Update app.py

app.py

@@ -31,6 +31,7 @@ model = AutoModelForCausalLM.from_pretrained(
 model = PeftModel.from_pretrained(model, lora_path)
 model = model.merge_and_unload()
 model.eval()
+model = torch.compile(model)
 
 model.config.use_cache = True
 
@@ -56,6 +57,8 @@ def call(related_materials, materials, question):
     start_time = time.time()
     generated_ids = model.generate(
         model_inputs.input_ids,
+        attention_mask=model_inputs.attention_mask,
+        do_sample=False,
         max_length=8096
     )