BlackJAX

Leverages JAX's XLA compilation for seamless GPU/TPU support, enabling significantly faster sampling on modern hardware, as highlighted in the quickstart example with NUTS.

Provides building blocks like integrators and proposals that can be combined to create custom algorithms, facilitating flexible research and development, as explained in the philosophy section.

Includes well-tested, performant samplers such as NUTS for immediate use, with examples in the documentation showing easy integration into inference loops.

Accelerates experimentation by exposing low-level internals and integrates seamlessly with probabilistic programming languages that supply compatible log-density functions, as stated in the key features.