x

docker/Dockerfile

@@ -225,12 +225,12 @@ RUN set -ex && \
     # --no-build-isolation 可能漏装部分运行时依赖，显式补装
     ${VENV}/bin/pip install --no-cache-dir regex && \
     # ---- 补丁 6：vllm 平台检测 —— torch.version.hip 回退 ----
-    ${VENV}/bin/python -c "
+    cat > /tmp/patch6.py << 'PYEOF'
 import re
 f = '/opt/vllm/vllm/platforms/__init__.py'
 c = open(f).read()
-old = \"    return 'vllm.platforms.rocm.RocmPlatform' if is_rocm else None\"
+old = "    return 'vllm.platforms.rocm.RocmPlatform' if is_rocm else None"
-new = '''    # amdsmi fallback: also check torch.version.hip
+new = """    # amdsmi fallback: also check torch.version.hip
     if not is_rocm:
         try:
             import torch
@@ -238,28 +238,24 @@ new = '''    # amdsmi fallback: also check torch.version.hip
                 is_rocm = True
         except Exception:
             pass
-    return 'vllm.platforms.rocm.RocmPlatform' if is_rocm else None'''
+    return 'vllm.platforms.rocm.RocmPlatform' if is_rocm else None"""
 c2 = c.replace(old, new)
-if c2 != c:
+if c2 != c: open(f, 'w').write(c2); print('Patch 6 applied.')
-    open(f, 'w').write(c2)
+else: print('Patch 6 already applied.')
-    print('Patch 6: __init__.py platform fallback applied.')
+PYEOF
-else:
+    ${VENV}/bin/python /tmp/patch6.py && \
-    print('Patch 6: already applied or pattern not found.')
-" && \
     # ---- 补丁 7：vllm rocm.py 断开循环导入 ----
-    ${VENV}/bin/python -c "
+    cat > /tmp/patch7.py << 'PYEOF'
 f = '/opt/vllm/vllm/platforms/rocm.py'
 c = open(f).read()
-# 替换 logger.warning_once( 为 sys.stderr.write(
 old = 'logger.warning_once('
 new = 'import sys as _sys\n        _sys.stderr.write('
 c2 = c.replace(old, new)
-if c2 != c:
+if c2 != c: open(f, 'w').write(c2); print('Patch 7 applied.')
-    open(f, 'w').write(c2)
+else: print('Patch 7 already applied.')
-    print('Patch 7: rocm.py logger.warning_once → sys.stderr.write applied.')
+PYEOF
-else:
+    ${VENV}/bin/python /tmp/patch7.py && \
-    print('Patch 7: already applied or pattern not found.')
+    rm -f /tmp/patch6.py /tmp/patch7.py && \
-" && \
     # 验证 PyTorch 没被 vllm 依赖覆盖
     ${VENV}/bin/python -c "import torch; v=torch.__version__; assert 'rocm' in v, f'PyTorch overwritten by vllm deps: {v}'; print('PyTorch OK:', v)" && \
     # 先重装 ROCm PyTorch 覆盖可能的 CUDA 版，再清理 CUDA triton 元数据