Update app.py

app.py

@@ -10,7 +10,14 @@ 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")
-model= quantize_blockwise(model)
+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
+            module = quantize_blockwise(module)
+
+# Quantize the model (modified)
+quantize_model(model)
 # alpaca_prompt = You MUST copy from above!
 @spaces.GPU(duration=300)
 def chunk_it(input_command, item_list):