lllyasviel / lllyasviel/stable-diffusion-webui-forge
Training problems with flux1-dev-bnb-nf4-v2 model
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 13k
- Forks
- 1.7k
- PR merge metrics
- No merged PRs in 30d
Description
Hello lllyasviel, I tried to do some DreamBooth training on an RTX 4090 with your flux1-dev-bnb-nf4-v2 model in Kohya but couldnt get it to train.
It works fine with the regular flux1-dev model. Is this an expected consequence of the quantization and compression that was used or should it still train the same way the dev model does?
Ive been using the flux1-dev model this past week for training and it has been extremely slow and painful so I am eager to either try to get the flux1-dev-bnb-nf4-v2 working with Dream Booth OR keep using flux1-dev despite it being painfully slow and then do an nf4 conversion on the final model.
Thank you for any help or guidance on this, I am so grateful for everything you do!
Contributor guide
No contributing guide indexed for this repository
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start by reproducing DreamBooth training with the flux1-dev-bnb-nf4-v2 model in Kohya on an RTX 4090, then compare the result with the regular flux1-dev model. The issue provides no files, tests, logs, or entry point; done would require establishing whether NF4 quantization supports training or documenting the limitation and guidance for conversion.
Written by the indexing model from the issue text.
Assessment
- Domain
- machine-learning
- Issue type
- Bug
- Difficulty
- 5/5
- Estimated time
- Over a week
- Activity status
- Stale
- Clarity
- Needs clarification
- Newbie friendliness
- 15/100