A python package to register on the bittensor network using a CUDA device.
- Ubuntu 20.04 or higher
- bittensor>=3.0.0
- sm_61, sm_70, sm_75, sm_80, or sm_86 enabled CUDA GPU (See here)
- Change the base image to the target bittensor version
- Build the image:
sudo docker build -t opentensorfdn/bittensor:VERSION_BT-cubitVERSION_CUBIT -f ./docker/Dockerfile .
- Push the image
sudo docker push opentensorfdn/bittensor:VERSION_BT-cubitVERSION_CUBIT
Using the wheel for your version of python (3.8, etc.) from releases
For Python 3.8
pip install https://github.com/opentensor/cubit/releases/download/v1.1.2/cubit-1.1.2-cp38-cp38-linux_x86_64.whl
For Python 3.9
pip install https://github.com/opentensor/cubit/releases/download/v1.1.2/cubit-1.1.2-cp39-cp39-linux_x86_64.whl
For Python 3.10
pip install https://github.com/opentensor/cubit/releases/download/v1.1.2/cubit-1.1.2-cp310-cp310-linux_x86_64.whl
Install test extras from wheel
For Python 3.8
pip install https://github.com/opentensor/cubit/releases/download/v1.1.2/cubit-1.1.2-cp38-cp38-linux_x86_64.whl[test]
For Python 3.9
pip install https://github.com/opentensor/cubit/releases/download/v1.1.2/cubit-1.1.2-cp39-cp39-linux_x86_64.whl[test]
For Python 3.10
pip install https://github.com/opentensor/cubit/releases/download/v1.1.2/cubit-1.1.2-cp310-cp310-linux_x86_64.whl[test]
- cuda-toolkit 11.3 or higher
- nvcc
- gcc (9.3.1 or higher)
- python 3.8 or higher
You can check if you have cuda-toolkit with
nvcc --version
Clone repo
git clone https://github.com/opentensor/cubit.git
Enter dir
cd cubit/
Install as editable
pip install -e .
Install test extras as editable
pip install -e .[test]
Testing uses unittest as there is an issue with pytest and Cython compatability
python3 -m unittest test.py
https://github.com/rmcgibbo/npcuda-example/
https://github.com/mochimodev/cuda-hashing-algos/
https://github.com/camfairchild/bittensor_register_cuda/