Ye-Qing Zhang

Accelerating Disaggregated RL for Visual Generative LLMs with Diffusion-Based Parallelism and Trainer-Assisted Generation

DigenRL is introduced, a disaggregated RL framework for diffusion-based generative LLMs that supports flexible resource allocation, accommodates heterogeneous GPUs, and facilitates efficient task scheduling and an elastic trainer-assisted generation approach to enable the trainer GPU resources to dynamically assist in...