[Feature] Support Pipeline Parallism in torchrun SPMD offline inference for V1 (#17827)

Signed-off-by: Lucia Fang <fanglu@fb.com>
2025-05-15 22:28:27 -07:00
parent 6b31c84aff
commit 3d2779c29a
9 changed files with 55 additions and 27 deletions
--- a/vllm/executor/uniproc_executor.py
+++ b/vllm/executor/uniproc_executor.py
@@ -86,9 +86,6 @@ class ExecutorWithExternalLauncher(UniProcExecutor):
    def _init_executor(self) -> None:
        """Initialize the worker and load the model.
        """
-        assert self.vllm_config.parallel_config.pipeline_parallel_size == 1, \
-            ("ExecutorWithExternalLauncher does not "
-            "support pipeline parallelism.")
        assert self.vllm_config.scheduler_config.delay_factor == 0.0, \
            ("ExecutorWithExternalLauncher needs deterministic "
            "execution, so it"