diff --git a/vllm/patches/deepseek_v4_attention.py b/vllm/patches/deepseek_v4_attention.py
index 4bb2a64c..acb77d72 100644
--- a/vllm/patches/deepseek_v4_attention.py
+++ b/vllm/patches/deepseek_v4_attention.py
@@ -14,7 +14,6 @@ import torch.nn.functional as F
 from transformers import DeepseekV2Config, DeepseekV3Config
 
 import vllm.envs as envs
-from vllm.compilation.breakable_cudagraph import eager_break_during_capture
 from vllm.model_executor.layers.linear import (
     ReplicatedLinear,
 )
@@ -554,7 +553,6 @@ class DeepseekV4MultiHeadLatentAttentionWrapper(PluggableLayer):
         )
 
 
-@eager_break_during_capture
 def deepseek_v4_attention(
     hidden_states: torch.Tensor,
     positions: torch.Tensor,