Support torchrun and SPMD-style offline inference (#12071)

Signed-off-by: youkaichao <youkaichao@gmail.com>
2025-01-16 19:58:53 +08:00
parent dd7c9ad870
commit bf53e0c70b
14 changed files with 248 additions and 30 deletions
--- a/vllm/engine/arg_utils.py
+++ b/vllm/engine/arg_utils.py
@@ -388,7 +388,7 @@ class EngineArgs:
        # Parallel arguments
        parser.add_argument(
            '--distributed-executor-backend',
-            choices=['ray', 'mp'],
+            choices=['ray', 'mp', 'uni', 'external_launcher'],
            default=EngineArgs.distributed_executor_backend,
            help='Backend to use for distributed model '
            'workers, either "ray" or "mp" (multiprocessing). If the product '