使用镜像:harbor.sourcefind.cn:5443/dcu/admin/base/custom:vllm018-ubuntu22.04-dtk26.04-gemma4-0413
启动参数:
```Shell
export HIP_LAUNCH_BLOCKING=0 && export HIP_DEVICE_MAX_HW_QUEUES=8 && export HSA_FORCE_FINE_GRAIN_PCIE=1 && export GLOO_SOCKET_IFNAME=lo && export NCCL_SOCKET_IFNAME=lo && export NCCL_ALGO=Ring && export NCCL_PROTO=Simple && export PYTORCH_HIP_ALLOC_CONF=expandable_segments:True
vllm serve /workspace/Qwen3-32B --port 11435 --trust-remote-code --max-model-len 32768 -tp 8 --gpu-memory-utilization 0.5 --max-num-seqs 16 --disable-custom-all-reduce --enable-reasoning --reasoning-parser deepseek_r1 --tool-call-parser hermes --enable-auto-tool-choice --use-v2-block-manager > vllm_runtime.log 2>&1 &
```
手动测试结果:

首词输出70s,这个问题怎么修复呀?