Update app.py

app.py

@@ -55,11 +55,11 @@ latex_delimiters_set = [{
 def predict(message, history, system_prompt, temperature, max_new_tokens, top_k, repetition_penalty, top_p):
     # Format history with a given chat template
     
-    stop_tokens = [tokenizer.eos_token_id]
-    instruction = system_prompt + "\n\n"
+    stop_tokens = ["<|endoftext|>", "<|im_end|>"]
+    instruction = '<|im_start|>system\n' + system_prompt + '\n<|im_end|>\n'
     for user, assistant in history:
-        instruction += f"role:user, content: {user}\nrole:assistant, content: {assistant}\n"
-    instruction += f"role:user, content: {message}\nassistant:"
+        instruction += f'<|im_start|>user\n{user}\n<|im_end|>\n<|im_start|>assistant\n{assistant}\n<|im_end|>\n'
+    instruction += f'<|im_start|>user\n{message}\n<|im_end|>\n<|im_start|>assistant\n'
     
     print(instruction)