Skip to content

wsmoses/keras-benchmarks

 
 

Repository files navigation

keras-benchmarking

Scripts for Keras 3 benchmarking.

Hardware

  • Google Cloud Platform
  • Compute Engine
  • Machine type: a2-highgpu-1g
  • Host RAM: 85GB
  • GPUs: 1 x NVIDIA A100
  • GPU memory: 40GB

Software

  • Python 3.10 Refer to the text files under requirements directory for more detailed on Python package versions for each framework.

Permission setups

Kaggle setup

On the Kaggle Gemma model page, make sure you have accepted the license near the top of the page.

Sign in to Kaggle and go to Settings > API > Create New Token. After clicking, it will download a kaggle.json file.

In the file, you will find your username and key. Append the following lines to your ~/.bashrc file. Make sure you replace the <your_username> and <your_key> with the ones you found in kaggle.json.

export KAGGLE_USERNAME=<your_username>
export KAGGLE_KEY=<your_key>

Running the benchmarks

First, change directory to the root directory of the repository.

cd keras-benchmarking/

Then, create Python vritual environments for all the frameworks under ~/.venv/. Make sure you have pip and venv installed before running the script.

bash shell/install.sh

To run the benchmarks, you can run the following script.

bash shell/run.sh

If you want to remove all the virtual environments afterwards or if you encounter an error want to clean up the half-way installed dependencies, you can run shell/cleanup.sh.

Directories

  • benchmark contains the Python code for benchmarking each model. It is structured as a Python package. I needs pip install -e . before using. Most of the settings are in benchmark/__init__.py. You can run a single benchmark by calling each script, for example, python benchmark/gemma/predict.py results.txt
  • shell contains all the shell scripts for benchmarking.
  • requirements contains the version requirements for the PyPI packages in the dependencies.
  • configs contains the Keras config files for each backend.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 86.3%
  • Shell 13.7%