This directory contains a collection of Mojo notebooks designed to compare the accuracy and runtime performance of functions from the Specials package against those found in well-known Mojo and Python packages.
The benchmarks aim to highlight the correctness and efficiency of our implementations.
Please note that for us, Accuracy > Performance: when forced to choose between FLOPS and numerical accuracy, we always prefer numerical accuracy.
Before running the benchmarks, ensure you have completed the initial setup for the project. This includes installing Mojo, cloning the Specials repository, and setting up and activating a Python environment with Conda. Detailed setup instructions are available here.
Once the initial setup is complete, build the Specials package to ensure all components are ready for benchmarking. Run the following command in the specials/benchmarks directory of the cloned repository:
../scripts/prepare-benchmarks.shThe benchmarks can be run in Visual Studio Code or JupyterLab with Mojo kernel support. Follow the respective guides to set up your environment.
Once set up, open the desired notebook in the specials/benchmarks directory to begin running the benchmarks.