Update app.py

app.py

@@ -5,8 +5,10 @@ pipe1 = pipeline(model = "meta-llama/Llama-3.2-1B-Instruct")
 #pipe2 = pipeline(model = "meta-llama/Llama-3.2-3B-Instruct")
 
 def inference(input):
-    output = pipe1(input)
-    return output[0]['generated_text']
+    output = pipe1(
+        text_inputs = input
+    )
+    return output[0]['generated_text'][len(input):]
 
 app = gr.Interface(
     fn = inference,
@@ -16,6 +18,4 @@ app = gr.Interface(
     examples = [
         ["Hello, Llama."]
     ]
-)
-
-app.launch()
+).launch()