Update app.py

app.py

@@ -19,9 +19,9 @@ tokenizer = AutoTokenizer.from_pretrained(base_model)
 
 base = AutoModelForCausalLM.from_pretrained(
     base_model,
-    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
+    load_in_8bit=True,
     device_map="auto",
-    use_auth_token="hf_" +hf_token  # ensure it uses your auth
+    use_auth_token="hf_" +hf_token 
 )
 
 model = PeftModel.from_pretrained(base, adapter_model)