Elamur Research
☰
home / papers / 2023 / Orca 2: Teaching Small Language Models How to Reason

Orca 2: Teaching Small Language Models How to Reason

Written for the Research library (claude-opus-5-5) from the paper text. Original — arxiv.org/pdf/2311.11045

Summary

Small language models (below roughly 10B parameters) typically trail frontier LLMs in zero-shot reasoning. The usual remedy, training them to imitate a stronger teacher's outputs, tends to transfer the teacher's style more than its reasoning ability. Orca 2, from Microsoft Research, argues that a student model should not have to solve problems the same way a large teacher does. The authors use capable LLMs such as GPT-4 to generate synthetic training data showing several solution strategies, including step-by-step, recall-then-generate, recall-reason-generate, extract-generate and direct answer. For each task they select the behaviour that a small model can realistically reproduce. The detailed instructions that triggered the teacher's behaviour are stripped from the training examples, a technique called Prompt Erasure, so the student must learn to choose a suitable strategy on its own. Fine-tuned from LLaMA-2 7B and 13B, Orca 2 clearly outperforms models of the same size. It also matches or beats models 5–10 times larger on zero-shot reasoning tasks. This suggests that well-designed synthetic data can make cheaper, smaller models far more useful.

Key points

  • Two models, Orca-2-7B and Orca-2-13B, are built on LLaMA-2 base models. Their weights are released for research at aka.ms/orca-lm.
  • The core ideas are teaching multiple reasoning strategies tailored per task and using Prompt Erasure, which together produce what the authors call a "Cautious Reasoner."
  • Evaluation spans 15 benchmarks, covering about 100 tasks and more than 36K unique prompts. These include AGIEval, BBH, MMLU, ARC-E, ARC-C, RACE and GSM8K, plus tests of summarization, groundedness, truthfulness and toxicity.
  • Orca 2 is compared against LLaMA-2-Chat (13B, 70B) and WizardLM (13B, 70B), all sharing LLaMA-2 bases. In zero-shot settings, Orca 2 matches or surpasses even the 70B variants.
  • The paper reports several limitations:
    • The models received no RLHF safety training.
    • Few-shot prompting brings weaker gains than it does for other models.
    • Performance varies with the system message.
    • Knowledge is bounded by pre-training, so the authors recommend use as a reasoning engine over in-context information, and in research settings only.

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: Mathematics & statistics

Cited by

130 papers of our library cite it (of the 140673 whose bibliographies we read); 5 of them have a page here, 125 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2311.11045 · arXiv 2311.11045

@misc{mitra2023orca,
  title = {{Orca 2: Teaching Small Language Models How to Reason}},
  author = {Mitra, Arindam and Corro, Luciano Del and Mahajan, Shweti and Codas, Andres and Simoes, Clarisse and Agarwal, Sahaj and Chen, Xuxi and Razdaibiedina, Anastasia and Jones, Erik and Aggarwal, Kriti and Palangi, Hamid and Zheng, Guoqing and Rosset, Corby and Khanpour, Hamed and Awadallah, Ahmed},
  year = {2023},
  doi = {10.48550/arxiv.2311.11045},
  eprint = {2311.11045},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2311.11045}
}

References

69 references, 28 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Rohan Anil, Andrew M. Dai, Orhan Firat, Melvin Johnson, Dmitry Lepikhin, Alexandre Passos, Siamak Shakeri, Emanuel Taropa, Paige Bailey, Zhifeng Chen, Eric Chu, Jonathan H. Clark, Laurent El Shafey, Yanping Huang, Kathy Meier-Hellstern, Gaurav Mishra, Erica Moreira, Mark Omernick, Kevin Robinson, Se… find our card
  2. Payal Bajaj, Daniel Campos, Nick Craswell, Li Deng, Jianfeng Gao, Xiaodong Liu, Rangan Majumder, Andrew McNamara, Bhaskar Mitra, Tri Nguyen, Mir Rosenberg, Xia Song, Alina Stoica, Saurabh Tiwary, and Tong Wang. Ms marco: A human generated machine reading comprehension dataset, 2018. find our card
  3. Christian Bird, Denae Ford, Thomas Zimmermann, Nicole Forsgren, Eirini Kalliamvakou, Travis Lowdermilk, and Idan Gazit. Taking flight with copilot: Early insights and opportunities of ai-powered pair-programming tools. Queue, 20(6):35–57, jan 2023. ISSN 1542-7730. doi: 10.1145/3582083. URL https://d… — [doi:10.1145/3582083](https://doi.org/10.1145/3582083) find
  4. Tommaso Caselli, Valerio Basile, Jelena Mitrovic, and M. Granitzer. Hatebert: Retraining bert for abusive language detection in english. ArXiv, abs/2010.12472, 2021. find
  5. Yew Ken Chia, Pengfei Hong, Lidong Bing, and Soujanya Poria. Instructeval: Towards holistic evaluation of instruction-tuned large language models. arXiv preprint arXiv:2306.04757, 2023. 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://vicuna.lmsys.org. ↗
  7. Hyung Won Chung, Le Hou, Shayne Longpre, Barret Zoph, Yi Tay, William Fedus, Yunxuan Li, Xuezhi Wang, Mostafa Dehghani, Siddhartha Brahma, Albert Webson, Shixiang Shane Gu, Zhuyun Dai, Mirac Suzgun, Xinyun Chen, Aakanksha Chowdhery, Alex Castro-Ros, Marie Pellat, Kevin Robinson, Dasha Valter, Sharan… find
  8. Peter Clark, Isaac Cowhey, Oren Etzioni, Tushar Khot, Ashish Sabharwal, Carissa Schoenick, and Oyvind Tafjord. Think you have solved question answering? try arc, the ai2 reasoning challenge. ArXiv, abs/1803.05457, 2018. URL https://api.semanticscholar.org/CorpusID: 3922816. ↗
  9. 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
  10. Dheeru Dua, Yizhong Wang, Pradeep Dasigi, Gabriel Stanovsky, Sameer Singh, and Matt Gardner. DROP: A reading comprehension benchmark requiring discrete reasoning over paragraphs. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Hum… find our card
  11. Jörg Frohberg and Frank Binder. Crass: A novel data set and benchmark to test counterfactual reasoning of large language models, 2022. find
  12. Xinyang Geng, Arnav Gudibande, Hao Liu, Eric Wallace, Pieter Abbeel, Sergey Levine, and Dawn Song. Koala: A dialogue model for academic research. Blog post, April 2023. URL https://bair.berkeley.edu/blog/2023/04/03/koala/. ↗
  13. Arnav Gudibande, Eric Wallace, Charlie Snell, Xinyang Geng, Hao Liu, Pieter Abbeel, Sergey Levine, and Dawn Song. The false promise of imitating proprietary llms, 2023. find our card
  14. Himanshu Gupta, Neeraj Varshney, Swaroop Mishra, Kuntal Kumar Pal, Saurabh Arjun Sawant, Kevin Scaria, Siddharth Goyal, and Chitta Baral. " john is 50 years old, can his son be 65?" evaluating nlp models’ understanding of feasibility. arXiv preprint arXiv:2210.07471, 2022. find
  15. Veronika Hackl, Alexandra Elena Müller, Michael Granitzer, and Maximilian Sailer. Is gpt-4 a reliable rater? evaluating consistency in gpt-4 text ratings, 2023. find
  16. Thomas Hartvigsen, Saadia Gabriel, Hamid Palangi, Maarten Sap, Dipankar Ray, and Ece Kamar. ToxiGen: A large-scale machine-generated dataset for adversarial and implicit hate speech detection. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long… find our card
  17. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. Measuring massive multitask language understanding. Proceedings of the International Conference on Learning Representations (ICLR), 2021. find our card
  18. Dan Hendrycks, Collin Burns, Saurav Kadavath, Akul Arora, Steven Basart, Eric Tang, Dawn Song, and Jacob Steinhardt. Measuring mathematical problem solving with the math dataset. arXiv preprint arXiv:2103.03874, 2021. find our card
  19. Mohammad Javad Hosseini, Hannaneh Hajishirzi, Oren Etzioni, and Nate Kushman. Learning to solve arithmetic word problems with verb categorization. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 523–533, 2014. find our card
  20. Frederick Jelinek, Robert L. Mercer, Lalit R. Bahl, and Janet M. Baker. Perplexity—a measure of the difficulty of speech recognition tasks. Journal of the Acoustical Society of America, 62, 1977. URL https://api.semanticscholar.org/CorpusID:121680873. ↗
  21. Erik Jones, Hamid Palangi, Clarisse Simões, Varun Chandrasekaran, Subhabrata Mukherjee, Arindam Mitra, Ahmed Awadallah, and Ece Kamar. Teaching language models to hallucinate less with synthetic tasks, 2023. find
  22. Daniel Kahneman. Thinking, fast and slow. Farrar, Straus and Giroux, New York, 2011. ISBN 9780374275631 0374275637. URL https://www.amazon.de/ Thinking-Fast-Slow-Daniel-Kahneman/dp/0374275637/ref=wl\_it\_dp\_o\_pdT1\_nS\_nC? ie=UTF8&colid=151193SNGKJT9&coliid=I3OCESLZCVDFL7. ↗
  23. Takeshi Kojima, Shixiang Shane Gu, Machel Reid, Yutaka Matsuo, and Yusuke Iwasawa. Large language models are zero-shot reasoners, 2023. find our card
  24. Rik Koncel-Kedziorski, Hannaneh Hajishirzi, Ashish Sabharwal, Oren Etzioni, and Siena Dumas Ang. Parsing algebraic word problems into equations. Transactions of the Association for Computational Linguistics, 3:585–597, 2015. find
  25. Mario Michael Krell, Matej Kosec, Sergio P. Perez, and Andrew Fitzgibbon. Efficient sequence packing without cross-contamination: Accelerating large language models without impacting performance, 2022. find
  26. Nate Kushman, Yoav Artzi, Luke Zettlemoyer, and Regina Barzilay. Learning to automatically solve algebra word problems. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 271–281, 2014. find
  27. Guokun Lai, Qizhe Xie, Hanxiao Liu, Yiming Yang, and Eduard Hovy. RACE: Large-scale ReAding comprehension dataset from examinations. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 785–794, Copenhagen, Denmark, September 2017. Association for Computat… find our card
  28. Md Tahmid Rahman Laskar, M Saiful Bari, Mizanur Rahman, Md Amran Hossen Bhuiyan, Shafiq Joty, and Jimmy Huang. A systematic study and comprehensive evaluation of ChatGPT on benchmark datasets. In Anna Rogers, Jordan Boyd-Graber, and Naoaki Okazaki, editors, Findings of the Association for Computatio… find
  29. Chin-Yew Lin. ROUGE: A package for automatic evaluation of summaries. In Text Summarization Branches Out, pages 74–81, Barcelona, Spain, July 2004. Association for Computational Linguistics. URL https://aclanthology.org/W04-1013. ↗
  30. 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), pages 3214–3252, Dublin, Ireland, May 2022. Association for Computational Linguis… find our card
  31. Wang Ling, Dani Yogatama, Chris Dyer, and Phil Blunsom. Program induction by rationale generation: Learning to solve and explain algebraic word problems. ACL, 2017. find
  32. Yang Liu, Dan Iter, Yichong Xu, Shuohang Wang, Ruochen Xu, and Chenguang Zhu. G-eval: Nlg evaluation using gpt-4 with better human alignment, 2023. find
  33. Shayne Longpre, Le Hou, Tu Vu, Albert Webson, Hyung Won Chung, Yi Tay, Denny Zhou, Quoc V. Le, Barret Zoph, Jason Wei, and Adam Roberts. The flan collection: Designing data and methods for effective instruction tuning, 2023. find our card
  34. Ahmed Magooda, Alec Helyar, Kyle Jackson, David Sullivan, Chad Atalla, Emily Sheng, Dan Vann, Richard Edgar, Hamid Palangi, Roman Lutz, Hongliang Kong, Vincent Yun, Eslam Kamal, Federico Zarfati, Hanna Wallach, Sarah Bird, and Mei Chen. A framework for automated measurement of responsible ai harms i… find
  35. Dakota Mahan, Ryan Carlow, Louis Castricato, Nathan Cooper, and Christian Laforte. Stable beluga models. URL [https://huggingface.co/stabilityai/StableBeluga2](https: //huggingface.co/stabilityai/StableBeluga2). find
  36. Y. Mehdi. Reinventing search with a new ai-powered microsoft bing and edge, your copilot for the web. https://blogs.microsoft.com/blog/2023/02/07/ reinventing-search-with-a-new-ai-powered-microsoft-bing-and-edge-your-copilot-for-the-web/, 2023. (Accessed: 15 November 2023). ↗
  37. Alham Fikri Aji Minghao Wu. Style over substance: Evaluation biases for large language models, 2023. find
  38. Swaroop Mishra, Daniel Khashabi, Chitta Baral, and Hannaneh Hajishirzi. Cross-task generalization via natural language crowdsourcing instructions. arXiv preprint arXiv:2104.08773, 2021. find
  39. Swaroop Mishra, Matthew Finlayson, Pan Lu, Leonard Tang, Sean Welleck, Chitta Baral, Tanmay Rajpurohit, Oyvind Tafjord, Ashish Sabharwal, Peter Clark, et al. Lila: A unified benchmark for mathematical reasoning. arXiv preprint arXiv:2210.17517, 2022. find our card
  40. Swaroop Mishra, Arindam Mitra, Neeraj Varshney, Bhavdeep Sachdeva, Peter Clark, Chitta Baral, and Ashwin Kalyan. Numglue: A suite of fundamental yet challenging mathematical reasoning tasks. arXiv preprint arXiv:2204.05660, 2022. find
  41. Nasrin Mostafazadeh, Nathanael Chambers, Xiaodong He, Devi Parikh, Dhruv Batra, Lucy Vanderwende, Pushmeet Kohli, and James Allen. A corpus and cloze evaluation for deeper understanding of commonsense stories. In Proceedings of the 2016 Conference of the North American Chapter of the Association for… find
  42. Subhabrata Mukherjee, Arindam Mitra, Ganesh Jawahar, Sahaj Agarwal, Hamid Palangi, and Ahmed Awadallah. Orca: Progressive learning from complex explanation traces of gpt-4. arXiv preprint arXiv:2306.02707, 2023. find our card
  43. Ben Naismith, Phoebe Mulcaire, and Jill Burstein. Automated evaluation of written discourse coherence using gpt-4, 2023. find
  44. OpenAI. Gpt-4 technical report, 2023. find our card
  45. OpenAI. Chatgpt (sep 25 version). https://chat.openai.com/chat, 2023. [Large language model]. ↗
  46. 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… find
  47. 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 E. Miller, Maddie Simens, Amanda Askell, Peter Welinder, Paul Francis Christiano, Jan Leike, and Ryan J. Low… find
  48. Denis Paperno, Germán Kruszewski, Angeliki Lazaridou, Ngoc Quan Pham, Raffaella Bernardi, Sandro Pezzelle, Marco Baroni, Gemma Boleda, and Raquel Fernández. The LAMBADA dataset: Word prediction requiring a broad discourse context. In Proceedings of the 54th Annual Meeting of the Association for Comp… find our card
  49. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. Bleu: a method for automatic evaluation of machine translation. In Pierre Isabelle, Eugene Charniak, and Dekang Lin, editors, Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311–318, Philadelp… find our card
  50. David Saxton, Edward Grefenstette, Felix Hill, and Pushmeet Kohli. Analysing mathematical reasoning abilities of neural models. arXiv preprint arXiv:1904.01557, 2019. find our card
  51. Karan Singhal, Tao Tu, Juraj Gottweis, Rory Sayres, Ellery Wulczyn, Le Hou, Kevin Clark, Stephen Pfohl, Heather Cole-Lewis, Darlene Neal, Mike Schaekermann, Amy Wang, Mohamed Amin, Sami Lachgar, Philip Mansfield, Sushant Prakash, Bradley Green, Ewa Dominowska, Blaise Aguera y Arcas, Nenad Tomasev, Y… find our card
  52. Aarohi Srivastava, Abhinav Rastogi, Abhishek Rao, Abu Awal Md Shoeb, Abubakar Abid, Adam Fisch, Adam R Brown, Adam Santoro, Aditya Gupta, and Adria Garriga-Alonso et al. Beyond the imitation game: Quantifying and extrapolating the capabilities of language models, 2022. find our card
  53. Gabriel Stanovsky, Noah A. Smith, and Luke Zettlemoyer. Evaluating gender bias in machine translation. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 1679–1684, Florence, Italy, July 2019. Association for Computational Linguistics. doi: 10.18653/v1/… find our card
  54. Mirac Suzgun, Nathan Scales, Nathanael Schärli, Sebastian Gehrmann, Yi Tay, Hyung Won Chung, Aakanksha Chowdhery, Quoc Le, Ed Chi, Denny Zhou, and Jason Wei. Challenging BIG-bench tasks and whether chain-of-thought can solve them. In Findings of the Association for Computational Linguistics: ACL 202… find our card
  55. 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. https://github.com/tatsu-lab/stanford\_alpaca, 2023. ↗
  56. Romal Thoppilan, Daniel De Freitas, Jamie Hall, Noam Shazeer, Apoorv Kulshreshtha, Heng-Tze Cheng, Alicia Jin, Taylor Bos, Leslie Baker, Yu Du, YaGuang Li, Hongrae Lee, Huaixiu Steven Zheng, Amin Ghafouri, Marcelo Menegali, Yanping Huang, Maxim Krikun, Dmitry Lepikhin, James Qin, Dehao Chen, Yuanzho… find our card
  57. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. Llama: Open and efficient foundation language models. arXiv preprint… find
  58. 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,… find
  59. Wen wai Yim, Yujuan Fu, Asma Ben Abacha, Neal Snider, Thomas Lin, and Meliha Yetisgen. Aci-bench: a novel ambient clinical intelligence dataset for benchmarking automatic visit note generation, 2023. find
  60. Peiyi Wang, Lei Li, Liang Chen, Dawei Zhu, Binghuai Lin, Yunbo Cao, Qi Liu, Tianyu Liu, and Zhifang Sui. Large language models are not fair evaluators, 2023. find
  61. Yizhong Wang, Swaroop Mishra, Pegah Alipoormolabashi, Yeganeh Kordi, Amirreza Mirzaei, Anjana Arunkumar, Arjun Ashok, Arut Selvan Dhanasekaran, Atharva Naik, David Stap, Eshaan Pathak, Giannis Karamanolakis, Haizhi Gary Lai, Ishan Purohit, Ishani Mondal, Jacob Anderson, Kirby Kuznia, Krima Doshi, Ma… find
  62. 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, 2022. find
  63. Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed H. Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, and William Fedus. Emergent abilities of large language models, 2022. find our card
  64. 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
  65. Canwen Xu, Daya Guo, Nan Duan, and Julian McAuley. Baize: An open-source chat model with parameter-efficient tuning on self-chat data, 2023. find our card
  66. Rowan Zellers, Ari Holtzman, Yonatan Bisk, Ali Farhadi, and Yejin Choi. Hellaswag: Can a machine really finish your sentence? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, 2019. find
  67. Lianmin Zheng, Wei-Lin Chiang, Ying Sheng, Siyuan Zhuang, Zhanghao Wu, Yonghao Zhuang, Zi Lin, Zhuohan Li, Dacheng Li, Eric. P Xing, Hao Zhang, Joseph E. Gonzalez, and Ion Stoica. Judging llm-as-a-judge with mt-bench and chatbot arena, 2023. find
  68. Ming Zhong, Da Yin, Tao Yu, Ahmad Zaidi, Mutethia Mutuma, Rahul Jha, Ahmed Hassan Awadallah, Asli Celikyilmaz, Yang Liu, Xipeng Qiu, and Dragomir Radev. QMSum: A new benchmark for query-based multi-domain meeting summarization. In Proceedings of the 2021 Conference of the North American Chapter of t… find our card
  69. Wanjun Zhong, Ruixiang Cui, Yiduo Guo, Yaobo Liang, Shuai Lu, Yanlin Wang, Amin Saied, Weizhu Chen, and Nan Duan. Agieval: A human-centric benchmark for evaluating foundation models, 2023. find our card