GraphPKU / PiSSA

PiSSA: Principal Singular Values and Singular Vectors Adaptation of Large Language Models(NeurIPS 2024 Spotlight)
https://arxiv.org/abs/2404.02948
261 stars 9 forks source link

Output singular value #13

Open GYDTTDYX opened 4 months ago

GYDTTDYX commented 4 months ago

Hi, thank you for the great tool. I read your paper and notice that you recommend finding the singular vectors that correspond to top-k singular values to initialize the W pri matrix. I'm wondering are there any method I could use to extract singular value after I use init_lora_weights="pissa" or I have to do the SVD using other package first and plot the distribution of singular value? Thanks