jax-ml
JAX is a high-performance numerical computing library that brings together Autograd and XLA for composable program transformations.
At a Glance
- AI Researchers
- Machine Learning Engineers
- Scientific Researchers
- Enterprise AI Teams
AI Tools by jax-ml
(1)JAX
NumPy ML Transformation Library
Discussions
No discussions yet
Be the first to start a discussion about jax-ml
Latest News
Google DeepMind announces Gemini, trained using JAX and TPUs.
Using JAX to accelerate our research at DeepMind.
Google Cloud announces A3 VMs and JAX optimizations at I/O.
JAX 0.4.26 released with enhanced Pallas support for custom kernels.
Products & Services
Core library for composable transformations of Python+NumPy programs (grad, jit, vmap, pmap).
A flexible and efficient neural network library for JAX.
A gradient processing and optimization library for JAX.
A simple, object-oriented neural network library for JAX (developed by DeepMind).
Market Position
Positions itself as a more flexible and research-oriented alternative to PyTorch and TensorFlow, leveraging XLA for superior performance on TPUs and GPUs.
Leadership
Founders
James Bradbury
Software Engineer at Google Research. Previously a researcher at Salesforce. Key author of the original JAX repository.
Roy Frostig
Research Scientist at Google Research. PhD from Stanford University. Focused on machine learning and optimization.
Matthew James Johnson
Research Scientist at Google Research. Postdoctoral researcher at Harvard University. PhD from MIT. Lead developer of JAX.
Peter Hawkins
Software Engineer at Google Research. PhD from Australian National University. Expert in XLA and compiler optimization.
Executive Team
Matthew Johnson
Software Engineer and JAX Project Lead
Core developer of Autograd and JAX at Google Research.
Roy Frostig
Research Scientist
Leading researcher in ML optimization and JAX core developer at Google.
Board of Directors
Founding Story
JAX was started at Google Brain to combine the flexibility of Autograd with the performance of XLA, specifically for research into neural network optimization and large-scale ML.
Business Model
Revenue Model
Open Source (Apache 2.0). Indirect revenue generated through Google Cloud Platform (TPU/GPU usage) and internal efficiency at Google.
Pricing Tiers
Full access to code and documentation under Apache 2.0 license.
Pay-as-you-go pricing for hardware acceleration optimized for JAX.
Target Markets
- AI Researchers
- Machine Learning Engineers
- Scientific Researchers
- Enterprise AI Teams
- Deep Learning Research
- Scientific Computing
- Large-scale Model Training
- Differential Privacy
- DeepMind
- Google Brain
- OpenAI
- Midjourney