[Misc][XPU] Avoid torch compile for XPU platform (#10747)

Signed-off-by: yan ma <[email protected]>
Co-authored-by: youkaichao <[email protected]>

.buildkite/run-xpu-test.sh

@@ -12,5 +12,7 @@ remove_docker_container() { docker rm -f xpu-test || true; }
 trap remove_docker_container EXIT
 remove_docker_container
 
-# Run the image and launch offline inference
-docker run --network host --name xpu-test --device /dev/dri -v /dev/dri/by-path:/dev/dri/by-path --entrypoint="" xpu-test python3 examples/offline_inference.py
+# Run the image and test offline inference/tensor parallel
+docker run -it -d --name xpu-test --device /dev/dri -v /dev/dri/by-path:/dev/dri/by-path xpu-test /bin/bash
+docker exec xpu-test bash -c "python3 examples/offline_inference.py"
+docker exec xpu-test bash -c "python3 examples/offline_inference_cli.py -tp 2" 

vllm/plugins/__init__.py

@@ -4,6 +4,7 @@
 import torch
 
 import vllm.envs as envs
+from vllm.platforms import current_platform
 
 logger = logging.getLogger(__name__)
 
@@ -25,6 +26,9 @@ def load_general_plugins():
     os.environ['TORCHINDUCTOR_COMPILE_THREADS'] = '1'
     # see https://github.com/vllm-project/vllm/issues/10619
     torch._inductor.config.compile_threads = 1
+    if current_platform.is_xpu():
+        # see https://github.com/pytorch/pytorch/blob/8cada5cbe5450e17c26fb8b358116785324537b2/torch/_dynamo/config.py#L158  # noqa
+        os.environ['TORCH_COMPILE_DISABLE'] = 'True'
     global plugins_loaded
     if plugins_loaded:
         return