https://github.com/RedKinda/mojo-benchmarks

https://github.com/RedKinda/mojo-benchmarks

Last synced: about 2 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/RedKinda/mojo-benchmarks
• Owner: RedKinda
• License: mit
• Created: 2024-04-08T17:24:28.000Z (7 months ago)
• Default Branch: master
• Last Pushed: 2024-07-07T13:30:55.000Z (4 months ago)
• Last Synced: 2024-07-07T14:48:35.158Z (4 months ago)
• Language: Jupyter Notebook
• Size: 9.86 MB
• Stars: 0
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-max-mojo - RedKinda/mojo-benchmarks - benchmarks?style=social"/> : mojo-benchmarks. (Performance Benchmark)
• awesome-max-mojo - RedKinda/mojo-benchmarks - benchmarks?style=social"/> : mojo-benchmarks. (Performance Benchmark)

README

        
# Benchmarking Mojo

*Note: this project is part of Bachelor's thesis of Red Kalab in 2024. Full work can be found [here](https://www.overleaf.com/read/yyzgdkcqhxhn#88d801).*

## Instructions for running

Build the Dockerfile with command such as `docker build . -t mojo-bench`. Image will take ~5GB on disk. 

To run full benchmarks, with each alg+bench taking ~5 minutes run `docker run -v ./bench_times:/app/bench_times mojo-bench ./_bench.py full run_name 300`. 

This configuration would take approximately 10 hours in total. Resulting benchmark timings will be saved in `./bench_times/run_name`. 

These results can be copied into `./results/data` and evaluated with the notebook in the `./results` folder.

## Customization

The bench utility takes 3 parameters, benchmarks to run, run name and time to run. Valid options for benchmarks to run are `full`, `all`, `softmax`, `matmul`, `crc16` and `quicksort`.

- `full` - runs all benchmarks on all defined sizes

- `all` - runs all benchmarks on default size

- `softmax`/`matmul`/`crc16`/`quicksort` - runs given algorithm on default size

Run name determines the folder name of where the results are stored. Time to run signifies time to run for each algorithm+size combination. 

The `_bench.py` script contains the default/full sizes for benchmarking sizes, and is fully customizable.