google/jaxopt

Hardware accelerated, batchable and differentiable optimizers in JAX.

What it solves

JAXopt provides a library of optimizers that are hardware accelerated, batchable, and differentiable. It allows users to solve optimization problems on GPUs and TPUs, and differentiate the solutions of these problems with respect to their inputs.

How it works

Built on top of JAX, the library leverages JAX's vmap for automatic vectorization to handle multiple instances of the same optimization problem simultaneously. It supports differentiation of optimization solutions through either implicit differentiation or by using autodiff on unrolled algorithm iterations.

Who it’s for

Researchers and developers working with JAX who need high-performance optimization tools that can be integrated into differentiable programming workflows.

Highlights

  • Hardware acceleration for CPU, GPU, and TPU.
  • Automatic vectorization via JAX's vmap for batching optimization problems.
  • Support for both implicit and unrolled differentiation of optimization solutions.

Related

  • Project
    Kotlin/dataframe

    A type-safe, in-memory tabular data processing library for the JVM that provides compile-time column access and functional transformations.

  • Project
    sybil-solutions/codex-shim

    codex‑shim is a cross‑platform Python/aiohttp shim that makes Codex Desktop think it’s talking to the official OpenAI Responses API, while actually routing each request to any user‑provided LLM (OpenAI, Anthropic, Ollama, OpenCode Go, etc.) or to ChatGPT’s Codex backend. It lets you add BYOK models to Codex’s picker, keep all Codex UI features (function calls, streaming, images), and optionally auto‑route tasks to the cheapest capable model.

  • Project
    kernel/kernel-images

    A project providing sandboxed, ready-to-use Chrome browsers for browser automation and AI web agents, featuring remote GUI access and ultra-fast unikernel-based restarts.

  • Project
    keslr/keslr_connect

    A set of TypeScript packages for integrating with the Keslr network to ensure applications are accessed only by verified humans via trust-graph authentication and private network addressing.