{ "base_url": "http://127.0.0.1:39272/v1", "completed": 3, "concurrency": 1, "context_distribution": { "32768": 3 }, "context_seed": 0, "context_tokens": 32768, "engine": "sglang", "failed": 0, "failures": [], "finished_at": "2026-09-05T03:25:31Z", "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": 60, "content_chunks_s": null, "decode_tok_s_max": 98.16805444473712, "decode_tok_s_mean": 91.67635072610749, "decode_tok_s_min": 87.91916098710153, "decode_tok_s_p25": 87.91916098710153, "decode_tok_s_p50": 88.9418367464838, "decode_tok_s_p75": 98.16805444473712, "decode_tok_s_p90": 98.16805444473712, "decode_tok_s_p95": 98.16805444473712, "decode_tok_s_p99": 98.16805444473712, "decode_tok_s_samples": 3, "e2e_max_ms": 5385.254899971187, "e2e_mean_ms": 4843.843533308245, "e2e_min_ms": 4276.954400003888, "e2e_p25_ms": 4276.954400003888, "e2e_p50_ms": 4869.321299949661, "e2e_p75_ms": 5385.254899971187, "e2e_p90_ms": 5385.254899971187, "e2e_p95_ms": 5385.254899971187, "e2e_p99_ms": 5385.254899971187, "e2e_samples": 3, "effective_prefill_tok_s_max": 7694.601163473137, "effective_prefill_tok_s_mean": 7166.930558809923, "effective_prefill_tok_s_min": 6148.557165337271, "effective_prefill_tok_s_p25": 6148.557165337271, "effective_prefill_tok_s_p50": 7657.633347619362, "effective_prefill_tok_s_p75": 7694.601163473137, "effective_prefill_tok_s_p90": 7694.601163473137, "effective_prefill_tok_s_p95": 7694.601163473137, "effective_prefill_tok_s_p99": 7694.601163473137, "effective_prefill_tok_s_samples": 3, "generation_max_ms": 978.1713000265881, "generation_mean_ms": 640.9402333277588, "generation_min_ms": 404.75889993831515, "generation_p25_ms": 404.75889993831515, "generation_p50_ms": 539.8905000183731, "generation_p75_ms": 978.1713000265881, "generation_p90_ms": 978.1713000265881, "generation_p95_ms": 978.1713000265881, "generation_p99_ms": 978.1713000265881, "generation_samples": 3, "mean_inter_token_latency_ms_max": 11.374084884030095, "mean_inter_token_latency_ms_mean": 10.934666955996079, "mean_inter_token_latency_ms_min": 10.186613207893833, "mean_inter_token_latency_ms_p25": 10.186613207893833, "mean_inter_token_latency_ms_p50": 11.24330277606431, "mean_inter_token_latency_ms_p75": 11.374084884030095, "mean_inter_token_latency_ms_p90": 11.374084884030095, "mean_inter_token_latency_ms_p95": 11.374084884030095, "mean_inter_token_latency_ms_p99": 11.374084884030095, "mean_inter_token_latency_ms_samples": 3, "output_token_sources": { "usage": 3 }, "output_tokens": 178, "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": 12.247491472929783, "time_to_first_output_max_ms": 4845.364399952814, "time_to_first_output_mean_ms": 4202.903299980487, "time_to_first_output_min_ms": 3872.1955000655726, "time_to_first_output_p25_ms": 3872.1955000655726, "time_to_first_output_p50_ms": 3891.1499999230728, "time_to_first_output_p75_ms": 4845.364399952814, "time_to_first_output_p90_ms": 4845.364399952814, "time_to_first_output_p95_ms": 4845.364399952814, "time_to_first_output_p99_ms": 4845.364399952814, "time_to_first_output_samples": 3, "tpot_max_ms": 11.374084884030095, "tpot_mean_ms": 10.934666955996079, "tpot_min_ms": 10.186613207893833, "tpot_p25_ms": 10.186613207893833, "tpot_p50_ms": 11.24330277606431, "tpot_p75_ms": 11.374084884030095, "tpot_p90_ms": 11.374084884030095, "tpot_p95_ms": 11.374084884030095, "tpot_p99_ms": 11.374084884030095, "tpot_samples": 3, "ttft_max_ms": 4845.36479995586, "ttft_mean_ms": 4202.9037666507065, "ttft_min_ms": 3872.1961000701413, "ttft_p25_ms": 3872.1961000701413, "ttft_p50_ms": 3891.1503999261186, "ttft_p75_ms": 4845.36479995586, "ttft_p90_ms": 4845.36479995586, "ttft_p95_ms": 4845.36479995586, "ttft_p99_ms": 4845.36479995586, "ttft_samples": 3 }, "model": "qwen38-27b-inferact-nvfp4-pro6000-dflash2-k12-chunk1k", "request_timings": [ { "content_chunks": 17, "decode_tok_s": 98.16805444473712, "e2e_ms": 5385.254899971187, "effective_prefill_tok_s": 6148.557165337271, "generation_ms": 539.8905000183731, "mean_inter_token_latency_ms": 10.186613207893833, "output_token_source": "usage", "output_tokens": 54, "planned_context_tokens": 32768, "prompt_tokens": 29792, "reasoning_chunks": 0, "request_index": 0, "time_to_first_output_ms": 4845.364399952814, "tpot_ms": 10.186613207893833, "ttft_ms": 4845.36479995586, "visible_generation_ms": 539.8901000153273 }, { "content_chunks": 13, "decode_tok_s": 88.9418367464838, "e2e_ms": 4276.954400003888, "effective_prefill_tok_s": 7694.601163473137, "generation_ms": 404.75889993831515, "mean_inter_token_latency_ms": 11.24330277606431, "output_token_source": "usage", "output_tokens": 37, "planned_context_tokens": 32768, "prompt_tokens": 29795, "reasoning_chunks": 0, "request_index": 1, "time_to_first_output_ms": 3872.1955000655726, "tpot_ms": 11.24330277606431, "ttft_ms": 3872.1961000701413, "visible_generation_ms": 404.75829993374646 }, { "content_chunks": 30, "decode_tok_s": 87.91916098710153, "e2e_ms": 4869.321299949661, "effective_prefill_tok_s": 7657.633347619362, "generation_ms": 978.1713000265881, "mean_inter_token_latency_ms": 11.374084884030095, "output_token_source": "usage", "output_tokens": 87, "planned_context_tokens": 32768, "prompt_tokens": 29797, "reasoning_chunks": 0, "request_index": 2, "time_to_first_output_ms": 3891.1499999230728, "tpot_ms": 11.374084884030095, "ttft_ms": 3891.1503999261186, "visible_generation_ms": 978.1709000235423 } ], "requests": 3, "run_id": "benchmark-20260905T032517Z", "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": "2026-09-05T03:25:17Z", "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": 14533.588399994187 }