Skip to content

a light but powerful trainer for both LoRA and Dreambooth

Notifications You must be signed in to change notification settings

alexzhengyc/magic-trainer-autodl

Repository files navigation

magic-trainer-autodl

Autodl version of the SD-trainer

  1. 如果是tmp里没有模型,cp autodl-fs/models -r autodl-tmp

  2. 如果仍要读取autodl-fs下的sd模型,修改 train_lora/train_dreambooth 的 {sd_path} (基本不需要)如果有需要修改blip模型和vae模型的位置,修改 Dreambooth/Lora 中的{self.vae_path/self.blip_path}

  3. 运行程序使用automatic1111的venv, python interpreter 输入 stable-diffusion-webui/venv/bin

  4. 单次训练文件夹在autodl-tmp/training-outputs 下的{dir_name}, 训练出的模型在 autodl-tmp/models/ 下的lora或者dreambooth 文件夹,可以直接被automatic1111读取

  5. 修改configs 中的模型参数,可以设置多个dict来实现多次训练

  6. 只运行model = Lora(**config) / model = Dreambooth(**config), 来生成文件夹

  7. 运行model.prepare() 来自动打标签(已有标签注释这行)

  8. 运行model.train() 开始训练

About

a light but powerful trainer for both LoRA and Dreambooth

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages