PyJuice Pushes HMMs and Image Models Beyond State-of-the-Art

PyJuice dramatically accelerates the training and inference of probabilistic circuits (PCs), outperforming established frameworks like SPFlow, EiNet, and Juice.jl across multiple benchmarks. Leveraging GPU efficiency, PyJuice handles billion-edge models without exhausting memory, scales sparse and block-sparse architectures, and enables fine-tuning of large HMMs and image models that were previously impractical. The results show PyJuice not only reduces runtime by orders of magnitude but also advances the state-of-the-art in PC modeling at scale.