From e1d97c38f8689da0b11da0fac54cc277c237d5c4 Mon Sep 17 00:00:00 2001 From: Runkai Tao <129432511+RunkaiTao@users.noreply.github.com> Date: Wed, 11 Feb 2026 22:30:57 -0500 Subject: [PATCH] [Bug Fix] Fix `naive_block_assignment` always defaulting to False due to arg misalignment (#33848) Signed-off-by: Runkai Tao --- vllm/lora/layers/fused_moe.py | 2 +- vllm/lora/punica_wrapper/punica_base.py | 1 + 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/vllm/lora/layers/fused_moe.py b/vllm/lora/layers/fused_moe.py index 4d4e053cff..e3d9894de8 100644 --- a/vllm/lora/layers/fused_moe.py +++ b/vllm/lora/layers/fused_moe.py @@ -219,7 +219,7 @@ class FusedMoEWithLoRA(BaseLayerWithLoRA): self.max_loras, self.adapter_enabled, expert_map, - naive_block_assignment, + naive_block_assignment=naive_block_assignment, ) moe_state_dict["sorted_token_ids_lora"] = sorted_token_ids_lora diff --git a/vllm/lora/punica_wrapper/punica_base.py b/vllm/lora/punica_wrapper/punica_base.py index fdcf6c0cb1..facbd681a0 100644 --- a/vllm/lora/punica_wrapper/punica_base.py +++ b/vllm/lora/punica_wrapper/punica_base.py @@ -458,6 +458,7 @@ class PunicaWrapperBase(PunicaWrapperABC): adapter_enabled: torch.Tensor, expert_map: torch.Tensor | None = None, pad_sorted_ids: bool = False, + naive_block_assignment: bool = False, ) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]: """ Aligns tokens and experts into block-sized chunks for LoRA-based