AI Architecture & Systems

Parallel, Blockwise, and Diffusion Language Generation

6.19.1Autoregressive, Semi-Autoregressive, and Non-Autoregressive Generation#

6.19.2Blockwise Parallel Decoding and Iterative Refinement#

6.19.3Set Block Decoding (SBD)#

6.19.4Masked Diffusion Language Modeling#

6.19.5LLaDA, LLaDA 2.0, Dream, and Block Diffusion#

6.19.6Token Update Order, Remasking, and Sampling Steps#

6.19.7Trade-offs Between KV Cache Compatibility and Generation Quality#