From d90bc65e3075fad138f426ac08af29fadf4e4645 Mon Sep 17 00:00:00 2001 From: Kurkur <102506892+litmei@users.noreply.github.com> Date: Tue, 19 May 2026 09:09:11 +0800 Subject: [PATCH] [NPU] Fix TypeError in get_state_buf_infos when index_head_dim is None on MLA (#25383) --- python/sglang/srt/hardware_backend/npu/memory_pool_npu.py | 2 ++ 1 file changed, 2 insertions(+) diff --git a/python/sglang/srt/hardware_backend/npu/memory_pool_npu.py b/python/sglang/srt/hardware_backend/npu/memory_pool_npu.py index 9d116c555..145993f7e 100644 --- a/python/sglang/srt/hardware_backend/npu/memory_pool_npu.py +++ b/python/sglang/srt/hardware_backend/npu/memory_pool_npu.py @@ -354,6 +354,8 @@ class NPUMLATokenToKVPool(MLATokenToKVPool): ) def get_state_buf_infos(self): + if self.index_head_dim is None: + return [], [], [] data_ptrs = [self.index_k_buffer[i].data_ptr() for i in range(self.layer_num)] data_lens = [self.index_k_buffer[i].nbytes for i in range(self.layer_num)] item_lens = [self.index_k_buffer[i][0].nbytes for i in range(self.layer_num)]