JAX
NumPy-like array library with automatic differentiation and XLA compilation.
Overview
JAX provides composable function transformations: automatic differentiation, vectorisation and just-in-time compilation to XLA, all applicable to ordinary Python functions. It solves the problem that research code needs to be differentiated and accelerated without being rewritten around a framework's graph API. It exists because functional composition turns out to be a cleaner foundation for large-scale training, and it underlies several current model families.
Technology Graph
How JAX connects to the rest of the stack.
Uses and depends on
- NVIDIA CUDA platform Hardware JAX uses NVIDIA CUDA High ✓ source
Applied to
- Deep Learning concept Algorithms JAX used for Deep Learning Medium ✓ source
Tools
Software that implements or integrates with JAX.