoning_chunks": 0, "throughput_tok_s": 89.77813436601868, "time_to_first_output_max_ms": 124.47250005789101, "time_to_first_output_mean_ms": 110.92590000480413, "time_to_first_output_min_ms": 104.52739999163896, "time_to_first_output_p25_ms": 107.07010002806783, "time_to_first_output_p50_ms": 109.02139998506755, "time_to_first_output_p75_ms": 109.53809996135533, "time_to_first_output_p90_ms": 124.47250005789101, "time_to_first_output_p95_ms": 124.47250005789101, "time_to_first_output_p99_ms": 124.47250005789101, "time_to_first_output_samples": 5, "tpot_max_ms": 9.882517391815782, "tpot_mean_ms": 8.990276244107907, "tpot_min_ms": 8.275731915033402, "tpot_p25_ms": 8.417859574423192, "tpot_p50_ms": 8.993472339228747, "tpot_p75_ms": 9.381800000038412, "tpot_p90_ms": 9.882517391815782, "tpot_p95_ms": 9.882517391815782, "tpot_p99_ms": 9.882517391815782, "tpot_samples": 5, "ttft_max_ms": 124.47280006017536, "ttft_mean_ms": 110.92619998380542, "ttft_min_ms": 104.52759999316186, "ttft_p25_ms": 107.07039991393685, "ttft_p50_ms": 109.02159998659045, "ttft_p75_ms": 109.53859996516258, "ttft_p90_ms": 124.47280006017536, "ttft_p95_ms": 124.47280006017536, "ttft_p99_ms": 124.47280006017536, "ttft_samples": 5 }, "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k12-chunk1k", "request_timings": [ { "content_chunks": 15, "decode_tok_s": 101.18879232412495, "e2e_ms": 579.068300081417, "effective_prefill_tok_s": 29026.491781876535, "generation_ms": 454.59580002352595, "mean_inter_token_latency_ms": 9.882517391815782, "output_token_source": "usage", "output_tokens": 47, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 0, "time_to_first_output_ms": 124.47250005789101, "tpot_ms": 9.882517391815782, "ttft_ms": 124.47280006017536, "visible_generation_ms": 454.5955000212416 }, { "content_chunks": 13, "decode_tok_s": 120.83523370101383, "e2e_ms": 498.49749996792525, "effective_prefill_tok_s": 32983.95719183238, "generation_ms": 388.9594000065699, "mean_inter_token_latency_ms": 8.275731915033402, "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": 109.53809996135533, "tpot_ms": 8.275731915033402, "ttft_ms": 109.53859996516258, "visible_generation_ms": 388.9589000027627 }, { "content_chunks": 13, "decode_tok_s": 106.58935385490052, "e2e_ms": 531.2023999867961, "effective_prefill_tok_s": 33140.28255457062, "generation_ms": 422.18100000172853, "mean_inter_token_latency_ms": 9.381800000038412, "output_token_source": "usage", "output_tokens": 46, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 2, "time_to_first_output_ms": 109.02139998506755, "tpot_ms": 9.381800000038412, "ttft_ms": 109.02159998659045, "visible_generation_ms": 422.18080000020564 }, { "content_chunks": 13, "decode_tok_s": 111.19175800853768, "e2e_ms": 527.22059993539, "effective_prefill_tok_s": 34565.09967997865, "generation_ms": 422.6931999437511, "mean_inter_token_latency_ms": 8.993472339228747, "output_token_source": "usage", "output_tokens": 48, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 3, "time_to_first_output_ms": 104.52739999163896, "tpot_ms": 8.993472339228747, "ttft_ms": 104.52759999316186, "visible_generation_ms": 422.6929999422282 }, { "content_chunks": 13, "decode_tok_s": 118.79504417469711, "e2e_ms": 502.7095000259578, "effective_prefill_tok_s": 33744.24791844663, "generation_ms": 395.63939999789, "mean_inter_token_latency_ms": 8.417859574423192, "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": 107.07010002806783, "tpot_ms": 8.417859574423192, "ttft_ms": 107.07039991393685, "visible_generation_ms": 395.63910011202097 } ], "requests": 5, "run_id": "benchmark-20260905T032532Z", "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": "2026-09-05T03:25:32Z", "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": 2639.8409999674186 }