huggingface / huggingface/diffusers

SD3 dreambooth training - caching of embeddings and vae representations

Open
#8,540 2 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

stale
Dominant language
Python
Stars
34.5k
Forks
7.3k
Avg merge
3d 3h
Merged PRs (30d)
91

Description

Is your feature request related to a problem? Please describe.
The current implementation of the dreambooth trainings for both loras and finetuning is very memory intensive.

Describe the solution you'd like.
I would like the option to pre-cache the VAE representations of the images and the text encoder representations so that training could be done without the text encoders being in VRAM.

Describe alternatives you've considered.
I don't believe there are alternatives to reducing the VRAM requirement for training.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by locating the SD3 DreamBooth training entry points and how they currently encode images and text during training. The requested result is optional pre-caching of VAE and text-encoder representations so training can run without the text encoders in VRAM; the issue names no specific files or tests.

Written by the indexing model from the issue text.

Assessment

Tech stack
python, pytorch
Domain
machine-learning
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.