So I discovered I can still bake stuff and use stuff like gradient checkpointing and higher batch size, which is pretty cool, can pump these out a bit faster. Though I guess now I have no clue how many steps I should put for the model, now I know that my previous step counts were all wrong because I used the calculated steps, but that's not the 'true' step count since it's changed by the batch size. This model was probably around 4k-6k steps at the rate I previously did it at a batch size of 2, but with my current thing it took only 410 steps, so I'm not even sure if I should bother with that information section of the LoRA's because I don't think it really means anything to anyone. You can't tell how good a model is based off step counts and epochs alone.
