Elamur Research
☰
home / papers / 2023 / How Far Can Camels Go? Exploring the State of Instruction Tuning on Open Resources

How Far Can Camels Go? Exploring the State of Instruction Tuning on Open Resources

Written for the Research library (claude-opus-5-5) from the paper text. Original — arxiv.org/pdf/2306.04751

Summary

Open instruction-tuned models are often promoted as rivals to ChatGPT. Those claims usually rest on narrow tests that mostly use model-based judging. This makes it hard to tell which public instruction datasets are actually useful. The authors ran a controlled comparison. They fine-tuned LLaMA, LLaMA-2, OPT and Pythia base models (6.7B–65B parameters) on 12 open instruction corpora. The corpora range from human-written (Dolly, Open Assistant) to NLP-derived (SuperNI, Flan V2) and model-distilled (Alpaca, ShareGPT, Baize). All data were converted to a single chat format, with loss computed only on assistant replies. Evaluation used benchmarks, GPT-4 judging and human ratings to measure the models' grasp of facts, their reasoning ability, their performance across languages, code generation, safety, and how well they follow free-form instructions. Each dataset tends to boost particular skills, and no single source or mixture wins everywhere. Stronger base models help consistently. The resulting suite, TÜLU, has the best average results but still falls short of ChatGPT. The study matters because it shows that preference-style evaluation can mask capability gaps. It also releases models, data and code for reproducible research.

Key points

  • Training covered 12 instruction datasets, with CoT and Flan V2 subsampled to 100K examples. Base models were LLaMA-1 (7B–65B), LLaMA-2 (7B, 13B), OPT 6.7B and Pythia 6.9B.
  • TÜLU combines 7 open datasets. A separate human-only mixture uses Flan V2, CoT, Dolly and Open Assistant. TÜLU 65B was reportedly the largest fully instruction-tuned public LLaMA variant at release.
  • On any given evaluation, the top open model reaches on average about 87% of ChatGPT's score and 73% of GPT-4's.
  • TÜLU stays within 29% of the best model on every individual task while having the best average.
  • GPT-4-based preference scores correlate strongly with how many unique tokens a model generates. Such judgments can therefore favour long, varied answers over real capability.
  • All checkpoints, code and the evaluation framework are released at github.com/allenai/open-instruct.

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

262 papers of our library cite it (of the 140673 whose bibliographies we read); 24 of them have a page here, 238 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2306.04751 · arXiv 2306.04751

@misc{wang2023how,
  title = {{How Far Can Camels Go? Exploring the State of Instruction Tuning on Open Resources}},
  author = {Wang, Yizhong and Ivison, Hamish and Dasigi, Pradeep and Hessel, Jack and Khot, Tushar and Chandu, Khyathi Raghavi and Wadden, David and MacMillan, Kelsey and Smith, Noah A. and Beltagy, Iz and Hajishirzi, Hannaneh},
  year = {2023},
  doi = {10.48550/arxiv.2306.04751},
  eprint = {2306.04751},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2306.04751}
}

References

56 references, 22 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. E. Almazrouei, H. Alobeidli, A. Alshamsi, A. Cappelli, R. Cojocaru, M. Debbah, E. Goffinet, D. Heslow, J. Launay, Q. Malartic, B. Noune, B. Pannier, and G. Penedo. Falcon-40B: an open large language model with state-of-the-art performance. Huggingface Model Release, 2023. URL https://huggingface.co/… ↗
  2. R. Anil, A. M. Dai, O. Firat, M. Johnson, D. Lepikhin, A. Passos, S. Shakeri, E. Taropa, P. Bailey, Z. Chen, et al. Palm 2 technical report. arXiv preprint arXiv:2305.10403, 2023. find our card
  3. Y. Bai, A. Jones, K. Ndousse, A. Askell, A. Chen, N. DasSarma, D. Drain, S. Fort, D. Ganguli, T. Henighan, et al. Training a helpful and harmless assistant with reinforcement learning from human feedback. arXiv preprint arXiv:2204.05862, 2022. find
  4. S. Biderman, H. Schoelkopf, Q. G. Anthony, H. Bradley, K. O’Brien, E. Hallahan, M. A. Khan, S. Purohit, U. S. Prashanth, E. Raff, et al. Pythia: A suite for analyzing large language models across training and scaling. In International Conference on Machine Learning, pages 2397–2430. PMLR, 2023. find
  5. T. Cai, X. Wang, T. Ma, X. Chen, and D. Zhou. Large language models as tool makers. arXiv preprint arXiv:2305.17126, 2023. find
  6. S. Chaudhary. Code alpaca: An instruction-following llama model for code generation. GitHub repository, 2023. URL https://github.com/sahil280114/codealpaca. ↗
  7. M. Chen, J. Tworek, H. Jun, Q. Yuan, H. P. d. O. Pinto, J. Kaplan, H. Edwards, Y. Burda, N. Joseph, G. Brockman, et al. Evaluating large language models trained on code. arXiv preprint arXiv:2107.03374, 2021. find our card
  8. W.-L. Chiang, Z. Li, Z. Lin, Y. Sheng, Z. Wu, H. Zhang, L. Zheng, S. Zhuang, Y. Zhuang, J. E. Gonzalez, I. Stoica, and E. P. Xing. Vicuna: An open-source chatbot impressing gpt-4 with 90%* chatgpt quality. Blog post, March 2023. URL https://lmsys.org/blog/ 2023-03-30-vicuna/. ↗
  9. H. W. Chung, L. Hou, S. Longpre, B. Zoph, Y. Tay, W. Fedus, E. Li, X. Wang, M. Dehghani, S. Brahma, et al. Scaling instruction-finetuned language models. arXiv preprint arXiv:2210.11416, 2022. find
  10. J. H. Clark, E. Choi, M. Collins, D. Garrette, T. Kwiatkowski, V. Nikolaev, and J. Palomaki. TyDi QA: A benchmark for information-seeking question answering in typologically diverse languages. TACL, 2020. URL https://arxiv.org/abs/2003.05002. ↗ our card
  11. K. Cobbe, V. Kosaraju, M. Bavarian, M. Chen, H. Jun, L. Kaiser, M. Plappert, J. Tworek, J. Hilton, R. Nakano, C. Hesse, and J. Schulman. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168, 2021. find our card
  12. Databricks. Free dolly: Introducing the world’s first truly open instruction-tuned llm. Blog post, 2023. URL https://www.databricks.com/blog/2023/04/12/ dolly-first-open-commercially-viable-instruction-tuned-llm. ↗
  13. T. Dettmers, M. Lewis, Y. Belkada, and L. Zettlemoyer. LLM.int8(): 8-bit matrix multiplication for transformers at scale. In Advances in Neural Information Processing Systems, 2022. find
  14. T. Dettmers, A. Pagnoni, A. Holtzman, and L. Zettlemoyer. Qlora: Efficient finetuning of quantized llms. arXiv preprint arXiv:2305.14314, 2023. find
  15. N. Ding, Y. Chen, B. Xu, S. Hu, Y. Qin, Z. Liu, M. Sun, and B. Zhou. Ultrachat: A large-scale auto-generated multi-round dialogue data. GitHub Repository, 2023. URL https: //github.com/thunlp/ultrachat. find
  16. Y. Dubois, X. Li, R. Taori, T. Zhang, I. Gulrajani, J. Ba, C. Guestrin, P. Liang, and T. B. Hashimoto. Alpacafarm: A simulation framework for methods that learn from human feedback. arXiv preprint arXiv:2305.14387, 2023. find our card
  17. L. Gao, J. Tow, S. Biderman, S. Black, A. DiPofi, C. Foster, L. Golding, J. Hsu, K. McDonell, N. Muennighoff, J. Phang, L. Reynolds, E. Tang, A. Thite, B. Wang, K. Wang, and A. Zou. A framework for few-shot language model evaluation, Sept. 2021. URL https://doi.org/10. 5281/zenodo.5371628. find
  18. X. Geng and H. Liu. Openllama: An open reproduction of llama. GitHub Repository, 2023. URL https://github.com/openlm-research/open\_llama. ↗
  19. X. Geng, A. Gudibande, H. Liu, E. Wallace, P. Abbeel, S. Levine, and D. Song. Koala: A dialogue model for academic research. Blog post, April 2023. URL https://bair.berkeley. edu/blog/2023/04/03/koala/. ↗
  20. A. Gudibande, E. Wallace, C. Snell, X. Geng, H. Liu, P. Abbeel, S. Levine, and D. Song. The false promise of imitating proprietary llms. arXiv preprint arXiv:2305.15717, 2023. find our card
  21. T. Hartvigsen, S. Gabriel, H. Palangi, M. Sap, D. Ray, and E. Kamar. TOXIGEN: Controlling Language Models to Generate Implied and Adversarial Toxicity. In ACL, 2022. URL https: //arxiv.org/abs/2203.09509. find our card
  22. D. Hendrycks, C. Burns, S. Basart, A. Zou, M. Mazeika, D. Song, and J. Steinhardt. Measuring massive multitask language understanding. In International Conference on Learning Representations (ICLR), 2020. find our card
  23. O. Honovich, T. Scialom, O. Levy, and T. Schick. Unnatural instructions: Tuning language models with (almost) no human labor. arXiv preprint arXiv:2212.09689, 2022. find our card
  24. S. Iyer, X. V. Lin, R. Pasunuru, T. Mihaylov, D. Simig, P. Yu, K. Shuster, T. Wang, Q. Liu, P. S. Koura, et al. Opt-iml: Scaling language model instruction meta learning through the lens of generalization. arXiv preprint arXiv:2212.12017, 2022. find our card
  25. A. Köksal, T. Schick, A. Korhonen, and H. Schütze. Longform: Optimizing instruction tuning for long text generation with corpus extraction. arXiv preprint arXiv:2304.08460, 2023. — [arXiv:2304.08460](https://arxiv.org/abs/2304.08460) find
  26. A. Köpf, Y. Kilcher, D. von Rütte, S. Anagnostidis, Z.-R. Tam, K. Stevens, A. Barhoum, N. M. Duc, O. Stanley, R. Nagyfi, et al. Openassistant conversations–democratizing large language model alignment. arXiv preprint arXiv:2304.07327, 2023. find our card
  27. X. Li, T. Zhang, Y. Dubois, R. Taori, I. Gulrajani, C. Guestrin, P. Liang, and T. B. Hashimoto. Alpacaeval: An automatic evaluator of instruction-following models. Github repository, 2023. URL https://github.com/tatsu-lab/alpaca\_eval. ↗
  28. P. Liang, R. Bommasani, T. Lee, D. Tsipras, D. Soylu, M. Yasunaga, Y. Zhang, D. Narayanan, Y. Wu, A. Kumar, B. Newman, B. Yuan, B. Yan, C. Zhang, C. Cosgrove, C. D. Manning, C. R’e, D. Acosta-Navas, D. A. Hudson, E. Zelikman, E. Durmus, F. Ladhak, F. Rong, H. Ren, H. Yao, J. Wang, K. Santhanam, L. J… find
  29. S. Lin, J. Hilton, and O. 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 Linguistics. doi: 10… find our card
  30. S. Lin, J. Hilton, and O. 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, 2022. find our card
  31. S. Longpre, L. Hou, T. Vu, A. Webson, H. W. Chung, Y. Tay, D. Zhou, Q. V. Le, B. Zoph, J. Wei, et al. The flan collection: Designing data and methods for effective instruction tuning. arXiv preprint arXiv:2301.13688, 2023. find our card
  32. S. Mishra, D. Khashabi, C. Baral, and H. Hajishirzi. Cross-Task Generalization via Natural Language Crowdsourcing Instructions. In Annual Meeting of the Association for Computational Linguistics (ACL), 2022. find
  33. MosaicML. Introducing mpt-7b: A new standard for open-source, commercially usable llms. Blog post, 2023. URL https://www.mosaicml.com/blog/mpt-7b. ↗
  34. OpenAI. Gpt-4 technical report. arXiv preprint arXiv:2303.08774, 2023. find our card
  35. L. Ouyang, J. Wu, X. Jiang, D. Almeida, C. L. Wainwright, P. Mishkin, C. Zhang, S. Agarwal, K. Slama, A. Ray, et al. Training Language Models to Follow Instructions with Human Feedback. In Advances in Neural Information Processing Systems (NeurIPS), 2022. find
  36. B. Peng, C. Li, P. He, M. Galley, and J. Gao. Instruction tuning with gpt-4. arXiv preprint arXiv:2304.03277, 2023. find our card
  37. S. Rajbhandari, J. Rasley, O. Ruwase, and Y. He. Zero: Memory optimizations toward training trillion parameter models. In Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis, SC ’20. IEEE Press, 2020. ISBN 9781728199986. find
  38. J. Rasley, S. Rajbhandari, O. Ruwase, and Y. He. Deepspeed: System optimizations enable training deep learning models with over 100 billion parameters. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, 2020. find
  39. V. Sanh, A. Webson, C. Raffel, S. Bach, L. Sutawika, Z. Alyafeai, A. Chaffin, A. Stiegler, A. Raja, M. Dey, M. S. Bari, C. Xu, U. Thakker, S. S. Sharma, E. Szczechla, T. Kim, G. Chhablani, N. Nayak, D. Datta, J. Chang, M. T.-J. Jiang, H. Wang, M. Manica, S. Shen, Z. X. Yong, H. Pandey, R. Bawden, T.… find our card
  40. N. Shazeer, A. Mirhoseini, K. Maziarz, A. Davis, Q. Le, G. Hinton, and J. Dean. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. In International Conference on Learning Representations, 2017. find
  41. A. Srivastava, A. Rastogi, A. Rao, A. A. M. Shoeb, A. Abid, A. Fisch, A. R. Brown, A. Santoro, and et al. Beyond the imitation game: Quantifying and extrapolating the capabilities of language models. arXiv preprint arXiv:2206.04615, 2022. find our card
  42. M. Suzgun, N. Scales, N. Schärli, S. Gehrmann, Y. Tay, H. W. Chung, A. Chowdhery, Q. V. Le, E. H. Chi, D. Zhou, et al. Challenging big-bench tasks and whether chain-of-thought can solve them. arXiv preprint arXiv:2210.09261, 2022. find our card
  43. R. Taori, I. Gulrajani, T. Zhang, Y. Dubois, X. Li, C. Guestrin, P. Liang, and T. B. Hashimoto. Stanford alpaca: An instruction-following llama model. GitHub repository, 2023. URL https: //github.com/tatsu-lab/stanford\_alpaca. find
  44. H. Touvron, T. Lavril, G. Izacard, X. Martinet, M.-A. Lachaux, T. Lacroix, B. Rozière, N. Goyal, E. Hambro, F. Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023. find
  45. H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y. Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosale, et al. Llama 2: Open foundation and fine-tuned chat models. arXiv preprint arXiv:2307.09288, 2023. find
  46. P. Wang, L. Li, L. Chen, D. Zhu, B. Lin, Y. Cao, Q. Liu, T. Liu, and Z. Sui. Large language models are not fair evaluators. arXiv preprint arXiv:2305.17926, 2023. find
  47. Y. Wang, Y. Kordi, S. Mishra, A. Liu, N. A. Smith, D. Khashabi, and H. Hajishirzi. Self-instruct: Aligning language model with self generated instructions. arXiv preprint arXiv:2212.10560, 2022. find
  48. Y. Wang, S. Mishra, P. Alipoormolabashi, Y. Kordi, A. Mirzaei, A. Arunkumar, A. Ashok, A. S. Dhanasekaran, A. Naik, D. Stap, et al. Super-NaturalInstructions: Generalization via Declarative Instructions on 1600+ Tasks. In EMNLP, 2022. find
  49. J. Wei, M. Bosma, V. Zhao, K. Guu, A. W. Yu, B. Lester, N. Du, A. M. Dai, and Q. V. Le. Finetuned Language Models are Zero-Shot Learners. In International Conference on Learning Representations (ICLR), 2022. find
  50. J. Wei, X. Wang, D. Schuurmans, M. Bosma, E. Chi, Q. Le, and D. Zhou. Chain of thought prompting elicits reasoning in large language models. arXiv preprint arXiv:2201.11903, 2022. find our card
  51. O. Weller, N. Lourie, M. Gardner, and M. E. Peters. Learning from task descriptions. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1361–1375, Online, Nov. 2020. Association for Computational Linguistics. doi: 10.18653/ v1/2020.emnlp-main.105… find
  52. C. Xu, D. Guo, N. Duan, and J. McAuley. Baize: An open-source chat model with parameter-efficient tuning on self-chat data. arXiv preprint arXiv:2304.01196, 2023. find our card
  53. C. Xu, Q. Sun, K. Zheng, X. Geng, P. Zhao, J. Feng, C. Tao, and D. Jiang. Wizardlm: Empowering large language models to follow complex instructions. arXiv preprint arXiv:2304.12244, 2023. find our card
  54. S. Zhang, S. Roller, N. Goyal, M. Artetxe, M. Chen, S. Chen, C. Dewan, M. Diab, X. Li, X. V. Lin, et al. Opt: Open pre-trained transformer language models. arXiv preprint arXiv:2205.01068, 2022. find
  55. L. Zheng, Y. Sheng, W.-L. Chiang, H. Zhang, J. E. Gonzalez, and I. Stoica. Chatbot Arena: Benchmarking LLMs in the Wild with Elo Ratings. Blog post, May 2023. URL https: //lmsys.org/blog/2023-05-03-arena/. find
  56. C. Zhou, P. Liu, P. Xu, S. Iyer, J. Sun, Y. Mao, X. Ma, A. Efrat, P. Yu, L. Yu, et al. Lima: Less is more for alignment. arXiv preprint arXiv:2305.11206, 2023. find