fix import

Signed-off-by: reidliu41 <reid201711@gmail.com>

Author: reidliu41

vllm/__init__.py

@@ -8,21 +8,6 @@
 
 import torch
 
-from vllm.engine.arg_utils import AsyncEngineArgs, EngineArgs
-from vllm.engine.async_llm_engine import AsyncLLMEngine
-from vllm.engine.llm_engine import LLMEngine
-from vllm.entrypoints.llm import LLM
-from vllm.executor.ray_utils import initialize_ray_cluster
-from vllm.inputs import PromptType, TextPrompt, TokensPrompt
-from vllm.model_executor.models import ModelRegistry
-from vllm.outputs import (ClassificationOutput, ClassificationRequestOutput,
-                          CompletionOutput, EmbeddingOutput,
-                          EmbeddingRequestOutput, PoolingOutput,
-                          PoolingRequestOutput, RequestOutput, ScoringOutput,
-                          ScoringRequestOutput)
-from vllm.pooling_params import PoolingParams
-from vllm.sampling_params import SamplingParams
-
 # set some common config/environment variables that should be set
 # for all processes created by vllm and all processes
 # that interact with vllm workers.
@@ -36,6 +21,41 @@
 # see https://github.com/vllm-project/vllm/issues/10619
 torch._inductor.config.compile_threads = 1
 
+_lazy_imports_module_list = {
+    "LLM": "vllm.entrypoints.llm.LLM",
+    "ModelRegistry": "vllm.model_executor.models.ModelRegistry",
+    "PromptType": "vllm.inputs.PromptType",
+    "TextPrompt": "vllm.inputs.TextPrompt",
+    "TokensPrompt": "vllm.inputs.TokensPrompt",
+    "SamplingParams": "vllm.sampling_params.SamplingParams",
+    "RequestOutput": "vllm.outputs.RequestOutput",
+    "CompletionOutput": "vllm.outputs.CompletionOutput",
+    "PoolingOutput": "vllm.outputs.PoolingOutput",
+    "PoolingRequestOutput": "vllm.outputs.PoolingRequestOutput",
+    "EmbeddingOutput": "vllm.outputs.EmbeddingOutput",
+    "EmbeddingRequestOutput": "vllm.outputs.EmbeddingRequestOutput",
+    "ClassificationOutput": "vllm.outputs.ClassificationOutput",
+    "ClassificationRequestOutput": "vllm.outputs.ClassificationRequestOutput",
+    "ScoringOutput": "vllm.outputs.ScoringOutput",
+    "ScoringRequestOutput": "vllm.outputs.ScoringRequestOutput",
+    "LLMEngine": "vllm.engine.llm_engine.LLMEngine",
+    "EngineArgs": "vllm.engine.arg_utils.EngineArgs",
+    "AsyncLLMEngine": "vllm.engine.async_llm_engine.AsyncLLMEngine",
+    "AsyncEngineArgs": "vllm.engine.arg_utils.AsyncEngineArgs",
+    "initialize_ray_cluster": "vllm.executor.ray_utils.initialize_ray_cluster",
+    "PoolingParams": "vllm.pooling_params.PoolingParams",
+}
+
+
+def __getattr__(name: str):
+    if name in _lazy_imports_module_list:
+        import importlib
+        module_path, attr = _lazy_imports_module_list[name].rsplit(".", 1)
+        mod = importlib.import_module(module_path)
+        return getattr(mod, attr)
+    raise AttributeError(f"module {__name__!r} has no attribute {name!r}")
+
+
 __all__ = [
     "__version__",
     "__version_tuple__",

vllm/engine/arg_utils.py

@@ -1834,8 +1834,11 @@ def add_cli_args(parser: FlexibleArgumentParser,
         parser.add_argument('--disable-log-requests',
                             action='store_true',
                             help='Disable logging requests.')
-        from vllm.platforms import current_platform
-        current_platform.pre_register_and_update(parser)
+        # Skip to avoid triggering platform detection
+        import sys
+        if not any(arg in sys.argv for arg in ["-h", "--help"]):
+            from vllm.platforms import current_platform
+            current_platform.pre_register_and_update(parser)
         return parser
 
 

vllm/executor/ray_utils.py

@@ -11,7 +11,6 @@
 from vllm.config import ParallelConfig
 from vllm.executor.msgspec_utils import decode_hook, encode_hook
 from vllm.logger import init_logger
-from vllm.platforms import current_platform
 from vllm.sequence import ExecuteModelRequest, IntermediateTensors
 from vllm.utils import get_ip
 from vllm.worker.worker_base import WorkerWrapperBase
@@ -109,7 +108,7 @@ def setup_device_if_necessary(self):
             # We can remove this API after it is fixed in compiled graph.
             assert self.worker is not None, "Worker is not initialized"
             if not self.compiled_dag_cuda_device_set:
-                if current_platform.is_tpu():
+                if vllm.platforms.current_platform.is_tpu():
                     # Not needed
                     pass
                 else:

vllm/model_executor/layers/spec_decode_base_sampler.py

@@ -7,8 +7,6 @@
 import torch.jit
 import torch.nn as nn
 
-from vllm.platforms import current_platform
-
 
 class SpecDecodeBaseSampler(nn.Module):
     """Base class for samplers used for Speculative Decoding verification
@@ -37,6 +35,7 @@ def __init__(self, strict_mode: bool = False):
     def init_gpu_tensors(self, device: Union[int, str]) -> None:
         assert self.num_accepted_tokens is None
         if isinstance(device, int):
+            from vllm.platforms import current_platform
             device = f"{current_platform.device_type}:{device}"
         elif not isinstance(device, str):
             raise ValueError(f"Device must be int or str, get {type(device)}")

vllm/plugins/__init__.py

@@ -57,24 +57,28 @@ def load_general_plugins():
         return
     plugins_loaded = True
 
-    # some platform-specific configurations
-    from vllm.platforms import current_platform
+    # Skip to avoid triggering platform detection
+    import sys
+    if not any(arg in sys.argv for arg in ["-h", "--help"]):
+
+        # some platform-specific configurations
+        from vllm.platforms import current_platform
 
-    if current_platform.is_xpu():
-        # see https://github.com/pytorch/pytorch/blob/43c5f59/torch/_dynamo/config.py#L158
-        torch._dynamo.config.disable = True
-    elif current_platform.is_hpu():
-        # NOTE(kzawora): PT HPU lazy backend (PT_HPU_LAZY_MODE = 1)
-        # does not support torch.compile
-        # Eager backend (PT_HPU_LAZY_MODE = 0) must be selected for
-        # torch.compile support
-        is_lazy = os.environ.get('PT_HPU_LAZY_MODE', '1') == '1'
-        if is_lazy:
+        if current_platform.is_xpu():
+            # see https://github.com/pytorch/pytorch/blob/43c5f59/torch/_dynamo/config.py#L158
             torch._dynamo.config.disable = True
-            # NOTE(kzawora) multi-HPU inference with HPUGraphs (lazy-only)
-            # requires enabling lazy collectives
-            # see https://docs.habana.ai/en/latest/PyTorch/Inference_on_PyTorch/Inference_Using_HPU_Graphs.html # noqa: E501
-            os.environ['PT_HPU_ENABLE_LAZY_COLLECTIVES'] = 'true'
+        elif current_platform.is_hpu():
+            # NOTE(kzawora): PT HPU lazy backend (PT_HPU_LAZY_MODE = 1)
+            # does not support torch.compile
+            # Eager backend (PT_HPU_LAZY_MODE = 0) must be selected for
+            # torch.compile support
+            is_lazy = os.environ.get('PT_HPU_LAZY_MODE', '1') == '1'
+            if is_lazy:
+                torch._dynamo.config.disable = True
+                # NOTE(kzawora) multi-HPU inference with HPUGraphs (lazy-only)
+                # requires enabling lazy collectives
+                # see https://docs.habana.ai/en/latest/PyTorch/Inference_on_PyTorch/Inference_Using_HPU_Graphs.html # noqa: E501
+                os.environ['PT_HPU_ENABLE_LAZY_COLLECTIVES'] = 'true'
 
     plugins = load_plugins_by_group(group='vllm.general_plugins')
     # general plugins, we only need to execute the loaded functions