Try something a little different

app.py

@@ -42,7 +42,14 @@ def predict(code):
 
     print(f"Prompt: {str(prompt)}")
     print(f"Tokenized: {tokenizer.tokenize(prompt)}")
-    return pipe(prompt)
+    pipe_out =  pipe(prompt)
+    print(f"Pipe out: {str(pipe_out)}")
+
+    return model.generate(
+        tokenizer.tokenize(prompt),
+        return_text=True,
+        max_length=10**6
+    )
 
 
 demo = gr.Interface(