Disable FlashInfer autotuning for stable cold starts

entrypoint.sh

@@ -23,6 +23,8 @@ exec python3 -m vllm.entrypoints.openai.api_server \
   --max-num-batched-tokens 1024 \
   --kv-cache-dtype fp8 \
   --gpu-memory-utilization 0.95 \
+  --kernel-config '{"enable_jit_warmup":false,"enable_cutedsl_warmup":false}' \
+  --no-enable-flashinfer-autotune \
   --limit-mm-per-prompt '{"image":0,"video":0}' \
   --trust-remote-code \
   --enforce-eager \