Support (1 <= dp < tp) in the dp attention in DeepEP (#4770)

Co-authored-by: Cheng Wan <cwan39@gatech.edu>
2025-03-27 20:09:35 -04:00
parent 98a2cfa9b2
commit 7f19e083c1
10 changed files with 238 additions and 47 deletions
--- a/python/sglang/srt/model_executor/model_runner.py
+++ b/python/sglang/srt/model_executor/model_runner.py
@@ -281,9 +281,6 @@ class ModelRunner:

        if server_args.enable_deepep_moe:
            logger.info("DeepEP is turned on.")
-            assert (
-                server_args.enable_dp_attention == True
-            ), "Currently DeepEP is bind to Attention DP. Set '--enable-dp-attention --enable-deepep-moe'"

    def init_torch_distributed(self):
        logger.info("Init torch distributed begin.")