From 74898a70159f4217f1abf2a738c5cc6c4695cc44 Mon Sep 17 00:00:00 2001 From: Danielle Robinson Date: Fri, 30 Jan 2026 07:27:42 -0800 Subject: [PATCH] [BugFix][LoRA] TritonExperts is ModularMoEPath for FP8 models (#33393) Signed-off-by: Danielle Robinson Co-authored-by: Danielle Robinson --- vllm/lora/layers/fused_moe.py | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-) diff --git a/vllm/lora/layers/fused_moe.py b/vllm/lora/layers/fused_moe.py index 8f5cb14c3..be1fd7cdb 100644 --- a/vllm/lora/layers/fused_moe.py +++ b/vllm/lora/layers/fused_moe.py @@ -143,9 +143,7 @@ class FusedMoEWithLoRA(BaseLayerWithLoRA): m_fused_moe_fn.fused_experts, (MarlinExperts, UnfusedOAITritonExperts) ) else: - assert isinstance( - m_fused_moe_fn.fused_experts, (MarlinExperts, TritonExperts) - ) + assert isinstance(m_fused_moe_fn.fused_experts, TritonExperts) def fwd_decorator(layer, func): def wrapper(*args, **kwargs):