Difference between training / generation input format.

[hbin0701]

Hi guys! :)
Thanks for making your code open-source.
One quick question. I've realized that in run_spin.py, you use apply_chat_template which differs from the format
of generate.py or generate_vllm.py which just tokenizes the following formatted string

"### Instruction: " + data[idx][0]['content'] + "\n\n### Response: " for idx in range(len(data))]

Please correct me if I am wrong😅
Thanks in advance!