English

From Dense to Dynamic: Token-Difficulty Driven MoEfication of Pre-Trained LLMs

Computation and Language 2025-02-19 v1

Abstract

Training large language models (LLMs) for different inference constraints is computationally expensive, limiting control over efficiency-accuracy trade-offs. Moreover, once trained, these models typically process tokens uniformly, regardless of their complexity, leading to static and inflexible behavior. In this paper, we introduce a post-training optimization framework, DynaMoE, that adapts a pre-trained dense LLM to a token-difficulty-driven Mixture-of-Experts model with minimal fine-tuning cost. This adaptation makes the model dynamic, with sensitivity control to customize the balance between efficiency and accuracy. DynaMoE features a token-difficulty-aware router that predicts the difficulty of tokens and directs them to the appropriate sub-networks or experts, enabling larger experts to handle more complex tokens and smaller experts to process simpler ones. Our experiments demonstrate that DynaMoE can generate a range of adaptive model variants of the existing trained LLM with a single fine-tuning step, utilizing only 10B10B tokens, a minimal cost compared to the base model's training. Each variant offers distinct trade-offs between accuracy and performance. Compared to the baseline post-training optimization framework, Flextron, our method achieves similar aggregated accuracy across downstream tasks, despite using only 19th\frac{1}{9}\text{th} of their fine-tuning cost.

Keywords

Cite

@article{arxiv.2502.12325,
  title  = {From Dense to Dynamic: Token-Difficulty Driven MoEfication of Pre-Trained LLMs},
  author = {Kumari Nishu and Sachin Mehta and Samira Abnar and Mehrdad Farajtabar and Maxwell Horton and Mahyar Najibi and Moin Nabi and Minsik Cho and Devang Naik},
  journal= {arXiv preprint arXiv:2502.12325},
  year   = {2025}
}
R2 v1 2026-06-28T21:47:56.986Z