AI Architecture & Systems

Diffusion and Non-Autoregressive Model Systems

5.33.1Denoising Steps, Schedulers, and Multi-Stage Execution#

5.33.2CFG Parallelism and Model Parallelism#

5.33.3Sequence/Patch/Temporal Parallelism#

5.33.4Cross-Step Feature Cache and Compute Reuse#

5.33.5Block Scheduling and Caching for Diffusion LLMs#

5.33.6Memory, Throughput, and Service Orchestration for Image/Video Generation#