Update app.py

app.py

@@ -64,6 +64,7 @@ bnb_config = BitsAndBytesConfig(
     bnb_4bit_quant_type=bnb_4bit_quant_type,
     bnb_4bit_compute_dtype=compute_dtype,
     bnb_4bit_use_double_quant=use_nested_quant,
+    bnb_4bit_disable_gpu=True,  # Add this line to disable GPU quantization
 )
 
 # Remove GPU-specific check for bfloat16