Ok, try this

app.py

@@ -42,14 +42,19 @@ def predict(code):
 
     print(f"Prompt: {str(prompt)}")
     print(f"Tokenized: {tokenizer.tokenize(prompt)}")
-    pipe_out =  pipe(prompt)
+    pipe_out = pipe(
+        prompt,
+        max_length=10**6,
+        top_k=100,
+        top_p=0.09,
+        temperature=1.2,
+        repetition_penalty=1.1,
+    )
     print(f"Pipe out: {str(pipe_out)}")
 
-    return model.generate(
-        tokenizer.tokenize(prompt),
-        return_text=True,
-        max_length=10**6
-    )
+    return tokenizer.decode(model.generate(
+        tokenizer.encode(prompt), max_length=10**6
+    ))
 
 
 demo = gr.Interface(