NVIDIA/framework-reproducibility
NVIDIA tool providing determinism patches and variance-reduction utilities for achieving bit-accurate reproducibility in PyTorch and TensorFlow frameworks.
Not currently ranked — collecting fresh signals.
star history
This repository provides documentation, status tracking, patches, and utilities for achieving run-to-run reproducibility in deep learning frameworks, with a focus on GPU determinism. It includes a Seeder tool designed to reduce variance in training runs. The project maintains compatibility with both PyTorch and TensorFlow ecosystems.
Frequently asked
- What is NVIDIA/framework-reproducibility?
- NVIDIA tool providing determinism patches and variance-reduction utilities for achieving bit-accurate reproducibility in PyTorch and TensorFlow frameworks.
- Is framework-reproducibility open source?
- Yes — NVIDIA/framework-reproducibility is open source, released under the Apache-2.0 license.
- What language is framework-reproducibility written in?
- NVIDIA/framework-reproducibility is primarily written in Python.
- How popular is framework-reproducibility?
- NVIDIA/framework-reproducibility has 431 stars on GitHub.
- Where can I find framework-reproducibility?
- NVIDIA/framework-reproducibility is on GitHub at https://github.com/NVIDIA/framework-reproducibility.