search

pratyushasharma / laser

The Truth Is In There: Improving Reasoning in Language Models with Layer-Selective Rank Reduction
https://pratyushasharma.github.io/laser/
MIT License
371 stars 28 forks source link

Does LASER have future work? #29

Open pangsg opened 2 months ago

pangsg commented 2 months ago

Hello, I have read the LASER paper carefully, i think it's an excellent paper. i notice that , your team have plan to do some experiment on combine these interventions，"There are many other ways in which one can combine these interventions, however, and we defer a this to future work.", which i am looking forward.

dkmisra commented 2 months ago

Hi @pangsg . There is a bunch of interesting directions that emerge from LASER and there is already a long line of work that has built up in this space since our paper came out. You can look at some of the citations for reference here.

I'd love to work again on extending LASER including the question you mentioned. However, last several months have had different priorities take over my time. Hopefully, I have more time in the next few months.

If you do publish in this space, I'd love to read it. I do think the idea of combining interventions is a great one and one where a lot can be done.

pangsg commented 2 months ago

thanks a lot