Aurora
Together AI Blog
•
Reinforcement Learning
1.25x over a well-trained static speculator. Aurora is an open-source RL framework that turns speculative decoding from a one-time offline setup into a self-improving system that learns from every request it serves.