Fetching the paper…

FR-Spec: Accelerating Large-Vocabulary Language Models via Frequency-Ranked Speculative Sampling · Around