cjw/rocm_vllm_nightly
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

x

Browse Source
This commit is contained in:
SZCJW
2026-04-05 02:25:25 +08:00
parent fd579f68ba
commit 549cc05603
2 changed files with 3 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Dockerfile
+1 -1
View File
@@ -1,4 +1,4 @@
FROM vllm/vllm-openai-rocm:latest FROM vllm/vllm-openai-rocm:nightly
WORKDIR /workspace WORKDIR /workspace
docker-compose.yml
+2 -2
View File
@@ -3,7 +3,7 @@ services:
build: build:
context: . context: .
dockerfile: Dockerfile dockerfile: Dockerfile
image: rocm-vllm-inference:latest image: rocm-vllm-inference:nightly
container_name: rocm-vllm-openai container_name: rocm-vllm-openai
entrypoint: ["python"] entrypoint: ["python"]
command: ["-m", "app.start_openai"] command: ["-m", "app.start_openai"]
@@ -23,7 +23,7 @@ services:
VLLM_WORKER_MULTIPROC_METHOD: "spawn" VLLM_WORKER_MULTIPROC_METHOD: "spawn"
PYTHONUNBUFFERED: "1" PYTHONUNBUFFERED: "1"
# 启用 AITER 加速（R9700 属于 gfx1201 架构） # 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" VLLM_ROCM_USE_AITER_MHA: "0"
# 增强 All-Reduce 性能（双卡通信优化） # 增强 All-Reduce 性能（双卡通信优化）
VLLM_ROCM_QUICK_REDUCE_QUANTIZATION: "INT4" VLLM_ROCM_QUICK_REDUCE_QUANTIZATION: "INT4"