VictorMylle/Kohya-ss-sd-scripts
1
0
Fork 1
mirror of https://github.com/kohya-ss/sd-scripts.git synced 2026-04-09 06:45:09 +00:00
Code Issues Packages Projects Releases Wiki Activity

add t5xxl max token length, support schnell

Browse Source
This commit is contained in:
Kohya S
2024-08-16 17:06:05 +09:00
parent 739a8969bc
commit 3921a4efda
3 changed files with 44 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
README.md
View File

@@ -9,6 +9,14 @@ __Please update PyTorch to 2.4.0. We have tested with `torch==2.4.0` and `torchv
The command to install PyTorch is as follows:
`pip3 install torch==2.4.0 torchvision==0.19.0 --index-url https://download.pytorch.org/whl/cu124`
Aug 16, 2024:
FLUX.1 schnell model based training is now supported (but not tested). If the name of the model file contains `schnell`, the model is treated as a schnell model.
Added `--t5xxl_max_token_length` option to specify the maximum token length of T5XXL. The default is 512 in dev and 256 in schnell.
Previously, when `--max_token_length` was specified, that value was used, and 512 was used when omitted (default). Therefore, there is no impact if `--max_token_length` was not specified. If `--max_token_length` was specified, please specify `--t5xxl_max_token_length` instead. `--max_token_length` is ignored during FLUX.1 training.
Aug 14, 2024: Sample image generation during training is now supported. Specify options such as `--sample_prompts` and `--sample_every_n_epochs`. It will be very slow when `--split_mode` is specified.
Aug 13, 2024: