Update app.py

app.py

@@ -26,7 +26,7 @@ try:
     # Load base model and tokenizer with HF auth token
     base_model = AutoModelForCausalLM.from_pretrained(
         peft_config.base_model_name_or_path,
-        torch_dtype=torch.float32,
+        torch_dtype=torch.bfloat16,
         device_map={"": "cpu"},
         use_auth_token=True
     )