wip

Signed-off-by: wenxindongwork <[email protected]>

Author: wenxindongwork

vllm/entrypoints/llm.py

@@ -66,6 +66,7 @@
     RequestOutput,
     ScoringRequestOutput,
 )
+from vllm.platforms import current_platform
 from vllm.pooling_params import PoolingParams
 from vllm.sampling_params import BeamSearchParams, RequestOutputKind, SamplingParams
 from vllm.tasks import PoolingTask
@@ -80,7 +81,6 @@
 from vllm.v1.engine import EngineCoreRequest
 from vllm.v1.engine.llm_engine import LLMEngine
 from vllm.v1.sample.logits_processor import LogitsProcessor
-from vllm.platforms import current_platform
 
 if TYPE_CHECKING:
     from vllm.v1.metrics.reader import Metric
@@ -289,7 +289,11 @@ def __init__(
         # warn about single-process data parallel usage.
         _dp_size = int(kwargs.get("data_parallel_size", 1))
         _distributed_executor_backend = kwargs.get("distributed_executor_backend")
-        if _dp_size > 1 and not _distributed_executor_backend == "external_launcher" and not current_platform.is_tpu():
+        if (
+            _dp_size > 1
+            and not _distributed_executor_backend == "external_launcher"
+            and not current_platform.is_tpu()
+        ):
             raise ValueError(
                 f"LLM(data_parallel_size={_dp_size}) is not supported for single-"
                 "process usage and may hang. Please use "