@inproceedings{tan-etal-2026-end,
    title = "End-to-End On-Device Quantization-Aware Training for {LLM}s at Inference Cost",
    author = "Tan, Qitao  and
      Song, Xiaoying  and
      Cheng, Ningxi  and
      Li, Guoming  and
      Liu, Jun  and
      Hong, Lingzi  and
      Ding, Caiwen  and
      Li, Jundong  and
      Zhai, Xiaoming  and
      Huang, Shaoyi  and
      Lu, Jin  and
      Liu, Ninghao  and
      Niu, Wei  and
      Yuan, Geng",
    editor = "Lee, Hung-Yi  and
      Dev, Sunipa  and
      Li, Jessy  and
      Elliott, Desmond",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
    month = oct,
    year = "2026",
    address = "Budapest, Hungary",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2026.findings-emnlp.28/",
    pages = "556--577",
    ISBN = "979-8-89176-462-0"
}