fix import error

setup.py

@@ -395,6 +395,7 @@ def get_version_add(sha: Optional[str] = None) -> str:
     version += ".torch" + torch.__version__[:5]
 
     with open(add_version_path, encoding="utf-8",mode="w") as file:
+        file.write("__commit__='705f6a35aa92800b9c69c8b346168c3055fdebbd'\n")
         file.write("__version__='0.5.2'\n")
         file.write("__dcu_version__='0.5.2+{}'\n".format(version))
     file.close()

vllm/__init__.py

@@ -11,9 +11,7 @@ from vllm.outputs import (CompletionOutput, EmbeddingOutput,
                           EmbeddingRequestOutput, RequestOutput)
 from vllm.pooling_params import PoolingParams
 from vllm.sampling_params import SamplingParams
-from vllm.version import __dcu_version__
-
-from .version import __commit__, __version__
+from vllm.version import __commit__, __version__, __dcu_version__
 
 __all__ = [
     "__commit__",

vllm/distributed/device_communicators/custom_all_reduce.py

@@ -12,10 +12,14 @@ from vllm.distributed.device_communicators.custom_all_reduce_utils import (
 from vllm.distributed.parallel_state import in_the_same_node_as
 from vllm.logger import init_logger
 from vllm.utils import cuda_device_count_stateless, is_full_nvlink
+from vllm.utils import is_hip
 
 try:
-    assert ops.is_custom_op_supported("_C_custom_ar::meta_size")
-    custom_ar = True
+    if (not is_hip()):
+        assert ops.is_custom_op_supported("_C_custom_ar::meta_size")
+        custom_ar = True
+    else:
+        custom_ar = False
 except Exception:
     # For AMD GPUs and CPUs
     custom_ar = False