mcx

Models are JIT-compiled with JAX, enabling transparent execution on CPU, GPU, or TPU for high-speed inference, as highlighted in the README's focus on leveraging modern hardware.

Provides a generator runtime for real-time monitoring and dynamic control of sampling, allowing users to interrupt inference or debug on-the-fly, which is unique among PPLs.

Uses NumPy-like syntax for expressive, reusable model definitions, making it easy to build and share probabilistic programs, as demonstrated in the linear regression example.

Sampling functions support batching for parallel computation, enabling efficient handling of large datasets and vectorized operations, as noted in the key features.