Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
38 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
41 tokens/sec
o3 Pro
7 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Evaluating the Zero-shot Robustness of Instruction-tuned Language Models (2306.11270v2)

Published 20 Jun 2023 in cs.CL and cs.LG

Abstract: Instruction fine-tuning has recently emerged as a promising approach for improving the zero-shot capabilities of LLMs on new tasks. This technique has shown particular strength in improving the performance of modestly sized LLMs, sometimes inducing performance competitive with much larger model variants. In this paper we ask two questions: (1) How sensitive are instruction-tuned models to the particular phrasings of instructions, and, (2) How can we make them more robust to such natural language variation? To answer the former, we collect a set of 319 instructions manually written by NLP practitioners for over 80 unique tasks included in widely used benchmarks, and we evaluate the variance and average performance of these instructions as compared to instruction phrasings observed during instruction fine-tuning. We find that using novel (unobserved) but appropriate instruction phrasings consistently degrades model performance, sometimes substantially so. Further, such natural instructions yield a wide variance in downstream performance, despite their semantic equivalence. Put another way, instruction-tuned models are not especially robust to instruction re-phrasings. We propose a simple method to mitigate this issue by introducing ``soft prompt'' embedding parameters and optimizing these to maximize the similarity between representations of semantically equivalent instructions. We show that this method consistently improves the robustness of instruction-tuned models.

PDF HTML Abstract
Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize File Document Download Save Streamline Icon: https://streamlinehq.com PDF File Document Download Save Streamline Icon: https://streamlinehq.com Markdown Bookmark Chat Bubble Oval Streamline Icon: https://streamlinehq.com Chat (Pro)
Definition Search Book Streamline Icon: https://streamlinehq.com
References (40)
  1. What learning algorithm is in-context learning? investigations with linear models. arXiv preprint arXiv:2211.15661, 2022.
  2. Training a helpful and harmless assistant with reinforcement learning from human feedback. arXiv preprint arXiv:2204.05862, 2022.
  3. Language models are few-shot learners. Advances in neural information processing systems, 33:1877–1901, 2020.
  4. Palm: Scaling language modeling with pathways. arXiv preprint arXiv:2204.02311, 2022.
  5. Scaling instruction-finetuned language models. arXiv preprint arXiv:2210.11416, 2022.
  6. Why can gpt learn in-context? language models secretly perform gradient descent as meta optimizers. arXiv preprint arXiv:2212.10559, 2022.
  7. Improving alignment of dialogue agents via targeted human judgements. arXiv preprint arXiv:2209.14375, 2022.
  8. Robustness of learning from task instructions. 2023.
  9. Measuring massive multitask language understanding. arXiv preprint arXiv:2009.03300, 2020.
  10. Training compute-optimal large language models. arXiv preprint arXiv:2203.15556, 2022.
  11. OPT-IML: Scaling language model instruction meta learning through the lens of generalization. arXiv preprint arXiv:2212.12017, 2022.
  12. Can large language models truly understand prompts? a case study with negated prompts. In Transfer Learning for Natural Language Processing Workshop, pages 52–62. PMLR, 2023.
  13. Unifying question answering, text classification, and regression via span extraction. arXiv preprint arXiv:1904.09286, 2019.
  14. Unifiedqa: Crossing format boundaries with a single qa system. arXiv preprint arXiv:2005.00700, 2020.
  15. Quantifying the carbon emissions of machine learning. arXiv preprint arXiv:1910.09700, 2019.
  16. Prefix-tuning: Optimizing continuous prompts for generation. arXiv preprint arXiv:2101.00190, 2021.
  17. The flan collection: Designing data and methods for effective instruction tuning. arXiv preprint arXiv:2301.13688, 2023.
  18. The natural language decathlon: Multitask learning as question answering. arXiv preprint arXiv:1806.08730, 2018.
  19. Rethinking the role of demonstrations: What makes in-context learning work? arXiv preprint arXiv:2202.12837, 2022.
  20. Cross-task generalization via natural language crowdsourcing instructions. arXiv preprint arXiv:2104.08773, 2021.
  21. Webgpt: Browser-assisted question-answering with human feedback. arXiv preprint arXiv:2112.09332, 2021.
  22. Training language models to follow instructions with human feedback. arXiv preprint arXiv:2203.02155, 2022.
  23. Exploring the limits of transfer learning with a unified text-to-text transformer. The Journal of Machine Learning Research, 21(1):5485–5551, 2020.
  24. Multitask prompted training enables zero-shot task generalization. arXiv preprint arXiv:2110.08207, 2021.
  25. Beyond the imitation game: Quantifying and extrapolating the capabilities of language models. arXiv preprint arXiv:2206.04615, 2022.
  26. Alpaca: A strong, replicable instruction-following model, 2023.
  27. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023.
  28. Laurens Van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of machine learning research, 9(11), 2008.
  29. Large language models are implicitly topic models: Explaining and finding good demonstrations for in-context learning. arXiv preprint arXiv:2301.11916, 2023.
  30. Self-instruct: Aligning language model with self generated instructions. arXiv preprint arXiv:2212.10560, 2022.
  31. Benchmarking generalization via in-context instructions on 1,600+ language tasks. arXiv preprint arXiv:2204.07705, 2022.
  32. Do prompt-based models really understand the meaning of their prompts? In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 2300–2344, Seattle, United States, July 2022. Association for Computational Linguistics.
  33. Finetuned language models are zero-shot learners. arXiv preprint arXiv:2109.01652, 2021.
  34. Multiinstruct: Improving multi-modal zero-shot learning via instruction tuning. arXiv preprint arXiv:2212.10773, 2022.
  35. Guess the instruction! making language models stronger zero-shot learners. arXiv preprint arXiv:2210.02969, 2022.
  36. Alert: Adapting language models to reasoning tasks. arXiv preprint arXiv:2212.08286, 2022.
  37. Glm-130b: An open bilingual pre-trained model. arXiv preprint arXiv:2210.02414, 2022.
  38. Aligning instruction tasks unlocks large language models as zero-shot relation extractors. arXiv preprint arXiv:2305.11159, 2023.
  39. The wisdom of hindsight makes language models better instruction followers. arXiv preprint arXiv:2302.05206, 2023.
  40. Large language models are human-level prompt engineers. arXiv preprint arXiv:2211.01910, 2022.
User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (3)
  1. Jiuding Sun (11 papers)
  2. Chantal Shaib (7 papers)
  3. Byron C. Wallace (82 papers)
Citations (36)
View on Semantic Scholar