3 ms·
LoRA training/merging basically is "crank up the batch size ridiculously high" in a nutshell, right? What actually breaks when you do that?
by FeepingCreature 3y ago
LoRA training/merging basically is "crank up the batch size ridiculously high" in a nutshell, right? What actually breaks when you do that?
- brrrrrm 3y agoCranking up the batch size kills convergence.
- FeepingCreature 3y agoWonder if that can be avoided by modifying the training approach. Ideas offhand: group by topic, train a subset of weights per node; figure out which layers have the most divergence and reduce lr on those only.
- brrrrrm 3y agoA provable way to recover convergence is to calculate the hessian. It’s computationally expensive but there are approximation methods.