Live data from Hacker News

LoRA Without Regret

thinkingmachines.ai

61–62 of 62 posts

Re: LoRA Without Regret

#61
post #53

Earlier quoted context omitted.

Standard LoRA (W_delta = B@A with standard inits) generally underperforms FT, primarily because of "intruder dimensions" (new high-ranking singular vectors which misalign with the singular vectors of the underlying weights) as outlined in the paper. There are techniques like PiCa and SVFT which can mitigate much of the loss, though.

pica came out two days ago, how did you find out about it?

The one I was referring to was from this paper, first published in May: https://arxiv.org/abs/2505.20211v1

I don't recall how I found out about it, but it was either paperswithcode or an LLM research session working through the intruder dimensions problem.

In my Stable Diffusion tests, it substantially improves LoRA training speed and fidelity, though I've got some experiments that seem to even further substantially improve on it by adding learnable rotations of the singular vectors.

Re: LoRA Without Regret

#62
post #28

Earlier quoted context omitted.

To say that the 'literature is clear on that' while citing a single paper, which has been rejected from ICLR, is a bit of an overstatement.

> which has been rejected from ICLR Oh, you mean rejected just like these papers? Efficient Estimation of Word Representations in Vector Space[1], one of the most influential papers in the space with tens of thousands of citations[2]? Or the RoBERTa[3] paper (dramatically improved upon BERT; RoBERTa and derived models currently have tens of millions of downloads on HF and still serve as a reliable industry workhorse)…

Based.

This guys knows his stuff.

Post reply on HN