Adaptive Multi-Token Prediction in Language Models
The study examines how multi-token prediction horizons can adapt dynamically to context to optimize language model inference efficiency.
The study examines how multi-token prediction horizons can adapt dynamically to context to optimize language model inference efficiency.