Accelerating Inference of Discrete Autoregressive Normalizing Flows by Selective Jacobi Decoding

Strict sequential inference is unnecessary in parts of autoregressive normalizing flows. We retain sequential decoding for the first layer and apply parallel Jacobi iteration to later layers with more redundant dependencies. The method has convergence guarantees and achieves up to 4.7 times faster inference without retraining or changing the model architecture.

Recommended citation: Zhang, J., Lu, J., Wu, X., Wang, Z., and Zhang, R. (2026). "Accelerating Inference of Discrete Autoregressive Normalizing Flows by Selective Jacobi Decoding." Transactions on Machine Learning Research. Featured Certification.
Download Paper