Clone Cerebras Model Zoo#
Cerebras Model Zoo repository contains examples of common deep learning models that can be trained on the Cerebras Wafer-Scale Cluster.
This repository can be used for:
Getting started. It contains:
Reference implementations in PyTorch and TensorFlow of common deep learning models.
Data preparation scripts to use your own data.
Configuration files for multiple model sizes.
- Benchmarking. It contains implementations of common models optimized for Cerebras hardware.
Developing new models. It contains:
Cerebras-customized PyTorch and Tensorflow Layer APIs that simplify the compilation process.
Examples that demonstrate best practices for coding your own model.
Note
The Cerebras Model Zoo is available at https://github.com/Cerebras/modelzoo and a list of supported models can be found here.
To clone the Cerebras Model Zoo Repository in your preferred directory, use:
git clone https://github.com/Cerebras/modelzoo.git
Note
If your user node requires a proxy to access to external packages, you will need to configure a proxy in your git configuration to clone any repository.
Setup environment for Cerebras Model Zoo#
After Setup Cerebras virtual environment, you will need to install additional packages to your Python environment to support all functionality of Cerebras Model Zoo
For the desired environment, please install the packages available in Cerebras Model Zoo by running:
source venv_cerebras_pt/bin/activate
cd modelzoo
pip install -r requirements_cb_pt.txt
source venv_cerebras_tf/bin/activate
cd modelzoo
pip install --no-deps -r requirements_cb_tf.txt
# the --no-deps above is intentional due to conflicts in protobuf version between tfds and cerebras_tensorflow