Update app.py

app.py

@@ -11,9 +11,8 @@ from bitsandbytes.functional import quantize_blockwise
 tokenizer = AutoTokenizer.from_pretrained("FlawedLLM/Bhashini_gemma_merged4bit_clean_final")
 model = AutoModelForCausalLM.from_pretrained("FlawedLLM/Bhashini_gemma_merged4bit_clean_final")
 def quantize_model(model):
-    # Iterate over model parameters, not the entire model object
     for name, module in model.named_modules():
-        if isinstance(module, torch.nn.Linear):  # Quantize only Linear layers
+        if isinstance(module, torch.nn.Linear) and not isinstance(module, torch.nn.Linear4bit):
             module = quantize_blockwise(module)
 
 # Quantize the model (modified)