From b0d1bfd924f5a9f968aca117e11a3bc5d69e7c6a Mon Sep 17 00:00:00 2001 From: ybyang Date: Mon, 28 Apr 2025 18:33:50 +0800 Subject: [PATCH 1/2] [PD] Fix https://github.com/deepseek-ai/DeepEP/issues/134 --- python/sglang/srt/layers/moe/ep_moe/token_dispatcher.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) 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 1f27b339337f..132bfa0b8ed3 100644 --- a/python/sglang/srt/layers/moe/ep_moe/token_dispatcher.py +++ b/python/sglang/srt/layers/moe/ep_moe/token_dispatcher.py @@ -85,7 +85,7 @@ def get_deepep_buffer( num_rdma_bytes, low_latency_mode=deepep_mode.enable_low_latency(), num_qps_per_rank=( - num_experts // group.size() if deepep_mode.enable_low_latency() else 1 + max(num_experts // group.size(), Buffer.num_sms // 2) ), ) return cls._buffer From 12825c7d9f5d6c18a447c0f2dc823e8af38f4846 Mon Sep 17 00:00:00 2001 From: ybyang Date: Wed, 30 Apr 2025 09:47:10 +0800 Subject: [PATCH 2/2] fix lint --- python/sglang/srt/layers/moe/ep_moe/token_dispatcher.py | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) 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 132bfa0b8ed3..f273c55cb3d1 100644 --- a/python/sglang/srt/layers/moe/ep_moe/token_dispatcher.py +++ b/python/sglang/srt/layers/moe/ep_moe/token_dispatcher.py @@ -84,9 +84,7 @@ def get_deepep_buffer( num_nvl_bytes, num_rdma_bytes, low_latency_mode=deepep_mode.enable_low_latency(), - num_qps_per_rank=( - max(num_experts // group.size(), Buffer.num_sms // 2) - ), + num_qps_per_rank=(max(num_experts // group.size(), Buffer.num_sms // 2)), ) return cls._buffer