auto-vicuna 0.0.3

An experiment with Vicuna.

Auto-Vicuna

Fast Chat and transformers must be installed via requirements file for now.

Installing PyTorch and NVIDIA dependencies

First, upgrade wheel and friends.

pip install --upgrade setuptools pip wheel

For CUDA 11.X:

pip install nvidia-cuda-runtime-cu11 --index-url https://pypi.ngc.nvidia.com

For CUDA 12.x

pip install nvidia-cuda-runtime-cu12 --index-url https://pypi.ngc.nvidia.com

For PyTorch cu117

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu117

For PyTorch cu118

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

You might need to add:

--upgrade --force-reinstall

To the PyTorch installation commands if you have a previous version of PyTorch installed.

LLaMA Weights

If your weights used LLaMATokenizer you may need a different version of transformers.

I have fork that should work located here: https://github.com/BillSchumacher/transformers

You should clone this repo and:

pip uninstall transformers
cd path/to/my/transformers/cloned/repo
python setup.py install

For some reason pip install -e . causes issues.

If you do that, delete pycache, dist and build folders and try the setup.py method instead.