DynaMo: Accelerating Language Model Inference with Dynamic Multi-Token Sampling
This work addresses the inference efficiency bottleneck for users of large language models, offering a novel method to accelerate generation with incremental improvements over existing techniques.
The authors tackled the problem of high inference times in large language models by proposing DynaMo, a suite of multi-token prediction models that dynamically predict multiple tokens based on confidence, achieving a 2.57× speed-up with minimal parameter and training overheads while maintaining text quality comparable to a baseline model.
Traditional language models operate autoregressively, i.e., they predict one token at a time. Rapid explosion in model sizes has resulted in high inference times. In this work, we propose DynaMo, a suite of multi-token prediction language models that reduce net inference times. Our models $\textit{dynamically}$ predict multiple tokens based on their confidence in the predicted joint probability distribution. We propose a lightweight technique to train these models, leveraging the weights of traditional autoregressive counterparts. Moreover, we propose novel ways to enhance the estimated joint probability to improve text generation quality, namely co-occurrence weighted masking and adaptive thresholding. We also propose systematic qualitative and quantitative methods to rigorously test the quality of generated text for non-autoregressive generation. One of the models in our suite, DynaMo-7.3B-T3, achieves same-quality generated text as the baseline (Pythia-6.9B) while achieving 2.57$\times$ speed-up with only 5.87% and 2.67% parameter and training time overheads, respectively.