Add GB200 FP8 8k prefill batch limit

jasonlizhengjian · jasonlizhengjian · commit f601cfed3cef · 2026-06-02T14:17:45.000-07:00
diff --git a/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m2.5-gb200-fp8/8k1k/disagg-gb200-1p1d-tp4ep.yaml b/benchmarks/multi_node/srt-slurm-recipes/vllm/minimax-m2.5-gb200-fp8/8k1k/disagg-gb200-1p1d-tp4ep.yaml
@@ -44,6 +44,7 @@ backend:
       tensor-parallel-size: 1
       pipeline-parallel-size: 1
       data-parallel-size: 2
+      max-num-batched-tokens: 16384
       data-parallel-rpc-port: 13346
       enable-expert-parallel: true
       safetensors-load-strategy: "prefetch"
