Try different flash attention version

app.py

@@ -10,7 +10,7 @@ from qwen_vl_utils import process_vision_info
 from transformers.utils import is_flash_attn_2_available
 
 subprocess.run(
-    "pip install 'flash-attn>=2.2.0' --no-build-isolation",
+    "pip install 'flash-attn==2.2.0' --no-build-isolation",
     env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
     shell=True,
 )