Communication-Budgeted Chunk-Aware Mixed Model Parallelism for Collaborative Inference

Splits a model across devices with chunk-aware mixed parallelism under a fixed communication budget for collaborative inference.