使用groupgemm完成高吞吐模式适配

python/sglang/srt/layers/moe/token_dispatcher/deepep.py

@@ -541,7 +541,7 @@ class _DeepEPDispatcherImplLowLatency(_DeepEPDispatcherImplBase):
         num_max_dispatch_tokens_per_rank: the actual batch size in the decoding engine should be less than 256
         https://github.com/deepseek-ai/DeepEP?tab=readme-ov-file#example-use-in-inference-decoding
         """
-        self.return_recv_hook = return_recv_hook
+        self.return_recv_hook = False
         self.device_module = torch.get_device_module()
         self.quant_config = {}