From 1e0806f30b99205bffe453b0caef77bc21e99baa Mon Sep 17 00:00:00 2001 From: fzyzcjy <5236035+fzyzcjy@users.noreply.github.com> Date: Sat, 19 Apr 2025 13:38:07 +0800 Subject: [PATCH] Fix DeepGEMM masked cannot be run on groups not being multiple or 4 (#5340) --- python/sglang/srt/layers/moe/ep_moe/layer.py | 3 --- 1 file changed, 3 deletions(-) diff --git a/python/sglang/srt/layers/moe/ep_moe/layer.py b/python/sglang/srt/layers/moe/ep_moe/layer.py index a35d0b8d0..a5a9eb738 100644 --- a/python/sglang/srt/layers/moe/ep_moe/layer.py +++ b/python/sglang/srt/layers/moe/ep_moe/layer.py @@ -972,9 +972,6 @@ class DeepEPMoE(EPMoE): ): assert self.quant_method is not None assert self.activation == "silu" - assert ( - hidden_states_fp8[0].size(0) % 4 == 0 - ), f"TMA alignment error: {hidden_states_fp8[0].size(0)}" # GroupGemm-0 num_groups, m, k = hidden_states_fp8[0].size()