Publications

To appear in Findings of EMNLP 2026.

DPA-MoE is a runtime precision control framework that dynamically allocates different precision levels to the routed experts of a mixture-of-experts model. Because the allocation happens at inference time, a single checkpoint supports flexible weight-size–quality trade-offs with no retraining and no per-precision copy of the weights.

Jaeyun Kim and Hoonki Lee contributed equally; Sangyeob Kim is the corresponding author.

@inproceedings{kim2026dpamoe,
  title     = {{DPA}-MoE: A Runtime Precision Control Framework for {M}o{E} Inference},
  author    = {Kim, Jaeyun and Lee, Hoonki and Shin, Susie and Kim, Hyunmin and
               Lee, Dahyun and Kim, Sangyeob},
  booktitle = {Findings of the 2026 Conference on Empirical Methods in Natural Language Processing},
  year      = {2026},
  url       = {https://openreview.net/forum?id=BrcO2Gg9A9}
}