[cheald]

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.

[tangjurine]

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

[cheald]

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.