diff --git a/src/transformers/integrations/flex_attention.py b/src/transformers/integrations/flex_attention.py
index 5181b2c1a0ad..1d5f35d9fe55 100644
--- a/src/transformers/integrations/flex_attention.py
+++ b/src/transformers/integrations/flex_attention.py
@@ -7,6 +7,8 @@
 
 if is_torch_flex_attn_available():
     from torch.nn.attention.flex_attention import flex_attention
+    flex_attention = torch.compile(flex_attention, dynamic=False)
+
 
 
 def flex_attention_forward(