Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
144 tokens/sec
GPT-4o
8 tokens/sec
Gemini 2.5 Pro Pro
46 tokens/sec
o3 Pro
4 tokens/sec
GPT-4.1 Pro
38 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Does Instruction Tuning Make LLMs More Consistent? (2404.15206v3)

Published 23 Apr 2024 in cs.CL

Abstract: The purpose of instruction tuning is enabling zero-shot performance, but instruction tuning has also been shown to improve chain-of-thought reasoning and value alignment (Si et al., 2023). Here we consider the impact on $\textit{consistency}$, i.e., the sensitivity of LLMs to small perturbations in the input. We compare 10 instruction-tuned LLaMA models to the original LLaMA-7b model and show that almost across-the-board they become more consistent, both in terms of their representations and their predictions in zero-shot and downstream tasks. We explain these improvements through mechanistic analyses of factual recall.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (39)
  1. A large annotated corpus for learning natural language inference. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 632–642, Lisbon, Portugal. Association for Computational Linguistics.
  2. Evaluating Predictive Uncertainty, Visual Objects Classification and Recognising Textual Entailment: Selected Proceedings of the First PASCAL Machine Learning Challenges Workshop.
  3. Vicuna: An open-source chatbot impressing gpt-4 with 90%* chatgpt quality.
  4. Chatbot arena: An open platform for evaluating llms by human preference. arXiv preprint arXiv:2403.04132.
  5. Databricks. 2023. Free dolly: Introducing the world’s first truly open instruction-tuned llm. Blog post.
  6. Bill Dolan and Chris Brockett. 2005a. Automatically constructing a corpus of sentential paraphrases. In Third International Workshop on Paraphrasing (IWP2005).
  7. William B. Dolan and Chris Brockett. 2005b. Automatically constructing a corpus of sentential paraphrases. In Proceedings of the Third International Workshop on Paraphrasing (IWP2005).
  8. Measuring and improving consistency in pretrained language models. Transactions of the Association for Computational Linguistics, 9:1012–1031.
  9. Constanza Fierro and Anders Søgaard. 2022. Factual consistency of multilingual pretrained language models. In Findings of the Association for Computational Linguistics: ACL 2022, pages 3046–3052, Dublin, Ireland. Association for Computational Linguistics.
  10. Dissecting recall of factual associations in auto-regressive language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 12216–12235, Singapore. Association for Computational Linguistics.
  11. Wes Gurnee and Max Tegmark. 2023. Language models represent space and time. arXiv preprint arXiv:2310.02207.
  12. The effect of scaling, retrieval augmentation and form on the factual consistency of language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 5457–5476, Singapore. Association for Computational Linguistics.
  13. Measuring massive multitask language understanding. In International Conference on Learning Representations.
  14. Unnatural instructions: Tuning language models with (almost) no human labor.
  15. BECEL: Benchmark for consistency evaluation of language models. In Proceedings of the 29th International Conference on Computational Linguistics, pages 3680–3696, Gyeongju, Republic of Korea. International Committee on Computational Linguistics.
  16. Myeongjun Jang and Thomas Lukasiewicz. 2023a. Consistency analysis of ChatGPT. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 15970–15985, Singapore. Association for Computational Linguistics.
  17. Myeongjun Jang and Thomas Lukasiewicz. 2023b. Improving language models’ meaning understanding and consistency by learning conceptual roles from dictionary. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 8496–8510, Singapore. Association for Computational Linguistics.
  18. BeliefBank: Adding memory to a pre-trained language model for a systematic notion of belief. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 8849–8861, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics.
  19. Openassistant conversations – democratizing large language model alignment.
  20. Alpacaeval: An automatic evaluator of instruction-following models. https://github.com/tatsu-lab/alpaca_eval.
  21. The unlocking spell on base llms: Rethinking alignment via in-context learning. arXiv preprint arXiv:2312.01552.
  22. The flan collection: Designing data and methods for effective instruction tuning. arXiv preprint arXiv:2301.13688.
  23. Training language models to follow instructions with human feedback. Advances in Neural Information Processing Systems, 35:27730–27744.
  24. Instruction tuning with gpt-4. arXiv preprint arXiv:2304.03277.
  25. Mohammad Taher Pilehvar and Jose Camacho-Collados. 2019. WiC: the word-in-context dataset for evaluating context-sensitive meaning representations. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 1267–1273, Minneapolis, Minnesota. Association for Computational Linguistics.
  26. Cross-lingual consistency of factual knowledge in multilingual language models. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 10650–10666, Singapore. Association for Computational Linguistics.
  27. Multitask prompted training enables zero-shot task generalization. In International Conference on Learning Representations.
  28. Yves Scherrer. 2020. TaPaCo: A corpus of sentential paraphrases for 73 languages. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 6868–6873, Marseille, France. European Language Resources Association.
  29. Quantifying language models’ sensitivity to spurious features in prompt design or: How i learned to start worrying about prompt formatting. In The Twelfth International Conference on Learning Representations.
  30. An empirical study of instruction-tuning large language models in Chinese. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 4086–4107, Singapore. Association for Computational Linguistics.
  31. Stanford alpaca: An instruction-following llama model. https://github.com/tatsu-lab/stanford_alpaca.
  32. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971.
  33. How far can camels go? exploring the state of instruction tuning on open resources. In Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track.
  34. How far can camels go? exploring the state of instruction tuning on open resources.
  35. Self-instruct: Aligning language model with self generated instructions. arXiv preprint arXiv:2212.10560.
  36. Super-naturalinstructions:generalization via declarative instructions on 1600+ tasks. In EMNLP.
  37. Finetuned language models are zero-shot learners. In International Conference on Learning Representations.
  38. Baize: An open-source chat model with parameter-efficient tuning on self-chat data. arXiv preprint arXiv:2304.01196.
  39. Representation engineering: A top-down approach to ai transparency. arXiv preprint arXiv:2310.01405.
Citations (1)
View on Semantic Scholar

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now
X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets