Taking some Huggingface datasets and transformers for a tour.
- The Mistral AI 7B base model, see https://huggingface.co/mistralai/Mistral-7B-v0.1
- The Llama2 models from Meta, see https://huggingface.co/meta-llama
This repo uses Python 3.11, PyTorch 2.1.0, Huggingface Transformers 4.34 or newer, and JupyterLab notebooks.
We use the transformers with PyTorch. If you are using Conda, install it like this:
conda install -c pytorch -c nvidia pytorch pytorch-cuda
At the time of writing, the Mistral AI Base Model requires Transformers 4.34, which is not available for
Conda on the main and huggingface channels, but only on conda-forge. You can install it like this:
conda install -c conda-forge 'transformers=4.34.0' accelerate=0.23.0 safetensors
We install the optional transformers dependency accelerate to speed up running on GPUs as it allows the model to be
mapped to CPU and GPU as memory permits. It in turn, requires safetensors.
MIT License. See the LICENSE file for details.