v0.9.8
版本发布时间: 2024-08-05 08:57:20
bghira/SimpleTuner最新发布版本:v0.9.8(2024-08-05 08:57:20)
Flux
It's here! Runs on 24G cards using Quanto's 8bit quantisation, or 25.7G on a Macbook system (slowly)!
If you're after accuracy, a 40G card will do Just Fine, with 80G cards being somewhat of a sweet spot for larger training efforts.
What you get:
- LoRA, full tuning (but probably just don't do that)
- Documentation to get you started fast
- Probably better for just square crop training for now - might artifact for weird resolutions
- Quantised base model unlocking the ability to safely use Adafactor, Prodigy, and other neat optimisers as a consolation prize for losing access to full bf16 training (AdamWBF16 just won't work with Quanto)
What's Changed
- trainer: simplify check by @bghira in https://github.com/bghira/SimpleTuner/pull/592
- documentation updates, apple pytorch 2.4 by @bghira in https://github.com/bghira/SimpleTuner/pull/595
- staged storage for image embed support by @bghira in https://github.com/bghira/SimpleTuner/pull/596
- fix: loading default image embed backend by @bghira in https://github.com/bghira/SimpleTuner/pull/597
- fix: loading default image embed backend by @bghira in https://github.com/bghira/SimpleTuner/pull/598
- multi-gpu console output improvements by @bghira in https://github.com/bghira/SimpleTuner/pull/599
- vae cache: hash_filenames option for image sets by @bghira in https://github.com/bghira/SimpleTuner/pull/601
- multi-gpu console output reduction by @bghira in https://github.com/bghira/SimpleTuner/pull/602
- fix for relative cache directories with NoneType being unsubscriptable by @bghira in https://github.com/bghira/SimpleTuner/pull/603
- multigpu / relative path fixes for caching by @bghira in https://github.com/bghira/SimpleTuner/pull/604
- backend for csv based datasets by @bghira in https://github.com/bghira/SimpleTuner/pull/600
- CSV data backend by @bghira in https://github.com/bghira/SimpleTuner/pull/605
- config file versioning to allow updating defaults without breaking backwards compat by @bghira in https://github.com/bghira/SimpleTuner/pull/607
- config file versioning for backwards compat by @bghira in https://github.com/bghira/SimpleTuner/pull/608
- experiment: small DiT model by @bghira in https://github.com/bghira/SimpleTuner/pull/609
- merge by @bghira in https://github.com/bghira/SimpleTuner/pull/610
- Fix crash when using jsonl files by @swkang-rp in https://github.com/bghira/SimpleTuner/pull/611
- merge by @bghira in https://github.com/bghira/SimpleTuner/pull/612
- flux training by @bghira in https://github.com/bghira/SimpleTuner/pull/614
- update base_dir to output_dir by @bghira in https://github.com/bghira/SimpleTuner/pull/615
- merge by @bghira in https://github.com/bghira/SimpleTuner/pull/616
- flux: validations should ignore any custom schedulers by @bghira in https://github.com/bghira/SimpleTuner/pull/618
- release: flux by @bghira in https://github.com/bghira/SimpleTuner/pull/617
- bugfix: correctly set hash_filenames to true or false for an initial dataset creation by @bghira in https://github.com/bghira/SimpleTuner/pull/620
- release: minor follow-up fixes by @bghira in https://github.com/bghira/SimpleTuner/pull/628
- Flux: Fix random validation errors due to some tensors being on the cpu by @mhirki in https://github.com/bghira/SimpleTuner/pull/629
- Improve config support for transformers with accelerate by @touchwolf in https://github.com/bghira/SimpleTuner/pull/630
- quanto: exploring low-precision training. by @bghira in https://github.com/bghira/SimpleTuner/pull/622
- remove all text encoders from memory correctly by @bghira in https://github.com/bghira/SimpleTuner/pull/637
New Contributors
- @swkang-rp made their first contribution in https://github.com/bghira/SimpleTuner/pull/611
- @touchwolf made their first contribution in https://github.com/bghira/SimpleTuner/pull/630
Full Changelog: https://github.com/bghira/SimpleTuner/compare/v0.9.7.8...v0.9.8