diff --git a/python/sglang/srt/layers/quantization/utils.py b/python/sglang/srt/layers/quantization/utils.py
index d2c7975ba..a7be39141 100644
--- a/python/sglang/srt/layers/quantization/utils.py
+++ b/python/sglang/srt/layers/quantization/utils.py
@@ -146,6 +146,10 @@ def requantize_with_max_scale(
     return max_w_scale, weight
 
 
+def update_tensor_inplace(old: torch.Tensor, new: torch.Tensor) -> None:
+    old.copy_(new)
+
+
 # Adapted from https://github.com/vllm-project/vllm/blob/main/vllm/model_executor/layers/quantization/utils/layer_utils.py
 # Newly generated tensors need to replace existing tensors that are
 # already registered as parameters by vLLM (and won't be freed)