Update app.py

app.py

@@ -5,6 +5,11 @@ import os
 from transformers import pipeline, WhisperProcessor, WhisperForConditionalGeneration, AutoModelForCausalLM, AutoProcessor
 from gtts import gTTS
 from langdetect import detect
+import subprocess
+from io import BytesIO
+
+# Install flash-attn
+subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
 
 # Disable CUDA initialization at import
 os.environ['CUDA_VISIBLE_DEVICES'] = ''