Question about LoRA Training Sources
#33
by
ZAHNGYUXUAN
- opened
Hi, thank you for sharing these LoRA weights and your work!
I’m curious about how these LoRA checkpoints were trained. Did you train them from scratch by re-implementing the methods from the original papers, or did you start with existing open-source weights released by those papers and then further finetune them?
If it’s the latter, I’d love to know a bit about your process—especially how you handle loading the original checkpoints and adapting them for LoRA training.
Thanks a lot for any info you can share!
Hey,
The LoRAs here are simply extracted from the original checkpoints, only modifications I've done in some cases were to prune some of the layers like with the case of CausVid.