Update app.py

app.py

@@ -13,7 +13,8 @@ tokenizer = AutoTokenizer.from_pretrained("FlawedLLM/Bhashini_gemma_merged4bit_c
 model = AutoModelForCausalLM.from_pretrained("FlawedLLM/Bhashini_gemma_merged4bit_clean_final")
 def quantize_model(model):
     for name, module in model.named_modules():
-        if isinstance(module, torch.nn.Linear) and not isinstance(module, torch.nn.Linear4bit):
+        # Quantize only Linear layers that haven't already been quantized
+        if isinstance(module, torch.nn.Linear) and not isinstance(module, Linear4bit):
             module = quantize_blockwise(module)
 
 # Quantize the model (modified)