jordach / cascadetuner

Implements a minimalistic version of Stable Cascade training

CascadeTuner

Implements a minimalistic version of Stable Cascade training

Installation Notes:

Install Anaconda or Miniconda.

At the terminal: conda env create -f environment.yaml

Wait a bit, as it takes time to install.

conda activate CTuner This will activate the virtual environment for CascadeTuner.

If bitsandbytes doesn't start under Windows 10/11: pip install https://github.com/jllllll/bitsandbytes-windows-webui/releases/download/wheels/bitsandbytes-0.41.1-py3-none-win_amd64.whl

Installing xformers: pip install --no-deps xformers --index-url https://download.pytorch.org/whl/cu121

Launching your script:

accelerate launch --mixed_precision="bf16" train_stage_c.py --yaml "configs/your_yaml_here.yaml"

If it doesn't launch: conda install -c defaults intel-openmp -f usually fixes PIL issues under Windows 10/11.

cascadetuner's Issues

Stage b training

Hello!
Have you finished the stage b and stage a training code?
The coded released are often include stage c training only. However, I find that only training stage c is not enough.
Fine-tuning the whole model may make sense.