diff --git a/python/sglang/srt/layers/moe/ep_moe/token_dispatcher.py b/python/sglang/srt/layers/moe/ep_moe/token_dispatcher.py index c8dac6810..8089168be 100644 --- a/python/sglang/srt/layers/moe/ep_moe/token_dispatcher.py +++ b/python/sglang/srt/layers/moe/ep_moe/token_dispatcher.py @@ -107,6 +107,8 @@ class DeepEPBuffer: num_rdma_bytes, low_latency_mode=deepep_mode.enable_low_latency(), num_qps_per_rank=num_qps_per_rank, + # TODO can be false when unneeded + allow_mnnvl=True, ) return cls._buffer