Any results about direct fine-tuning on the journey learning data?

[tongyx361]

It's possible that, like LIMA, direct fine-tuning on the high-quality journey learning data might produce competitive performance.

[john1226966735]

It's possible that, like LIMA, direct fine-tuning on the high-quality journey learning data might produce competitive performance.

It seems that in "6.8 Q8: How to Train our Models?" of the paper, the authors just apply SFT to finetune the model with journey learning data? Is that the same with your "direct fine-tuning on the journey learning data"?