[PP] Add pipeline parallelism (#5724)

2025-04-30 18:18:07 -07:00
parent e97e57e699
commit 11383cec3c
25 changed files with 1150 additions and 308 deletions
--- a/python/sglang/srt/speculative/eagle_worker.py
+++ b/python/sglang/srt/speculative/eagle_worker.py
@@ -106,11 +106,12 @@ class EAGLEWorker(TpModelWorker):
        # Init draft worker
        with empty_context():
            super().__init__(
+                server_args=server_args,
                gpu_id=gpu_id,
                tp_rank=tp_rank,
-                server_args=server_args,
-                nccl_port=nccl_port,
+                pp_rank=0,  # FIXME
                dp_rank=dp_rank,
+                nccl_port=nccl_port,
                is_draft_worker=True,
                req_to_token_pool=self.req_to_token_pool,
                token_to_kv_pool_allocator=self.token_to_kv_pool_allocator,