I had no trouble understanding this course except the chapter on Embedding Adaptors. I finished this course thinking it’s RLHF, because the course trained the adaptor using mock user feedback. I later learned that it’s more like a LoRA. I’m still very confused by this topic.
I have a couple questions listed below. If anyone can even answer 1-2, I’d really appreciate it:
- What exactly are embedding adaptors and what does it do?
- When should (or shouldn’t) you use embedding adaptors?
- Pros and cons of using embedding adaptors?
- Difference between embedding adaptor and fine-tuning an embedding model?