diff --git a/vllm/attention/ops/triton_flash_attention.py b/vllm/attention/ops/triton_flash_attention.py
index 3671c2f91e3b7..2e647a8e52278 100644
--- a/vllm/attention/ops/triton_flash_attention.py
+++ b/vllm/attention/ops/triton_flash_attention.py
@@ -394,7 +394,6 @@ def get_autotune_configs():
 @triton.autotune(
     configs=autotune_configs,
     key=autotune_keys,
-    use_cuda_graph=True,
 )
 @triton.jit
 def attn_fwd(