Elamur Research
☰
home / papers / 2023 / DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models

DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models

Written for the Research library (claude-opus-5-5) from the paper text, in our own words: the license "arXiv — non-exclusive license to distribute" does not allow redistributing the text itself. Original — arxiv.org/pdf/2309.03883

Summary

Large language models often produce statements that conflict with facts they saw during pretraining, which limits their use in high-stakes domains. Chuang et al. propose DoLa (Decoding by Contrasting Layers), a decoding-time method that needs neither retrieval of outside knowledge nor extra fine-tuning. It builds on interpretability findings that factual knowledge tends to be concentrated in particular, mostly upper, transformer layers.

At each generation step, the vocabulary head is applied to intermediate hidden states (early exit). The method then dynamically picks a "premature" layer whose prediction diverges most from the final "mature" layer, using Jensen-Shannon Divergence as the measure. The next-token score is the log-space difference between the mature and premature distributions, which amplifies information that only appears late in the network.

On LLaMA-family models, DoLa raises TruthfulQA performance by 12–17 absolute points. It also helps on FACTOR, on chain-of-thought tasks (StrategyQA, GSM8K) and in GPT-4-judged open-ended chat evaluation, while adding only minor latency. This makes it a cheap, plug-in way to make existing models more factual.

Key points

  • A preliminary analysis on 32-layer LLaMA-7B examined JSD between early-exit and final-layer outputs. For named entities and dates (e.g., "Wole Soyinka", "1986"), the divergence stays high even in upper layers. For function words and tokens copied from the input, it drops close to zero from the middle layers onward.
  • The premature layer is not fixed. It is reselected for every token from a candidate set of layers, choosing the one with the largest JSD from the mature layer.
  • The main result is a 12–17 absolute-point gain on TruthfulQA for LLaMA models, with no external knowledge and no training.
  • Evaluation covered:
    • multiple-choice factuality (TruthfulQA, FACTOR);
    • reasoning (StrategyQA, GSM8K);
    • GPT-4 ratings in the chatbot evaluation setup of Chiang et al. (2023).
  • The code is released at github.com/voidism/DoLa. The work comes from MIT and Microsoft.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Other

Cited by

239 papers of our library cite it (of the 140673 whose bibliographies we read); 8 of them have a page here, 231 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2309.03883 · arXiv 2309.03883

@misc{chuang2023dola,
  title = {{DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models}},
  author = {Chuang, Yung-Sung and Xie, Yujia and Luo, Hongyin and Kim, Yoon and Glass, James and He, Pengcheng},
  year = {2023},
  doi = {10.48550/arxiv.2309.03883},
  eprint = {2309.03883},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2309.03883}
}

References

44 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Sebastian Borgeaud, Arthur Mensch, Jordan Hoffmann, Trevor Cai, Eliza Rutherford, Katie Millican, George Bm Van Den Driessche, Jean-Baptiste Lespiau, Bogdan Damoc, Aidan Clark, et al. Improving language models by retrieving from trillions of tokens. In International conference on machine learning, p… find
  2. Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, Chri… find
  3. Curran Associates, Inc., 2020. URL https://proceedings.neurips.cc/paper files/paper/2020/file/ 1457c0d6bfcb4967418bfb8ac142f64a-Paper.pdf. ↗
  4. Cheng-Han Chiang and Hung-yi Lee. Can large language models be an alternative to human evaluations? arXiv preprint arXiv:2305.01937, 2023a. find
  5. Cheng-Han Chiang and Hung-yi Lee. A closer look into automatic evaluation using large language models. arXiv preprint arXiv:2310.05657, 2023b. — [arXiv:2310.05657](https://arxiv.org/abs/2310.05657) find
  6. Wei-Lin Chiang, Zhuohan Li, Zi Lin, Ying Sheng, Zhanghao Wu, Hao Zhang, Lianmin Zheng, Siyuan Zhuang, Yonghao Zhuang, Joseph E. Gonzalez, Ion Stoica, and Eric P. Xing. Vicuna: An open-source chatbot impressing gpt-4 with 90%* chatgpt quality, March 2023. URL https://lmsys.org/blog/2023-03-30-vicuna/… ↗
  7. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, et al. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168, 2021. find our card
  8. Damai Dai, Li Dong, Yaru Hao, Zhifang Sui, Baobao Chang, and Furu Wei. Knowledge neurons in pre-trained transformers. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 8493–8502, 2022. find
  9. Yilun Du, Shuang Li, Antonio Torralba, Joshua B Tenenbaum, and Igor Mordatch. Improving factuality and reasoning in language models through multiagent debate. arXiv preprint arXiv:2305.14325, 2023. find our card
  10. Maha Elbayad, Jiatao Gu, Edouard Grave, and Michael Auli. Depth-adaptive transformer. In ICLR 2020-Eighth International Conference on Learning Representations, pp. 1–14, 2020. find
  11. Mohsen Fayyaz, Ehsan Aghazadeh, Ali Modarressi, Hosein Mohebbi, and Mohammad Taher Pilehvar. Not all models localize linguistic knowledge in the same place: A layer-wise probing on bertoids’ representations. In Proceedings of the Fourth BlackboxNLP Workshop on Analyzing and Interpreting Neural Netwo… find
  12. Xinyang Geng and Hao Liu. Openllama: An open reproduction of llama, May 2023. URL https://github.com/openlm-research/open llama. ↗
  13. Ariel Gera, Roni Friedman, Ofir Arviv, Chulaka Gunasekara, Benjamin Sznajder, Noam Slonim, and Eyal Shnarch. The benefits of bad advice: Autocontrastive decoding across model layers. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), p… find
  14. Mor Geva, Daniel Khashabi, Elad Segal, Tushar Khot, Dan Roth, and Jonathan Berant. Did aristotle use a laptop? a question answering benchmark with implicit reasoning strategies. Transactions of the Association for Computational Linguistics, 9:346–361, 2021. find
  15. Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pp. 770–778, 2016. find
  16. Gautier Izacard, Patrick Lewis, Maria Lomeli, Lucas Hosseini, Fabio Petroni, Timo Schick, Jane Dwivedi-Yu, Armand Joulin, Sebastian Riedel, and Edouard Grave. Few-shot learning with retrieval augmented language models. arXiv preprint arXiv:2208.03299, 2022. find
  17. Ziwei Ji, Nayeon Lee, Rita Frieske, Tiezheng Yu, Dan Su, Yan Xu, Etsuko Ishii, Ye Jin Bang, Andrea Madotto, and Pascale Fung. Survey of hallucination in natural language generation. ACM Computing Surveys, 55(12):1–38, 2023. find our card
  18. Wei-Tsung Kao, Tsung-Han Wu, Po-Han Chi, Chun-Cheng Hsieh, and Hung-Yi Lee. Bert’s output layer recognizes all hidden layers? some intriguing phenomena and a simple way to boost bert. arXiv preprint arXiv:2001.09309, 2020. — [arXiv:2001.09309](https://arxiv.org/abs/2001.09309) find
  19. Nitish Shirish Keskar, Bryan McCann, Lav R Varshney, Caiming Xiong, and Richard Socher. Ctrl: A conditional transformer language model for controllable generation. arXiv preprint arXiv:1909.05858, 2019. find our card
  20. Kenneth Li, Oam Patel, Fernanda Viégas, Hanspeter Pfister, and Martin Wattenberg. Inference-time intervention: Eliciting truthful answers from a language model. arXiv preprint arXiv:2306.03341, 2023. find our card
  21. Xiang Lisa Li, Ari Holtzman, Daniel Fried, Percy Liang, Jason Eisner, Tatsunori Hashimoto, Luke Zettlemoyer, and Mike Lewis. Contrastive decoding: Open-ended text generation as optimization. arXiv preprint arXiv:2210.15097, 2022. find our card
  22. Tian Liang, Zhiwei He, Wenxiang Jiao, Xing Wang, Yan Wang, Rui Wang, Yujiu Yang, Zhaopeng Tu, and Shuming Shi. Encouraging divergent thinking in large language models through multi-agent debate. arXiv preprint arXiv:2305.19118, 2023. find our card
  23. Stephanie Lin, Jacob Hilton, and Owain Evans. Truthfulqa: Measuring how models mimic human falsehoods. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 3214–3252, 2022. find our card
  24. Yang Liu, Dan Iter, Yichong Xu, Shuohang Wang, Ruochen Xu, and Chenguang Zhu. G-eval: Nlg evaluation using gpt-4 with better human alignment. arXiv preprint arXiv:2303.16634, 2023. find
  25. Potsawee Manakul, Adian Liusie, and Mark JF Gales. Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models. arXiv preprint arXiv:2303.08896, 2023. find
  26. Kevin Meng, David Bau, Alex Andonian, and Yonatan Belinkov. Locating and editing factual associations in GPT. Advances in Neural Information Processing Systems, 36, 2022. find
  27. NLP Team MosaicML. Introducing mpt-7b: A new standard for open-source, commercially usable llms, 2023. URL www.mosaicml.com/blog/mpt-7b. Accessed: 2023-05-05. find
  28. Dor Muhlgay, Ori Ram, Inbal Magar, Yoav Levine, Nir Ratner, Yonatan Belinkov, Omri Abend, Kevin Leyton-Brown, Amnon Shashua, and Yoav Shoham. Generating benchmarks for factuality evaluation of language models. arXiv preprint arXiv:2307.06908, 2023. find
  29. Jingcheng Niu, Wenjie Lu, and Gerald Penn. Does bert rediscover a classical nlp pipeline? In Proceedings of the 29th International Conference on Computational Linguistics, pp. 3143–3153, 2022. find
  30. Sean O’Brien and Mike Lewis. Contrastive decoding improves reasoning in large language models. arXiv preprint arXiv:2309.09117, 2023. find
  31. OpenAI. Introducing chatgpt, November 2022. URL https://openai.com/blog/chatgpt. ↗
  32. OpenAI. Gpt-4 technical report. 2023. URL https://cdn.openai.com/papers/gpt-4.pdf. ↗ our card
  33. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, et al. Training language models to follow instructions with human feedback. Advances in Neural Information Processing Systems, 35:27730–27744, 2022. find
  34. Ori Ram, Yoav Levine, Itay Dalmedigos, Dor Muhlgay, Amnon Shashua, Kevin Leyton-Brown, and Yoav Shoham. In-context retrieval-augmented language models. arXiv preprint arXiv:2302.00083, 2023. find
  35. Erik Tjong Kim Sang and Fien De Meulder. Introduction to the conll-2003 shared task: Language-independent named entity recognition. In Proceedings of the Seventh Conference on Natural Language Learning at HLT-NAACL 2003, pp. 142–147, 2003. find
  36. Tal Schuster, Adam Fisch, Jai Gupta, Mostafa Dehghani, Dara Bahri, Vinh Tran, Yi Tay, and Donald Metzler. Confident adaptive language modeling. Advances in Neural Information Processing Systems, 35:17456– 17472, 2022. find
  37. Weijia Shi, Xiaochuang Han, Mike Lewis, Yulia Tsvetkov, Luke Zettlemoyer, and Scott Wen-tau Yih. Trusting your evidence: Hallucinate less with context-aware decoding. arXiv preprint arXiv:2305.14739, 2023. find our card
  38. Surat Teerapittayanon, Bradley McDanel, and Hsiang-Tsung Kung. Branchynet: Fast inference via early exiting from deep neural networks. In 2016 23rd International Conference on Pattern Recognition (ICPR), pp. 2464–2469. find
  39. Ian Tenney, Dipanjan Das, and Ellie Pavlick. Bert rediscovers the classical nlp pipeline. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pp. 4593–4601, 2019. find our card
  40. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023. find
  41. Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, et al. Emergent abilities of large language models. Transactions on Machine Learning Research, 2022a. find our card
  42. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Ed Chi, Quoc Le, and Denny Zhou. Chain of thought prompting elicits reasoning in large language models. arXiv preprint arXiv:2201.11903, 2022b. find our card
  43. Mengzhou Xia, Tianyu Gao, Zhiyuan Zeng, and Danqi Chen. Sheared llama: Accelerating language model pre-training via structured pruning. arXiv preprint arXiv:2310.06694, 2023. find
  44. Jin Xu, Xiaojiang Liu, Jianhao Yan, Deng Cai, Huayang Li, and Jian Li. Learning to break the loop: Analyzing and mitigating repetitions for neural text generation. Advances in Neural Information Processing Systems, 35:3082–3095, 2022. find