Minor fix

app.py

@@ -7,10 +7,9 @@ import gradio as gr
 import spaces
 from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
 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 --no-build-isolation",
     env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
     shell=True,
 )
@@ -49,7 +48,6 @@ def run_inference(model_key, input_type, text, image, video, fps, system_prompt,
         model_id,
         torch_dtype="auto",
         device_map="auto",
-        attn_implementation="flash_attention_2" if is_flash_attn_2_available() else None,
     )
     processor = AutoProcessor.from_pretrained(model_id)