Fix bugs (#4707)

Author: Bobholamovic

deploy/genai_vllm_server_docker/Dockerfile

@@ -13,9 +13,10 @@ RUN python -m pip install "paddlex${PADDLEX_VERSION}"
 
 ARG BUILD_FOR_SM120=false
 RUN if [ "${BUILD_FOR_SM120}" = 'true' ]; then \
-        python -m pip install https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.3.14/flash_attn-2.8.3+cu128torch2.8-cp310-cp310-linux_x86_64.whl \
+        python -m pip install https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.3.14/flash_attn-2.8.3+cu128torch2.8-cp310-cp310-linux_x86_64.whl; \
     else \
-        python -m pip install https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.3.14/flash_attn-2.8.2+cu128torch2.8-cp310-cp310-linux_x86_64.whl \
+        python -m pip install https://github.com/mjun0812/flash-attention-prebuild-wheels/releases/download/v0.3.14/flash_attn-2.8.2+cu128torch2.8-cp310-cp310-linux_x86_64.whl; \
+    fi \
     && paddlex --install genai-vllm-server
 
 EXPOSE 8080

paddlex/paddlex_cli.py

@@ -36,11 +36,7 @@
     is_dep_available,
     is_paddle2onnx_plugin_available,
 )
-from .utils.env import (
-    get_gpu_compute_capability,
-    get_paddle_cuda_version,
-    is_cuda_available,
-)
+from .utils.env import get_gpu_compute_capability, get_paddle_cuda_version
 from .utils.install import install_packages, uninstall_packages
 from .utils.interactive_get_pipeline import interactive_get_pipeline
 from .utils.pipeline_arguments import PIPELINE_ARGUMENTS
@@ -324,9 +320,6 @@ def _install_hpi_deps(device_type):
             )
 
     def _install_genai_deps(plugin_types):
-        if not is_cuda_available():
-            sys.exit("Currently, only GPU devices are supported.")
-
         fd_plugin_types = []
         not_fd_plugin_types = []
         for plugin_type in plugin_types: