x

Dockerfile

@@ -1,4 +1,4 @@
-FROM vllm/vllm-openai-rocm:latest
+FROM vllm/vllm-openai-rocm:nightly
 
 WORKDIR /workspace
 

docker-compose.yml

@@ -3,7 +3,7 @@ services:
     build:
       context: .
       dockerfile: Dockerfile
-    image: rocm-vllm-inference:latest
+    image: rocm-vllm-inference:nightly
     container_name: rocm-vllm-openai
     entrypoint: ["python"]
     command: ["-m", "app.start_openai"]
@@ -23,7 +23,7 @@ services:
       VLLM_WORKER_MULTIPROC_METHOD: "spawn"
       PYTHONUNBUFFERED: "1"
       # 启用 AITER 加速（R9700 属于 gfx1201 架构）
-      VLLM_ROCM_USE_AITER: "1"
+      VLLM_ROCM_USE_AITER: "0"
       VLLM_ROCM_USE_AITER_MHA: "0"
       # 增强 All-Reduce 性能（双卡通信优化）
       VLLM_ROCM_QUICK_REDUCE_QUANTIZATION: "INT4"