Post-training technique speeds up multi-token predictions in language models
Match the Distribution, Not the Compute: Post-Training Multi-Token Prediction Heads
Machine LearningArtificial Intelligence
Summary
Generating several words at once in language models can make them faster, but usually requires heavy training with lots of data. The authors show that a quick extra training step after the main training can achieve similar speedups using far less data. They also suggest smarter ways to check drafts and adjust how many words to predict to keep accuracy while improving speed.
What this means in practice
- •For natural language processing engineers: Improve inference speed of language models by applying a lightweight post-training step to enable multi-token drafts without retraining the entire model.
- •For software developers for ai assistants: Deploy more efficient next-word prediction by dynamically controlling multi-token predictions at runtime to balance speed and accuracy.
Authors
Prachi Badarayani, Aidan Jay, Chenghui Zhou, Dayquan Julienne, Yuan Gao, Tianwei Chen, George Zerveas, Ishmam Zabir, Xiren Zhou, Chris Quirk, Xia Song
Abstract
Multi-token prediction (MTP) improves the throughput of autoregressive generation by enabling the language model to draft multiple next tokens per forward pass, while a verification step over draft tokens ensures that token distribution of the backbone is preserved. Every open MTP-family release (MiMo-7B, DeepSeek-V3, Qwen3) trains its heads jointly with the backbone over the full pretraining run of tens of trillions of tokens, thus setting the drafter quality at pretraining time. We ask whether a lightweight post-training pass on target-generated chain-of-thought is enough to reach the same expected throughput speedup on a frozen reasoning model, and study how a serving-time system built on such a checkpoint can be optimized. We present three findings. 1) On a frozen Qwen3-8B with $K{=}3$ chained MTP heads, we show that a post-training recipe with plain cross-entropy on $\approx\!2.5$B tokens reaches or exceeds the expected speedup of jointly trained MiMo-7B on math, coding and knowledge benchmarks. Our post-training recipe utilizes $10^3$-$10^4\times$ less MTP-training tokens as compared with joint pre-training of MiMO-7B MTP baseline. 2) We propose a chain-aware relaxation of draft token verification rule that allows a bounded drift from backbone language model token distribution. We show that this relaxation lifts expected speedups by $+12$ to $+16\%$ per benchmark while preserving task accuracy. 3) We propose an adaptive controller that dynamically chooses the number of MTP heads to be engaged at inference time and demonstrate recovery of upto $11$--$14\%$ loss in speedup using fixed maximum MTP draft length.