Description of the feature request:

There is a lot of talk that fine-tuning can only teach the model to adapt to the task, and it is really difficult to inject domain knowledge when I actually try it (let alone lora), and the report only relies on command fine-tuning to achieve such excellent domain knowledge injection, it is amazing! Can you share some more details? For example, how to effectively inject domain knowledge (in the absence of much data)

What problem are you trying to solve with this feature?

No response

Any other information you'd like to share?

No response