{ "base_url": "http://127.0.0.1:39273/v1", "completed": 3, "concurrency": 1, "context_distribution": { "32768": 3 }, "context_seed": 0, "context_tokens": 32768, "engine": "sglang", "failed": 0, "failures": [], "finished_at": "2026-09-05T03:28:17Z", "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": 65, "content_chunks_s": null, "decode_tok_s_max": 94.60407354344403, "decode_tok_s_mean": 81.55738596581148, "decode_tok_s_min": 72.91187884342304, "decode_tok_s_p25": 72.91187884342304, "decode_tok_s_p50": 77.15620551056735, "decode_tok_s_p75": 94.60407354344403, "decode_tok_s_p90": 94.60407354344403, "decode_tok_s_p95": 94.60407354344403, "decode_tok_s_p99": 94.60407354344403, "decode_tok_s_samples": 3, "e2e_max_ms": 5047.080900054425, "e2e_mean_ms": 4770.781933368805, "e2e_min_ms": 4292.474000016227, "e2e_p25_ms": 4292.474000016227, "e2e_p50_ms": 4972.790900035761, "e2e_p75_ms": 5047.080900054425, "e2e_p90_ms": 5047.080900054425, "e2e_p95_ms": 5047.080900054425, "e2e_p99_ms": 5047.080900054425, "e2e_samples": 3, "effective_prefill_tok_s_max": 8604.419322368192, "effective_prefill_tok_s_mean": 7452.673397828244, "effective_prefill_tok_s_min": 6853.1962782576165, "effective_prefill_tok_s_p25": 6853.1962782576165, "effective_prefill_tok_s_p50": 6900.404592858924, "effective_prefill_tok_s_p75": 8604.419322368192, "effective_prefill_tok_s_p90": 8604.419322368192, "effective_prefill_tok_s_p95": 8604.419322368192, "effective_prefill_tok_s_p99": 8604.419322368192, "effective_prefill_tok_s_samples": 3, "generation_max_ms": 829.4861000031233, "generation_mean_ms": 728.1078333811214, "generation_min_ms": 655.3629001136869, "generation_p25_ms": 655.3629001136869, "generation_p50_ms": 699.4745000265539, "generation_p75_ms": 829.4861000031233, "generation_p90_ms": 829.4861000031233, "generation_p95_ms": 829.4861000031233, "generation_p99_ms": 829.4861000031233, "generation_samples": 3, "mean_inter_token_latency_ms_max": 13.715186275030469, "mean_inter_token_latency_ms_mean": 12.415425314750545, "mean_inter_token_latency_ms_min": 10.57036935667237, "mean_inter_token_latency_ms_p25": 10.57036935667237, "mean_inter_token_latency_ms_p50": 12.960720312548801, "mean_inter_token_latency_ms_p75": 13.715186275030469, "mean_inter_token_latency_ms_p90": 13.715186275030469, "mean_inter_token_latency_ms_p95": 13.715186275030469, "mean_inter_token_latency_ms_p99": 13.715186275030469, "mean_inter_token_latency_ms_samples": 3, "output_token_sources": { "usage": 3 }, "output_tokens": 180, "prefix_cache_hit_avg": null, "prompt_token_samples": 3, "prompt_tokens": 89384, "prompt_tokens_max": 29797.0, "prompt_tokens_mean": 29794.666666666668, "prompt_tokens_min": 29792.0, "prompt_tokens_p25": 29792.0, "prompt_tokens_p50": 29795.0, "prompt_tokens_p75": 29797.0, "prompt_tokens_p90": 29797.0, "prompt_tokens_p95": 29797.0, "prompt_tokens_p99": 29797.0, "prompt_tokens_samples": 3, "reasoning_chunks": 0, "throughput_tok_s": 12.574507361679707, "time_to_first_output_max_ms": 4347.606400027871, "time_to_first_output_mean_ms": 4042.6740999876833, "time_to_first_output_min_ms": 3462.987900013104, "time_to_first_output_p25_ms": 3462.987900013104, "time_to_first_output_p50_ms": 4317.427999922074, "time_to_first_output_p75_ms": 4347.606400027871, "time_to_first_output_p90_ms": 4347.606400027871, "time_to_first_output_p95_ms": 4347.606400027871, "time_to_first_output_p99_ms": 4347.606400027871, "time_to_first_output_samples": 3, "tpot_max_ms": 13.715186275030469, "tpot_mean_ms": 12.415425314750545, "tpot_min_ms": 10.57036935667237, "tpot_p25_ms": 10.57036935667237, "tpot_p50_ms": 12.960720312548801, "tpot_p75_ms": 13.715186275030469, "tpot_p90_ms": 13.715186275030469, "tpot_p95_ms": 13.715186275030469, "tpot_p99_ms": 13.715186275030469, "tpot_samples": 3, "ttft_max_ms": 4347.607200033963, "ttft_mean_ms": 4042.6748333266005, "ttft_min_ms": 3462.9884000169113, "ttft_p25_ms": 3462.9884000169113, "ttft_p50_ms": 4317.428899928927, "ttft_p75_ms": 4347.607200033963, "ttft_p90_ms": 4347.607200033963, "ttft_p95_ms": 4347.607200033963, "ttft_p99_ms": 4347.607200033963, "ttft_samples": 3 }, "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k12-chunk8k", "request_timings": [ { "content_chunks": 20, "decode_tok_s": 94.60407354344403, "e2e_ms": 4972.790900035761, "effective_prefill_tok_s": 6900.404592858924, "generation_ms": 655.3629001136869, "mean_inter_token_latency_ms": 10.57036935667237, "output_token_source": "usage", "output_tokens": 63, "planned_context_tokens": 32768, "prompt_tokens": 29792, "reasoning_chunks": 0, "request_index": 0, "time_to_first_output_ms": 4317.427999922074, "tpot_ms": 10.57036935667237, "ttft_ms": 4317.428899928927, "visible_generation_ms": 655.3620001068339 }, { "content_chunks": 21, "decode_tok_s": 72.91187884342304, "e2e_ms": 5047.080900054425, "effective_prefill_tok_s": 6853.1962782576165, "generation_ms": 699.4745000265539, "mean_inter_token_latency_ms": 13.715186275030469, "output_token_source": "usage", "output_tokens": 52, "planned_context_tokens": 32768, "prompt_tokens": 29795, "reasoning_chunks": 0, "request_index": 1, "time_to_first_output_ms": 4347.606400027871, "tpot_ms": 13.715186275030469, "ttft_ms": 4347.607200033963, "visible_generation_ms": 699.4737000204623 }, { "content_chunks": 24, "decode_tok_s": 77.15620551056735, "e2e_ms": 4292.474000016227, "effective_prefill_tok_s": 8604.419322368192, "generation_ms": 829.4861000031233, "mean_inter_token_latency_ms": 12.960720312548801, "output_token_source": "usage", "output_tokens": 65, "planned_context_tokens": 32768, "prompt_tokens": 29797, "reasoning_chunks": 0, "request_index": 2, "time_to_first_output_ms": 3462.987900013104, "tpot_ms": 12.960720312548801, "ttft_ms": 3462.9884000169113, "visible_generation_ms": 829.485599999316 } ], "requests": 3, "run_id": "benchmark-20260905T032803Z", "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": "2026-09-05T03:28:03Z", "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": 14314.676100038923 }