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
2026-04-16 00:15:55 +08:00
parent 101ccb3553
commit b12c6e6ec0
1 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files
docker-compose.yml
+4 -4
View File
@@ -3,7 +3,7 @@ services:
build: build:
context: . context: .
dockerfile: Dockerfile dockerfile: Dockerfile
image: rocm-vllm-inference:nightly image: vllm-openai-rocm: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"]
@@ -23,10 +23,10 @@ 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: "0" VLLM_ROCM_USE_AITER: "1"
# 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"
devices: devices:
- /dev/kfd - /dev/kfd
- /dev/dri - /dev/dri