Elamur Research⌕
☰
home / papers / 2024 / Self-play with Execution Feedback: Improving Instruction-following Capabilities of Large Language Models

Self-play with Execution Feedback: Improving Instruction-following Capabilities of Large Language Models

Paper · 2024 · Guanting Dong, Keming Lu, Chengpeng Li et al.

Summary

Teaching LLMs to obey complex instructions usually depends on costly human annotation or on distilling outputs from stronger proprietary models, and neither approach scales well or guarantees correctness. The authors propose AutoIF, which turns the question of whether a response follows an instruction into a code-checking problem. Starting from a small handwritten set of seed instructions, each with a single atomic constraint, an LLM expands the pool via self-instruct. It then writes Python verification functions and unit tests for every instruction. Candidates are kept only if the code executes, passes its tests, and survives a back-translation consistency check scored by an NLI model. At the query stage, the verified instructions are paired with ShareGPT queries. Sampled responses are filtered by executing the verifiers: passing outputs become SFT data, and pass/fail pairs serve as preference data for DPO. Applied to Qwen2 and LLaMA3 under self-alignment and strong-to-weak distillation, AutoIF improves instruction following without manual labels or closed-model judges.

Key points

  • The pipeline has two stages. The first augments and verifies instructions, keeping functions and test cases that run and reach accuracy above 0.5, followed by back-translation and NLI filtering. The second augments queries from ShareGPT and applies execution-feedback rejection sampling.
  • AutoIF was tested with three training algorithms (SFT, Offline DPO, Online DPO) on Qwen2-72B, LLaMA3-70B, Qwen2-7B and LLaMA3-8B.
  • On IFEval, Loose Instruction accuracy reached 88.0% for Qwen2-72B and 90.4% for LLaMA3-70B. The authors report this as the first result above 90%.
  • On FollowBench, the large models gained more than 5% in average SSR. Qwen2-7B and LLaMA3-8B gained more than 4% on average across both benchmarks, and using GPT-4 as the supervisor model improved results further.
  • The code is available at github.com/QwenLM/AutoIF. The authors plan to release SFT and DPO datasets built with Qwen2-72B, containing tens of thousands of examples. Baseline models were trained on 20K ShareGPT samples.

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

 Institutions

🏢Alibaba GroupHangzhou · China207 works · 17,449 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 19,503 worksArtificial Intelligence · 18,805 worksArts and Humanities · 17,959 worksLanguage and Linguistics · 17,789 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 4,145 works

Cite

Cite as Guanting Dong, Keming Lu, Chengpeng Li, Tingyu Xia, Bowen Yu, Chang Zhou and Jingren Zhou. 2024. Self-play with Execution Feedback: Improving Instruction-following Capabilities of Large Language Models. arXiv:2406.13542.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2406.13542 · arXiv 2406.13542

@misc{dong2024self,
  title = {{Self-play with Execution Feedback: Improving Instruction-following Capabilities of Large Language Models}},
  author = {Guanting Dong, Keming Lu, Chengpeng Li, Tingyu Xia, Bowen Yu, Chang Zhou and Jingren Zhou},
  year = {2024},
  doi = {10.48550/arxiv.2406.13542},
  eprint = {2406.13542},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2406.13542}
}

References

54 references, 27 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, et al. Gpt-4 technical report. arXiv preprint arXiv:2303.08774, 2023.
  2. Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, Binyuan Hui, Luo Ji, Mei Li, Junyang Lin, Runji Lin, Dayiheng Liu, Gao Liu, Chengqiang Lu, Keming Lu, Jianxin Ma, Rui Men, Xingzhang Ren, Xuancheng Ren, Chuanqi Tan, Sinan Tan, Jianhong Tu, P… — Qwen Technical Report
  3. Boxi Cao, Keming Lu, Xinyu Lu, Jiawei Chen, Mengjie Ren, Hao Xiang, Peilin Liu, Yaojie Lu, Ben He, Xianpei Han, Le Sun, Hongyu Lin, and Bowen Yu. Towards scalable automated alignment of llms: A survey, 2024a. find
  4. Boxi Cao, Keming Lu, Xinyu Lu, Jiawei Chen, Mengjie Ren, Hao Xiang, Peilin Liu, Yaojie Lu, Ben He, Xianpei Han, et al. Towards scalable automated alignment of llms: A survey. arXiv preprint arXiv:2406.01252, 2024b. — [arXiv:2406.01252](https://arxiv.org/abs/2406.01252) find
  5. Yupeng Chang, Xu Wang, Jindong Wang, Yuan Wu, Linyi Yang, Kaijie Zhu, Hao Chen, Xiaoyuan Yi, Cunxiang Wang, Yidong Wang, et al. A survey on evaluation of large language models. ACM Transactions on Intelligent Systems and Technology, 15(3):1–45, 2024.
  6. Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, Alex Ray, Raul Puri, Gretchen Krueger, Michael Petrov, Heidy Khlaaf, Girish Sastry, Pamela Mishkin, Brooke Chan, Scott Gray, Nick Ryder, Mikhai… — Evaluating Large Language Models Trained on Code
  7. Mark Chen, Jerry Tworek, Heewoo Jun, Qiming Yuan, Henrique Ponde de Oliveira Pinto, Jared Kaplan, Harri Edwards, Yuri Burda, Nicholas Joseph, Greg Brockman, Alex Ray, Raul Puri, Gretchen Krueger, Michael Petrov, Heidy Khlaaf, Girish Sastry, Pamela Mishkin, Brooke Chan, Scott Gray, Nick Ryder, Mikhai… — Evaluating Large Language Models Trained on Code
  8. Xinyun Chen, Maxwell Lin, Nathanael Schärli, and Denny Zhou. Teaching large language models to self-debug, 2023. find
  9. Zixiang Chen, Yihe Deng, Huizhuo Yuan, Kaixuan Ji, and Quanquan Gu. Self-play fine-tuning converts weak language models to strong language models, 2024. find
  10. 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… ↗
  11. Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, Mark Chen, Heewoo Jun, Lukasz Kaiser, Matthias Plappert, Jerry Tworek, Jacob Hilton, Reiichiro Nakano, Christopher Hesse, and John Schulman. Training verifiers to solve math word problems, 2021.
  12. Ganqu Cui, Lifan Yuan, Ning Ding, Guanming Yao, Wei Zhu, Yuan Ni, Guotong Xie, Zhiyuan Liu, and Maosong Sun. Ultrafeedback: Boosting language models with high-quality feedback. arXiv preprint arXiv:2310.01377, 2023. find
  13. Tri Dao. Flashattention-2: Faster attention with better parallelism and work partitioning. CoRR, 2023.
  14. Guanting Dong, Hongyi Yuan, Keming Lu, Chengpeng Li, Mingfeng Xue, Dayiheng Liu, Wei Wang, Zheng Yuan, Chang Zhou, and Jingren Zhou. How abilities in large language models are affected by supervised fine-tuning data composition. arXiv preprint arXiv:2310.05492, 2023.
  15. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding, 2021.
  16. Yuzhen Huang, Yuzhuo Bai, Zhihao Zhu, Junlei Zhang, Jinghan Zhang, Tangjun Su, Junteng Liu, Chuancheng Lv, Yikai Zhang, Jiayi Lei, Yao Fu, Maosong Sun, and Junxian He. C-eval: A multi-level multi-discipline chinese evaluation suite for foundation models, 2023.
  17. Albert Q Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, et al. Mistral 7b. arXiv preprint arXiv:2310.06825, 2023. find
  18. Albert Q Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, et al. Mixtral of experts. arXiv preprint arXiv:2401.04088, 2024a. find
  19. Yuxin Jiang, Yufei Wang, Xingshan Zeng, Wanjun Zhong, Liangyou Li, Fei Mi, Lifeng Shang, Xin Jiang, Qun Liu, and Wei Wang. Followbench: A multi-level fine-grained constraints following benchmark for large language models, 2024b.
  20. Hung Le, Yue Wang, Akhilesh Deepak Gotmare, Silvio Savarese, and Steven C. H. Hoi. Coderl: Mastering code generation through pretrained models and deep reinforcement learning, 2022.
  21. Renze Lou, Kai Zhang, and Wenpeng Yin. A comprehensive survey on instruction following. arXiv preprint arXiv:2303.10475, 2023. — [arXiv:2303.10475](https://arxiv.org/abs/2303.10475) find
  22. Meta. Introducing meta llama 3: The most capable openly available llm to date, 2024. URL https://ai.meta.com/blog/meta-llama-3/. ↗
  23. Niklas Muennighoff, Alexander Rush, Boaz Barak, Teven Le Scao, Nouamane Tazi, Aleksandra Piktus, Sampo Pyysalo, Thomas Wolf, and Colin A Raffel. Scaling data-constrained language models. Advances in Neural Information Processing Systems, 36, 2024. find
  24. Alhassan Mumuni and Fuseini Mumuni. Data augmentation: A comprehensive survey of modern approaches. Array, 16:100258, 2022. find
  25. OpenAI. Introducing chatgpt, 2022. URL https://openai.com/index/chatgpt/. ↗
  26. OpenAI, Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, Red Avila, Igor Babuschkin, Suchir Balaji, Valerie Balcom, Paul Baltescu, Haiming Bao, Mohammad Bavarian, Jeff Belgum, Irwan Bello, Ja… — GPT-4 Technical Report
  27. Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll L. Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, John Schulman, Jacob Hilton, Fraser Kelton, Luke Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Christiano, Jan Leike, and Ryan Lowe. Training la… — Training language models to follow instructions with human feedback
  28. Baolin Peng, Chunyuan Li, Pengcheng He, Michel Galley, and Jianfeng Gao. Instruction tuning with gpt-4. arXiv preprint arXiv:2304.03277, 2023.
  29. Shuofei Qiao, Honghao Gui, Chengfei Lv, Qianghuai Jia, Huajun Chen, and Ningyu Zhang. Making language models better tool learners with execution feedback, 2024. find
  30. Yiwei Qin, Kaiqiang Song, Yebowen Hu, Wenlin Yao, Sangwoo Cho, Xiaoyang Wang, Xuansheng Wu, Fei Liu, Pengfei Liu, and Dong Yu. Infobench: Evaluating instruction following ability in large language models, 2024.
  31. Rafael Rafailov, Archit Sharma, Eric Mitchell, Stefano Ermon, Christopher D. Manning, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model, 2023.
  32. Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. Direct preference optimization: Your language model is secretly a reward model. Advances in Neural Information Processing Systems, 36, 2024.
  33. Jeff Rasley, Samyam Rajbhandari, Olatunji Ruwase, and Yuxiong He. Deepspeed: System optimizations enable training deep learning models with over 100 billion parameters. KDD ’20, 2020. find
  34. Haoran Sun, Lixin Liu, Junjie Li, Fengyu Wang, Baohua Dong, Ran Lin, and Ruohui Huang. Conifer: Improving complex constrained instruction-following ability of large language models, 2024.
  35. Rohan Taori, Ishaan Gulrajani, Tianyi Zhang, Yann Dubois, Xuechen Li, Carlos Guestrin, Percy Liang, and Tatsunori B Hashimoto. Stanford alpaca: An instruction-following llama model, 2023. find
  36. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, Dan Bikel, Lukas Blecher, Cristian Canton Ferrer, Moya Chen, Guillem Cucurull, David Esiobu, Jude Fernandes, Jeremy Fu, Wenyin Fu, Brian Fuller,… — Llama 2: Open Foundation and Fine-Tuned Chat Models
  37. Xinru Wang, Hannah Kim, Sajjadur Rahman, Kushan Mitra, and Zhengjie Miao. Human-llm collaborative annotation through effective verification of llm labels. In Proceedings of the CHI Conference on Human Factors in Computing Systems, pp. 1–21, 2024a. find
  38. Yizhong Wang, Yeganeh Kordi, Swaroop Mishra, Alisa Liu, Noah A. Smith, Daniel Khashabi, and Hannaneh Hajishirzi. Self-instruct: Aligning language models with self-generated instructions, 2023. find
  39. Zifeng Wang, Chun-Liang Li, Vincent Perot, Long T. Le, Jin Miao, Zizhao Zhang, Chen-Yu Lee, and Tomas Pfister. Codeclm: Aligning language models with tailored synthetic data, 2024b. find
  40. Jason Wei, Maarten Bosma, Vincent Y Zhao, Kelvin Guu, Adams Wei Yu, Brian Lester, Nan Du, Andrew M Dai, and Quoc V Le. Finetuned language models are zero-shot learners. arXiv preprint arXiv:2109.01652, 2021.
  41. Jerry Wei, Chengrun Yang, Xinying Song, Yifeng Lu, Nathan Hu, Dustin Tran, Daiyi Peng, Ruibo Liu, Da Huang, Cosmo Du, et al. Long-form factuality in large language models. arXiv preprint arXiv:2403.18802, 2024.
  42. Congying Xia, Chen Xing, Jiangshu Du, Xinyi Yang, Yihao Feng, Ran Xu, Wenpeng Yin, and Caiming Xiong. Fofo: A benchmark to evaluate llms’ format-following capability, 2024. find
  43. Qizhe Xie, Zihang Dai, Eduard Hovy, Thang Luong, and Quoc Le. Unsupervised data augmentation for consistency training. Advances in neural information processing systems, 33:6256–6268, 2020.
  44. Can Xu, Qingfeng Sun, Kai Zheng, Xiubo Geng, Pu Zhao, Jiazhan Feng, Chongyang Tao, and Daxin Jiang. Wizardlm: Empowering large language models to follow complex instructions, 2023. find
  45. Jianhao Yan, Yun Luo, and Yue Zhang. Refutebench: Evaluating refuting instruction-following for large language models, 2024. find
  46. Shuo Yang, Wei-Lin Chiang, Lianmin Zheng, Joseph E. Gonzalez, and Ion Stoica. Rethinking benchmark and contamination for language models with rephrased samples, 2023.
  47. Weizhe Yuan, Richard Yuanzhe Pang, Kyunghyun Cho, Xian Li, Sainbayar Sukhbaatar, Jing Xu, and Jason Weston. Self-rewarding language models, 2024. find
  48. Zheng Yuan, Hongyi Yuan, Chengpeng Li, Guanting Dong, Chuanqi Tan, and Chang Zhou. Scaling relationship on learning mathematical reasoning with large language models. arXiv preprint arXiv:2308.01825, 2023.
  49. Shengyu Zhang, Linfeng Dong, Xiaoya Li, Sen Zhang, Xiaofei Sun, Shuhe Wang, Jiwei Li, Runyi Hu, Tianwei Zhang, Fei Wu, et al. Instruction tuning for large language models: A survey. arXiv preprint arXiv:2308.10792, 2023.
  50. Yingxiu Zhao, Bowen Yu, Binyuan Hui, Haiyang Yu, Fei Huang, Yongbin Li, and Nevin L. Zhang. A preliminary study of the intrinsic relationship between complexity and alignment. 2023. find
  51. Yingxiu Zhao, Bowen Yu, Binyuan Hui, Haiyang Yu, Minghao Li, Fei Huang, Nevin L Zhang, and Yongbin Li. Tree-instruct: A preliminary study of the intrinsic relationship between complexity and alignment. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language R… find
  52. Chenyu Zheng, Guoqiang Wu, and Chongxuan Li. Toward understanding generative data augmentation. Advances in Neural Information Processing Systems, 36, 2024. find
  53. Chunting Zhou, Pengfei Liu, Puxin Xu, Srinivasan Iyer, Jiao Sun, Yuning Mao, Xuezhe Ma, Avia Efrat, Ping Yu, Lili Yu, et al. Lima: Less is more for alignment. Advances in Neural Information Processing Systems, 36, 2024. find
  54. Jeffrey Zhou, Tianjian Lu, Swaroop Mishra, Siddhartha Brahma, Sujoy Basu, Yi Luan, Denny Zhou, and Le Hou. Instruction-following evaluation for large language models, 2023.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/dong-2024-self-play-execution-feedback-improving/