JAXopt
Hardware-accelerated, batchable, and differentiable optimization algorithms implemented in JAX for machine learning research.
What is JAXopt?
JAXopt is a library of optimization algorithms implemented in Google's JAX framework, providing hardware-accelerated, batchable, and differentiable solvers for machine learning and scientific computing. It enables researchers to solve optimization problems efficiently on GPUs and TPUs while maintaining differentiability for gradient-based learning pipelines.
Machine learning researchers and practitioners working with JAX who need efficient, differentiable optimization routines for training models, hyperparameter tuning, or solving implicit equations.
JAXopt offers GPU/TPU-native optimization algorithms with automatic vectorization and differentiation capabilities, making it uniquely suited for modern deep learning workflows where optimization is part of a larger differentiable computation graph.
Overview
Hardware accelerated, batchable and differentiable optimizers in JAX.
Use Cases
Best For
- Training machine learning models with custom optimization algorithms
- Hyperparameter optimization with gradient-based methods
- Solving implicit equations in differentiable programs
- Research requiring GPU-accelerated optimization routines
- Batch processing of multiple optimization problems simultaneously
- Implementing optimization layers in neural networks
Not Ideal For
- Projects requiring actively maintained libraries with ongoing updates and support
- Teams not already using JAX in their machine learning stack
- Production systems where long-term stability and bug fixes are critical
- Users needing out-of-the-box optimizers without deep integration into JAX's functional model
Pros & Cons
Pros
Optimizers run efficiently on GPU, TPU, and CPU, enabling fast computation for large-scale ML tasks, as highlighted in the README's focus on hardware acceleration.
Supports automatic vectorization using JAX's vmap, allowing parallel processing of multiple optimization problems, which is a key feature mentioned for efficiency.
Solutions can be differentiated implicitly or via autodiff, facilitating gradient-based learning in end-to-end pipelines, as emphasized in the differentiable optimizers section.
Provides an advanced framework for efficient gradient computation through fixed-point equations, described in the linked paper for modular implicit differentiation.
Cons
The library is no longer maintained or developed, as stated in the README's status section, meaning no future updates, bug fixes, or new features.
Requires familiarity with JAX's functional programming model, which has a steep learning curve and limits usability outside JAX-based workflows.
With maintenance ceased, community support and third-party integrations are likely minimal, and alternatives like optax have absorbed some features.
Frequently Asked Questions
Related Projects
ALXGoogle Research
OryxProbabilistic reasoning and statistical analysis in TensorFlow
BRAXMassively parallel rigidbody physics simulation on accelerator hardware.
MctxMonte Carlo tree search in JAX
Found a gem we're missing?
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.