Miles v0.1: Production-Level Post-Training
Miles v0.1 is presented, a full-stack, production-ready system for frontier post-training reinforcement-learning training, which supports LoRA RL, on-policy distillation, supervised fine-tuning, and true-on-policy rollout-training alignment, and extends the same architecture to diffusion models.
R. Chen, Ma Cheng, Shi Dong et al.
· 1 citation