NVIDIA/framework-reproducibility

Providing reproducibility in deep learning frameworks

50
/ 100
Established

When training deep learning models, slight variations in results between runs can make it hard to confidently compare different models or hyperparameter settings. This project helps scientists and machine learning engineers achieve consistent, bit-accurate results from their deep learning model training, especially on GPUs. It provides documentation, patches, and tools to ensure that if you run the same training setup twice, you'll get the exact same outcome.

434 stars. No commits in the last 6 months. Available on PyPI.

Use this if you are a deep learning practitioner struggling with non-deterministic model training runs and need to ensure identical, reproducible results for rigorous scientific comparison or debugging.

Not ideal if you are looking for general code reproducibility across different computing environments rather than bit-accurate run-to-run consistency within deep learning frameworks.

deep-learning-research model-training scientific-reproducibility GPU-acceleration machine-learning-engineering
Stale 6m No Dependents
Maintenance 0 / 25
Adoption 10 / 25
Maturity 25 / 25
Community 15 / 25

How are scores calculated?

Stars

434

Forks

38

Language

Python

License

Apache-2.0

Last pushed

May 13, 2024

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/NVIDIA/framework-reproducibility"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.