TL;DR: A new research from Apple, formalizes what “mid-training” should do before reinforcement learning RL post-training and introduces RA3 (Reasoning as Action Abstractions)—an EM-style procedure that learns temporally consistent latent actions from expert traces, then fine-tunes on those bootstrapped traces. It shows mid-training should (1) prune to a compact near-optimal action subspace and (2) shorten […]
The post RA3: Mid-Training with Temporal Action Abstractions for Faster Reinforcement Learning (RL) Post-Training in Code LLMs appeared first on MarkTechPost.