This package provides patches and tools related to determinism (bit-accurate, run-to-run reproducibility) in deep learning frameworks, with a focus on determinism when running on GPUs, and a tool (Seeder) for reducing variance in deep learning frameworks.
For further information, see the documentation in the associated open-source repository: GitHub/NVIDIA/framework-reproducibility