ja1496

Results 4 issues of ja1496

[Feat]: Custom Max Token Length for SD/SDXL

3 comment

### Describe your use-case. Increase the max number of tokens supported by SD1.5 and SDXL from 75. ### What would you like to see as a solution? Increased token length...

enhancement

根据kohya_ss中关于oft脚本的介绍，我认为需要修改gui的js代码中关于network_alpha的最小值，来正确执行oft训练，具体如图

![Image](https://github.com/user-attachments/assets/58c59908-00c9-410e-9b1a-a2bc58f5e18d) 如图中所示，使用kohya-ss 的oft代码进行训练，network_alpha会被等价为constrain ，而constrain 参数推介的最大值为1e-5. 我通过修改js代码，使得network_alpha最小值为1e-5后，进行训练，看到了打印参数中的constrain确实为network_alpha值 ![Image](https://github.com/user-attachments/assets/c32413ce-f06e-46b8-a2e5-0fe381c3f57e)

I would like to know if the current flux fine-tuning script supports fine-tuning training for T5 and CLIPL? Request assistance

[Bug] TypeError in prepare_buckets_latents.py with --skip_existing due to missing alpha_mask argument

Hello! I've identified a bug in finetune/prepare_buckets_latents.py that causes a crash when using the --skip_existing flag. Bug Description When running prepare_buckets_latents.py, if the --skip_existing argument is used, the script fails...