11750 steps.

9750 steps

6250 steps, improved aesthetics and coherence

4000 steps, same datasets, hparams

3500 steps, enabled mj-v6, bg20k-1024 combined datasets, better text generation.

2250 steps, probably last ckpt before it starts multi-aspect training

2000 steps

1000 steps

Upload pytorch_lora_weights.safetensors

Delete pytorch_lora_weights.safetensors

500 steps

250 steps of attempt 3 stage i lora, rank 32/alpha 16, same datasets. 64px, square aspect.

retraining on new settings. moving old model out of the way. was at 2500 steps, attempt 2.

2500 steps: re-trained with rank-32/alpha-16, typography dataset from CogVLM included

Add 2 files