kohya_ss · Issues· 533 open
Open on GitHubLocally synced open issues (discussions stay on GitHub)
- #3605
`torch` resolves to 2.14.0+xpu (download.pytorch.org/whl/xpu) or 2.9.1+cu128 (download.pytorch.org/whl/cu128) depending on install path (pip install .` with the index configuration committed in the repo)
Updated Sep 12, 2026 - #3602
No "Train U-Net Only" option for Anima training
Updated Sep 10, 2026 - #3603
HuggingFace token written in cleartext to setup.log on plain GUI startup
Updated Sep 7, 2026 - #3600
Consider using pre-built Flash Attention kernels via `kernels`
Updated Aug 21, 2026 - #3470
How to install for AMD ROCm on Linux?
Updated Aug 13, 2026 - #3275
Update via Stability Matrix fails and makes previous versions not work
Updated Aug 11, 2026 - #3590
ValueError: Unsupported attention mode: None
Updated Aug 3, 2026 - #3598
UnicodeEncodeError: 'charmap' codec can't encode characters in position 19-22: character maps
Updated Aug 1, 2026 - #3591
Bad keys in Anima finetune output model
Updated Jul 30, 2026 - #2771
Flux LoRA resumes wrong epochs
bugkohya_ss issuetrainingready-for-humanUpdated Jul 10, 2026 - #2520
49 sec/it on 4090
Troubleshootingneeds-triageUpdated Jul 10, 2026 - #3200
Severe Training Slowdown – xFormers Ignored, SDPA Used Instead
bugneeds-triageUpdated Jul 10, 2026 - #3467
How to speed up training
questionTroubleshootingneeds-infoUpdated Jul 10, 2026 - #3257
Large speed decrease with newest version
bugneeds-infoUpdated Jul 10, 2026 - #3378
The problem with training speed
bugneeds-infoUpdated Jul 10, 2026