{ "base_url": "http://127.0.0.1:39274/v1", "completed": 5, "concurrency": 1, "context_distribution": { "4096": 5 }, "context_seed": 0, "context_tokens": 4096, "engine": "sglang", "failed": 0, "failures": [], "finished_at": "2026-09-05T03:37:37Z", "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": 75, "content_chunks_s": null, "decode_tok_s_max": 122.35596408290178, "decode_tok_s_mean": 115.28376732490435, "decode_tok_s_min": 108.13382641663625, "decode_tok_s_p25": 111.84214449495036, "decode_tok_s_p50": 112.59413288229996, "decode_tok_s_p75": 121.49276874773341, "decode_tok_s_p90": 122.35596408290178, "decode_tok_s_p95": 122.35596408290178, "decode_tok_s_p99": 122.35596408290178, "decode_tok_s_samples": 5, "e2e_max_ms": 638.8129000551999, "e2e_mean_ms": 566.1494000349194, "e2e_min_ms": 475.3664000891149, "e2e_p25_ms": 536.4447999745607, "e2e_p50_ms": 580.1973000634462, "e2e_p75_ms": 599.9255999922752, "e2e_p90_ms": 638.8129000551999, "e2e_p95_ms": 638.8129000551999, "e2e_p99_ms": 638.8129000551999, "e2e_samples": 5, "effective_prefill_tok_s_max": 35303.026601119156, "effective_prefill_tok_s_mean": 28067.054233994753, "effective_prefill_tok_s_min": 16956.76517146871, "effective_prefill_tok_s_p25": 26507.139606523353, "effective_prefill_tok_s_p50": 30357.1863799284, "effective_prefill_tok_s_p75": 31211.153410934163, "effective_prefill_tok_s_p90": 35303.026601119156, "effective_prefill_tok_s_p95": 35303.026601119156, "effective_prefill_tok_s_p99": 35303.026601119156, "effective_prefill_tok_s_samples": 5, "generation_max_ms": 536.4704000530764, "generation_mean_ms": 428.8508200086653, "generation_min_ms": 359.6065000165254, "generation_p25_ms": 386.854299926199, "generation_p50_ms": 417.4285000190139, "generation_p75_ms": 443.8944000285119, "generation_p90_ms": 536.4704000530764, "generation_p95_ms": 536.4704000530764, "generation_p99_ms": 536.4704000530764, "generation_samples": 5, "mean_inter_token_latency_ms_max": 9.247800000593998, "mean_inter_token_latency_ms_mean": 8.694849666804366, "mean_inter_token_latency_ms_min": 8.172875000375576, "mean_inter_token_latency_ms_p25": 8.230942551621256, "mean_inter_token_latency_ms_p50": 8.88145744721306, "mean_inter_token_latency_ms_p75": 8.941173334217941, "mean_inter_token_latency_ms_p90": 9.247800000593998, "mean_inter_token_latency_ms_p95": 9.247800000593998, "mean_inter_token_latency_ms_p99": 9.247800000593998, "mean_inter_token_latency_ms_samples": 5, "output_token_sources": { "usage": 5 }, "output_tokens": 251, "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": 88.62816906112839, "time_to_first_output_max_ms": 213.07130006607622, "time_to_first_output_mean_ms": 137.29858002625406, "time_to_first_output_min_ms": 102.34250000212342, "time_to_first_output_p25_ms": 115.75990007258952, "time_to_first_output_p50_ms": 119.01629995554686, "time_to_first_output_p75_ms": 136.30290003493428, "time_to_first_output_p90_ms": 213.07130006607622, "time_to_first_output_p95_ms": 213.07130006607622, "time_to_first_output_p99_ms": 213.07130006607622, "time_to_first_output_samples": 5, "tpot_max_ms": 9.247800000593998, "tpot_mean_ms": 8.694849666804366, "tpot_min_ms": 8.172875000375576, "tpot_p25_ms": 8.230942551621256, "tpot_p50_ms": 8.88145744721306, "tpot_p75_ms": 8.941173334217941, "tpot_p90_ms": 9.247800000593998, "tpot_p95_ms": 9.247800000593998, "tpot_p99_ms": 9.247800000593998, "tpot_samples": 5, "ttft_max_ms": 213.07200007140636, "ttft_mean_ms": 137.2990800300613, "ttft_min_ms": 102.34270000364631, "ttft_p25_ms": 115.76020007487386, "ttft_p50_ms": 119.016999960877, "ttft_p75_ms": 136.30350003950298, "ttft_p90_ms": 213.07200007140636, "ttft_p95_ms": 213.07200007140636, "ttft_p99_ms": 213.07200007140636, "ttft_samples": 5 }, "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k12-compile", "request_timings": [ { "content_chunks": 14, "decode_tok_s": 121.49276874773341, "e2e_ms": 599.9255999922752, "effective_prefill_tok_s": 16956.76517146871, "generation_ms": 386.854299926199, "mean_inter_token_latency_ms": 8.230942551621256, "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": 213.07130006607622, "tpot_ms": 8.230942551621256, "ttft_ms": 213.07200007140636, "visible_generation_ms": 386.8535999208689 }, { "content_chunks": 15, "decode_tok_s": 108.13382641663625, "e2e_ms": 580.1973000634462, "effective_prefill_tok_s": 26507.139606523353, "generation_ms": 443.8944000285119, "mean_inter_token_latency_ms": 9.247800000593998, "output_token_source": "usage", "output_tokens": 49, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 1, "time_to_first_output_ms": 136.30290003493428, "tpot_ms": 9.247800000593998, "ttft_ms": 136.30350003950298, "visible_generation_ms": 443.8938000239432 }, { "content_chunks": 19, "decode_tok_s": 111.84214449495036, "e2e_ms": 638.8129000551999, "effective_prefill_tok_s": 35303.026601119156, "generation_ms": 536.4704000530764, "mean_inter_token_latency_ms": 8.941173334217941, "output_token_source": "usage", "output_tokens": 61, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 2, "time_to_first_output_ms": 102.34250000212342, "tpot_ms": 8.941173334217941, "ttft_ms": 102.34270000364631, "visible_generation_ms": 536.4702000515535 }, { "content_chunks": 15, "decode_tok_s": 112.59413288229996, "e2e_ms": 536.4447999745607, "effective_prefill_tok_s": 30357.1863799284, "generation_ms": 417.4285000190139, "mean_inter_token_latency_ms": 8.88145744721306, "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": 119.01629995554686, "tpot_ms": 8.88145744721306, "ttft_ms": 119.016999960877, "visible_generation_ms": 417.42780001368374 }, { "content_chunks": 12, "decode_tok_s": 122.35596408290178, "e2e_ms": 475.3664000891149, "effective_prefill_tok_s": 31211.153410934163, "generation_ms": 359.6065000165254, "mean_inter_token_latency_ms": 8.172875000375576, "output_token_source": "usage", "output_tokens": 45, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "request_index": 4, "time_to_first_output_ms": 115.75990007258952, "tpot_ms": 8.172875000375576, "ttft_ms": 115.76020007487386, "visible_generation_ms": 359.60620001424104 } ], "requests": 5, "run_id": "benchmark-20260905T033734Z", "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": "2026-09-05T03:37:34Z", "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": 2832.056700019166 }