== vLLM NVFP4 (max-num-seqs 8) (qwen38-nvfp4) warmup...
  N=1  agg=  56.9 tok/s  per-req= 56.9 tok/s  p50=13.91s p95=13.91s  (791 tok/13.91s)
  N=2  agg= 107.6 tok/s  per-req= 53.8 tok/s  p50=13.23s p95=13.23s  (1424 tok/13.23s)
  N=4  agg= 208.5 tok/s  per-req= 52.1 tok/s  p50=13.66s p95=13.66s  (2848 tok/13.66s)
  N=8  agg= 436.2 tok/s  per-req= 54.5 tok/s  p50=12.56s p95=12.56s  (5480 tok/12.56s)
JSON {"label": "vLLM NVFP4 (max-num-seqs 8)", "model": "qwen38-nvfp4", "rows": [{"n": 1, "wall_s": 13.91, "total_tok": 791, "agg_tok_s": 56.9, "per_req_tok_s": 56.9, "p50_lat_s": 13.91, "p95_lat_s": 13.91}, {"n": 2, "wall_s": 13.23, "total_tok": 1424, "agg_tok_s": 107.6, "per_req_tok_s": 53.8, "p50_lat_s": 13.23, "p95_lat_s": 13.23}, {"n": 4, "wall_s": 13.66, "total_tok": 2848, "agg_tok_s": 208.5, "per_req_tok_s": 52.1, "p50_lat_s": 13.66, "p95_lat_s": 13.66}, {"n": 8, "wall_s": 12.56, "total_tok": 5480, "agg_tok_s": 436.2, "per_req_tok_s": 54.5, "p50_lat_s": 12.56, "p95_lat_s": 12.56}]}