diff --git a/vllm-v0.6.2/vllm_mlu/vllm_mlu/model_executor/models/llama4.py b/vllm-v0.6.2/vllm_mlu/vllm_mlu/model_executor/models/llama4.py
index 604086f..495b2a4 100644
--- a/vllm-v0.6.2/vllm_mlu/vllm_mlu/model_executor/models/llama4.py
+++ b/vllm-v0.6.2/vllm_mlu/vllm_mlu/model_executor/models/llama4.py
@@ -9,7 +9,8 @@ from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm_mlu.model_executor.layers.feed_forward import FeedForward
 from vllm_mlu.mlu_hijack_utils import MluHijackObject
 from vllm.distributed import tensor_model_parallel_all_reduce
-from vllm.model_executor.model_loader.weight_utils import default_weight_loader
+from vllm.model_executor.model_loader.weight_utils import (
+    default_weight_loader, maybe_remap_kv_scale_name)
 from vllm.model_executor.models.llama4 import (
     Llama4Attention, Llama4DecoderLayer, Llama4ForCausalLM,
     Llama4Model, Llama4MoE)