[SDXL LoRA DreamBooth training] follow-up from #5508

[sayakpaul]

Some thoughts:

• Reflect the updates introduced in Add features to the Dreambooth LoRA SDXL training script #5508 in README_sdxl.md. In particular, focus on:
  • The simple improvements. If a blog post is being planned, just briefly mentioning the supported stuff and referencing the blog post work for me.
  • Interplay between repeats, max_train_epochs, and num_train_epochs. repeats doesn't likely affect the other two. But for a regular practitioner, it can be quite confusing when they see these three popping up in the same place.
• Rename to text_encoder_learning_rate for consistency with the other argument: learning_rate.
• When can't we cache the custom prompts as well? Saves memory, no? Or was it not done to keep the script simple? In that case, I understand no further explanation is needed.

Cc: @linoytsaban

[sayakpaul]

Gentle ping @linoytsaban.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.