--- title: "Rethinking Benchmark and Contamination for — Yang 2023" description: "Public LLM benchmarks can leak into training corpora." authors: ["Shuo Yang", "Wei-Lin Chiang", "Lianmin Zheng", "Joseph E. Gonzalez", "Ion Stoica"] year: 2023 venue: "arXiv:2311.04850" source: "https://arxiv.org/abs/2311.04850" doi: "10.48550/arxiv.2311.04850" arxiv: "2311.04850" licence: "arXiv — non-exclusive licence to distribute" licence_url: "http://arxiv.org/licenses/nonexclusive-distrib/1.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/yang-2023-rethinking-benchmark-contamination-language-models/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/yang-2023-rethinking-benchmark-contamination-language-models/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2023](https://md.research.elamur.ai/free/en/papers/2023/page/5/) / Rethinking Benchmark and Contamination for Language Models with Rephrased Samples # Rethinking Benchmark and Contamination for Language Models with Rephrased Samples **Authors:** Shuo Yang, [Wei-Lin Chiang](https://md.research.elamur.ai/free/en/authors/wei-lin-chiang/), [Lianmin Zheng](https://md.research.elamur.ai/free/en/authors/lianmin-zheng/), [Joseph E. Gonzalez](https://md.research.elamur.ai/free/en/authors/joseph-e-gonzalez/), [Ion Stoica](https://md.research.elamur.ai/free/en/authors/ion-stoica/) · **Year:** 2023 · **Venue:** arXiv:2311.04850 · **Licence:** [arXiv — non-exclusive licence to distribute](http://arxiv.org/licenses/nonexclusive-distrib/1.0/) **Links:** [Publisher's page](https://arxiv.org/abs/2311.04850) · [PDF](https://arxiv.org/pdf/2311.04850) · [DOI 10.48550/arxiv.2311.04850](https://doi.org/10.48550/arxiv.2311.04850) · [arXiv 2311.04850](https://arxiv.org/abs/2311.04850) · [OpenAlex](https://openalex.org/W4388555916) **Citations:** 7 citations (openalex-snapshot · 2026-10-14) · [93](#cited-by) in our library, of the 140673 papers whose bibliographies we read ## Summary Public LLM benchmarks can leak into training corpora. Standard decontamination looks for surface similarity, using either n-gram overlap or an embedding-similarity threshold. The authors show that these checks are easy to get around. They had GPT-4 reword test items, or rewrite code in a different style or programming language, so that the meaning stayed the same but n-gram tests no longer flagged them. They then fine-tuned Llama-2-13B and CodeLlama on these "rephrased samples" from MMLU, GSM-8k and HumanEval. Scores rose to about GPT-4 level, and string matching did not detect the leak. As a fix, they propose the LLM decontaminator. For each test case, it first uses embedding search to find the top-k most similar training items, and then a strong LLM such as GPT-4 decides whether any of them is really the same problem. Run on real datasets, it found overlap that had not been reported before in pre-training corpora, instruction data and GPT-generated synthetic sets. This matters because leaderboard numbers may overstate what models can do. The authors release their tool and call for new, single-use evaluations similar to coding contests. ## Key points - A Llama-2-13B fine-tuned on rephrased MMLU test questions reached 85.9 accuracy, and n-gram overlap did not detect the contamination. - The decontaminator works in two stages: embedding similarity search retrieves the top-k candidates, and then an LLM (e.g., GPT-4) judges them. The tool is open-sourced at github.com/lm-sys/llm-decontaminator. - About 8–18% of HumanEval appears in rephrased form in subsets of RedPajama-Data-1T and StarCoder-Data. - CodeAlpaca, a synthetic dataset produced with GPT-3.5, contains rephrased versions of 12.8% of HumanEval. This shows that training on LLM-generated data can introduce contamination by accident. - The method also finds overlap inside the MATH benchmark itself, with about 1.6% of test items reworded in its training split. In the FLAN CoT subset, it flagged 76 MMLU test items (≈0.54%). *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](https://md.research.elamur.ai/free/en/shelves/mathematics/) ## Cited by Cited by 93 papers of our library cite it (of the 140673 whose bibliographies we read); 6 of them have a page here, 87 not yet. - [ChatGLM: A Family of Large Language Models from GLM-130B to GLM-4 All Tools](https://md.research.elamur.ai/free/en/papers/glm-2024-chatglm-family-large-language-models/) - [LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for Code](https://md.research.elamur.ai/free/en/papers/jain-2024-livecodebench-holistic-contamination-free-evaluation/) - [Tulu 3: Pushing Frontiers in Open Language Model Post-Training](https://md.research.elamur.ai/free/en/papers/lambert-2024-tulu-3-pushing-frontiers-open/) - [From Crowdsourced Data to High-Quality Benchmarks: Arena-Hard and BenchBuilder Pipeline](https://md.research.elamur.ai/free/en/papers/li-2024-crowdsourced-data-high-quality-benchmarks/) - [DataComp-LM: In search of the next generation of training sets for language models](https://md.research.elamur.ai/free/en/papers/li-2024-datacomp-lm-search-next-generation/) - [When Flores Bloomz Wrong: Cross-Direction Contamination in Machine Translation Evaluation](https://md.research.elamur.ai/free/en/papers/tan-2026-when-flores-bloomz-wrong-cross/) ## Cite The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @misc{yang2023rethinking, title = {{Rethinking Benchmark and Contamination for Language Models with Rephrased Samples}}, author = {Yang, Shuo and Chiang, Wei-Lin and Zheng, Lianmin and Gonzalez, Joseph E. and Stoica, Ion}, year = {2023}, doi = {10.48550/arxiv.2311.04850}, eprint = {2311.04850}, archivePrefix = {arXiv}, url = {https://doi.org/10.48550/arxiv.2311.04850} } ``` - [← previous Results of WMT23 Metrics Shared Task: Metrics Might Be Guilty but References Are Not Innocent](https://md.research.elamur.ai/free/en/papers/freitag-2023-results-wmt23-metrics-shared-task/) - [next → Retrieval meets Long Context Large Language Models](https://md.research.elamur.ai/free/en/papers/xu-2023-retrieval-meets-long-context-large/) - [↑ back to papers of 2023](https://md.research.elamur.ai/free/en/papers/2023/page/5/) ## References 42 references, 12 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Anil, R., Dai, A. M., Firat, O., Johnson, M., Lepikhin, D., Passos, A., Shakeri, S., Taropa, E., Bailey, P., Chen, Z., Chu, E., Clark, J. H., Shafey, L. E., Huang, Y., Meier-Hellstern, K., Mishra, G., Moreira, E., Omernick, M., Robinson, K., Ruder, S., Tay, Y., Xiao, K., Xu, Y., Zhang, Y., Abrego, G… [find](https://scholar.google.com/scholar?q=Anil%2C+R%2E%2C+Dai%2C+A%2E+M%2E%2C+Firat%2C+O%2E%2C+Johnson%2C+M%2E%2C+Lepikhin%2C+D%2E%2C+Passos%2C+A%2E%2C+Shakeri%2C+S%2E%2C+Taropa%2C+E%2E%2C+Bailey%2C+P%2E%2C+Chen%2C+Z%2E%2C+Chu%2C+E%2E%2C+Clark%2C+J%2E+H%2E%2C+Shafey%2C+L%2E+E%2E%2C+Huang%2C+Y%2E%2C+Meier%2DHellstern%2C+K%2E%2C+Mishra%2C+G%2E%2C) [our card](https://md.research.elamur.ai/free/en/papers/anil-2023-palm-2-technical-report/) 2. Balle, B., Cherubin, G., and Hayes, J. Reconstructing training data with informed adversaries. In 2022 IEEE Symposium on Security and Privacy (SP), pp. 1138–1156. [find](https://scholar.google.com/scholar?q=Balle%2C+B%2E%2C+Cherubin%2C+G%2E%2C+and+Hayes%2C+J%2E+Reconstructing+training+data+with+informed+adversaries%2E+In+2022+IEEE+Symposium+on+Security+and+Privacy+%28SP%29%2C+pp%2E+1138%E2%80%931156%2E) 3. Brown, T. B., Mann, B., Ryder, N., Subbiah, M., Kaplan, J., Dhariwal, P., Neelakantan, A., Shyam, P., Sastry, G., Askell, A., Agarwal, S., Herbert-Voss, A., Krueger, G., Henighan, T., Child, R., Ramesh, A., Ziegler, D. M., Wu, J., Winter, C., Hesse, C., Chen, M., Sigler, E., Litwin, M., Gray, S., Ch… [find](https://scholar.google.com/scholar?q=Brown%2C+T%2E+B%2E%2C+Mann%2C+B%2E%2C+Ryder%2C+N%2E%2C+Subbiah%2C+M%2E%2C+Kaplan%2C+J%2E%2C+Dhariwal%2C+P%2E%2C+Neelakantan%2C+A%2E%2C+Shyam%2C+P%2E%2C+Sastry%2C+G%2E%2C+Askell%2C+A%2E%2C+Agarwal%2C+S%2E%2C+Herbert%2DVoss%2C+A%2E%2C+Krueger%2C+G%2E%2C+Henighan%2C+T%2E%2C+Child%2C+R%2E%2C+Rames) 4. Carlini, N., Tramer, F., Wallace, E., Jagielski, M., Herbert-Voss, A., Lee, K., Roberts, A., Brown, T., Song, D., Erlingsson, U., Oprea, A., and Raffel, C. Extracting training data from large language models, 2021. [find](https://scholar.google.com/scholar?q=Carlini%2C+N%2E%2C+Tramer%2C+F%2E%2C+Wallace%2C+E%2E%2C+Jagielski%2C+M%2E%2C+Herbert%2DVoss%2C+A%2E%2C+Lee%2C+K%2E%2C+Roberts%2C+A%2E%2C+Brown%2C+T%2E%2C+Song%2C+D%2E%2C+Erlingsson%2C+U%2E%2C+Oprea%2C+A%2E%2C+and+Raffel%2C+C%2E+Extracting+training+data+from+large+language) [our card](https://md.research.elamur.ai/free/en/papers/carlini-2020-extracting-training-data-large-language/) 5. Carlini, N., Ippolito, D., Jagielski, M., Lee, K., Tramer, F., and Zhang, C. Quantifying memorization across neural language models, 2023. [find](https://scholar.google.com/scholar?q=Carlini%2C+N%2E%2C+Ippolito%2C+D%2E%2C+Jagielski%2C+M%2E%2C+Lee%2C+K%2E%2C+Tramer%2C+F%2E%2C+and+Zhang%2C+C%2E+Quantifying+memorization+across+neural+language+models%2C+2023%2E) 6. Chang, Y., Wang, X., Wang, J., Wu, Y., Zhu, K., Chen, H., Yang, L., Yi, X., Wang, C., Wang, Y., et al. A survey on evaluation of large language models. arXiv preprint arXiv:2307.03109, 2023. [find](https://scholar.google.com/scholar?q=Chang%2C+Y%2E%2C+Wang%2C+X%2E%2C+Wang%2C+J%2E%2C+Wu%2C+Y%2E%2C+Zhu%2C+K%2E%2C+Chen%2C+H%2E%2C+Yang%2C+L%2E%2C+Yi%2C+X%2E%2C+Wang%2C+C%2E%2C+Wang%2C+Y%2E%2C+et+al%2E+A+survey+on+evaluation+of+large+language+models%2E+arXiv+preprint+arXiv%3A2307%2E03109%2C+2023%2E) [our card](https://md.research.elamur.ai/free/en/papers/chang-2023-survey-evaluation-large-language-models/) 7. Chaudhary, S. Code alpaca: An instruction-following llama model for code generation. [https://github.com/sahil280114/codealpaca](https://github.com/sahil280114/codealpaca), 2023. [↗](https://github.com/sahil280114/codealpaca) 8. Chen, M., Tworek, J., Jun, H., Yuan, Q., Pinto, H. P. d. O., Kaplan, J., Edwards, H., Burda, Y., Joseph, N., Brockman, G., et al. Evaluating large language models trained on code. arXiv preprint arXiv:2107.03374, 2021. [find](https://scholar.google.com/scholar?q=Chen%2C+M%2E%2C+Tworek%2C+J%2E%2C+Jun%2C+H%2E%2C+Yuan%2C+Q%2E%2C+Pinto%2C+H%2E+P%2E+d%2E+O%2E%2C+Kaplan%2C+J%2E%2C+Edwards%2C+H%2E%2C+Burda%2C+Y%2E%2C+Joseph%2C+N%2E%2C+Brockman%2C+G%2E%2C+et+al%2E+Evaluating+large+language+models+trained+on+code%2E+arXiv+preprint+arXiv) [our card](https://md.research.elamur.ai/free/en/papers/chen-2021-evaluating-large-language-models-trained/) 9. Cobbe, K., Kosaraju, V., Bavarian, M., Chen, M., Jun, H., Kaiser, L., Plappert, M., Tworek, J., Hilton, J., Nakano, R., et al. Training verifiers to solve math word problems. arXiv preprint arXiv:2110.14168, 2021. [find](https://scholar.google.com/scholar?q=Cobbe%2C+K%2E%2C+Kosaraju%2C+V%2E%2C+Bavarian%2C+M%2E%2C+Chen%2C+M%2E%2C+Jun%2C+H%2E%2C+Kaiser%2C+L%2E%2C+Plappert%2C+M%2E%2C+Tworek%2C+J%2E%2C+Hilton%2C+J%2E%2C+Nakano%2C+R%2E%2C+et+al%2E+Training+verifiers+to+solve+math+word+problems%2E+arXiv+preprint+arXiv%3A2110) [our card](https://md.research.elamur.ai/free/en/papers/cobbe-2021-training-verifiers-solve-math-word/) 10. Computer, T. Redpajama: An open source recipe to reproduce llama training dataset, April 2023. URL [https://github.com/togethercomputer/RedPajama-Data](https://github.com/togethercomputer/RedPajama-Data). [↗](https://github.com/togethercomputer/RedPajama-Data) 11. Dodge, J., Sap, M., Marasović, A., Agnew, W., Ilharco, G., Groeneveld, D., Mitchell, M., and Gardner, M. Documenting large webtext corpora: A case study on the colossal clean crawled corpus. arXiv preprint arXiv:2104.08758, 2021. [find](https://scholar.google.com/scholar?q=Dodge%2C+J%2E%2C+Sap%2C+M%2E%2C+Marasovic%CC%81%2C+A%2E%2C+Agnew%2C+W%2E%2C+Ilharco%2C+G%2E%2C+Groeneveld%2C+D%2E%2C+Mitchell%2C+M%2E%2C+and+Gardner%2C+M%2E+Documenting+large+webtext+corpora%3A+A+case+study+on+the+colossal+clean+crawled+corpus%2E+arXiv+pr) 12. Geng, X. and Liu, H. Openllama: An open reproduction of llama, May 2023. URL [https://github.com/openlm-research/open_llama](https://github.com/openlm-research/open_llama). [↗](https://github.com/openlm-research/open_llama) 13. Gunasekar, S., Zhang, Y., Aneja, J., Mendes, C. C. T., Giorno, A. D., Gopi, S., Javaheripi, M., Kauffmann, P., de Rosa, G., Saarikivi, O., Salim, A., Shah, S., Behl, H. S., Wang, X., Bubeck, S., Eldan, R., Kalai, A. T., Lee, Y. T., and Li, Y. Textbooks are all you need, 2023. [find](https://scholar.google.com/scholar?q=Gunasekar%2C+S%2E%2C+Zhang%2C+Y%2E%2C+Aneja%2C+J%2E%2C+Mendes%2C+C%2E+C%2E+T%2E%2C+Giorno%2C+A%2E+D%2E%2C+Gopi%2C+S%2E%2C+Javaheripi%2C+M%2E%2C+Kauffmann%2C+P%2E%2C+de+Rosa%2C+G%2E%2C+Saarikivi%2C+O%2E%2C+Salim%2C+A%2E%2C+Shah%2C+S%2E%2C+Behl%2C+H%2E+S%2E%2C+Wang%2C+X%2E%2C+Bubeck%2C+S%2E%2C+Eldan) 14. Hendrycks, D., Burns, C., Basart, S., Zou, A., Mazeika, M., Song, D., and Steinhardt, J. Measuring massive multitask language understanding. arXiv preprint arXiv:2009.03300, 2020. [find](https://scholar.google.com/scholar?q=Hendrycks%2C+D%2E%2C+Burns%2C+C%2E%2C+Basart%2C+S%2E%2C+Zou%2C+A%2E%2C+Mazeika%2C+M%2E%2C+Song%2C+D%2E%2C+and+Steinhardt%2C+J%2E+Measuring+massive+multitask+language+understanding%2E+arXiv+preprint+arXiv%3A2009%2E03300%2C+2020%2E) [our card](https://md.research.elamur.ai/free/en/papers/hendrycks-2020-measuring-massive-multitask-language-understanding/) 15. Hendrycks, D., Burns, C., Kadavath, S., Arora, A., Basart, S., Tang, E., Song, D., and Steinhardt, J. Measuring mathematical problem solving with the math dataset. NeurIPS, 2021. [find](https://scholar.google.com/scholar?q=Hendrycks%2C+D%2E%2C+Burns%2C+C%2E%2C+Kadavath%2C+S%2E%2C+Arora%2C+A%2E%2C+Basart%2C+S%2E%2C+Tang%2C+E%2E%2C+Song%2C+D%2E%2C+and+Steinhardt%2C+J%2E+Measuring+mathematical+problem+solving+with+the+math+dataset%2E+NeurIPS%2C+2021%2E) [our card](https://md.research.elamur.ai/free/en/papers/hendrycks-2021-measuring-mathematical-problem-solving-math/) 16. Kiela, D., Bartolo, M., Nie, Y., Kaushik, D., Geiger, A., Wu, Z., Vidgen, B., Prasad, G., Singh, A., Ringshia, P., Ma, Z., Thrush, T., Riedel, S., Waseem, Z., Stenetorp, P., Jia, R., Bansal, M., Potts, C., and Williams, A. Dynabench: Rethinking benchmarking in nlp, 2021. [find](https://scholar.google.com/scholar?q=Kiela%2C+D%2E%2C+Bartolo%2C+M%2E%2C+Nie%2C+Y%2E%2C+Kaushik%2C+D%2E%2C+Geiger%2C+A%2E%2C+Wu%2C+Z%2E%2C+Vidgen%2C+B%2E%2C+Prasad%2C+G%2E%2C+Singh%2C+A%2E%2C+Ringshia%2C+P%2E%2C+Ma%2C+Z%2E%2C+Thrush%2C+T%2E%2C+Riedel%2C+S%2E%2C+Waseem%2C+Z%2E%2C+Stenetorp%2C+P%2E%2C+Jia%2C+R%2E%2C+Bansal%2C+M%2E%2C+Potts) 17. Kocetkov, D., Li, R., Allal, L. B., Li, J., Mou, C., Ferrandis, C. M., Jernite, Y., Mitchell, M., Hughes, S., Wolf, T., Bahdanau, D., von Werra, L., and de Vries, H. The stack: 3 tb of permissively licensed source code, 2022. [find](https://scholar.google.com/scholar?q=Kocetkov%2C+D%2E%2C+Li%2C+R%2E%2C+Allal%2C+L%2E+B%2E%2C+Li%2C+J%2E%2C+Mou%2C+C%2E%2C+Ferrandis%2C+C%2E+M%2E%2C+Jernite%2C+Y%2E%2C+Mitchell%2C+M%2E%2C+Hughes%2C+S%2E%2C+Wolf%2C+T%2E%2C+Bahdanau%2C+D%2E%2C+von+Werra%2C+L%2E%2C+and+de+Vries%2C+H%2E+The+stack%3A+3+tb+of+permissively+li) 18. Koh, P. W. and Liang, P. Understanding black-box predictions via influence functions, 2020. [find](https://scholar.google.com/scholar?q=Koh%2C+P%2E+W%2E+and+Liang%2C+P%2E+Understanding+black%2Dbox+predictions+via+influence+functions%2C+2020%2E) 19. Lee, A. N., Hunter, C. J., and Ruiz, N. Platypus: Quick, cheap, and powerful refinement of llms, 2023. [find](https://scholar.google.com/scholar?q=Lee%2C+A%2E+N%2E%2C+Hunter%2C+C%2E+J%2E%2C+and+Ruiz%2C+N%2E+Platypus%3A+Quick%2C+cheap%2C+and+powerful+refinement+of+llms%2C+2023%2E) 20. Li, R., Allal, L. B., Zi, Y., Muennighoff, N., Kocetkov, D., Mou, C., Marone, M., Akiki, C., Li, J., Chim, J., Liu, Q., Zheltonozhskii, E., Zhuo, T. Y., Wang, T., Dehaene, O., Davaadorj, M., Lamy-Poirier, J., Monteiro, J., Shliazhko, O., Gontier, N., Meade, N., Zebaze, A., Yee, M.-H., Umapathi, L. K… [find](https://scholar.google.com/scholar?q=Li%2C+R%2E%2C+Allal%2C+L%2E+B%2E%2C+Zi%2C+Y%2E%2C+Muennighoff%2C+N%2E%2C+Kocetkov%2C+D%2E%2C+Mou%2C+C%2E%2C+Marone%2C+M%2E%2C+Akiki%2C+C%2E%2C+Li%2C+J%2E%2C+Chim%2C+J%2E%2C+Liu%2C+Q%2E%2C+Zheltonozhskii%2C+E%2E%2C+Zhuo%2C+T%2E+Y%2E%2C+Wang%2C+T%2E%2C+Dehaene%2C+O%2E%2C+Davaadorj%2C+M%2E%2C+Lamy%2DPoir) 21. Li, Y. Estimating contamination via perplexity: Quantifying memorisation in language model evaluation, 2023. [find](https://scholar.google.com/scholar?q=Li%2C+Y%2E+Estimating+contamination+via+perplexity%3A+Quantifying+memorisation+in+language+model+evaluation%2C+2023%2E) 22. Longpre, S., Hou, L., Vu, T., Webson, A., Chung, H. W., Tay, Y., Zhou, D., Le, Q. V., Zoph, B., Wei, J., and Roberts, A. The flan collection: Designing data and methods for effective instruction tuning, 2023. [find](https://scholar.google.com/scholar?q=Longpre%2C+S%2E%2C+Hou%2C+L%2E%2C+Vu%2C+T%2E%2C+Webson%2C+A%2E%2C+Chung%2C+H%2E+W%2E%2C+Tay%2C+Y%2E%2C+Zhou%2C+D%2E%2C+Le%2C+Q%2E+V%2E%2C+Zoph%2C+B%2E%2C+Wei%2C+J%2E%2C+and+Roberts%2C+A%2E+The+flan+collection%3A+Designing+data+and+methods+for+effective+instruction+tunin) [our card](https://md.research.elamur.ai/free/en/papers/longpre-2023-flan-collection-designing-data-methods/) 23. Ma, Z., Ethayarajh, K., Thrush, T., Jain, S., Wu, L., Jia, R., Potts, C., Williams, A., and Kiela, D. Dynaboard: An evaluation-as-a-service platform for holistic next-generation benchmarking. Advances in Neural Information Processing Systems, 34:10351–10367, 2021. [find](https://scholar.google.com/scholar?q=Ma%2C+Z%2E%2C+Ethayarajh%2C+K%2E%2C+Thrush%2C+T%2E%2C+Jain%2C+S%2E%2C+Wu%2C+L%2E%2C+Jia%2C+R%2E%2C+Potts%2C+C%2E%2C+Williams%2C+A%2E%2C+and+Kiela%2C+D%2E+Dynaboard%3A+An+evaluation%2Das%2Da%2Dservice+platform+for+holistic+next%2Dgeneration+benchmarking%2E+Advances) 24. Mukherjee, S., Mitra, A., Jawahar, G., Agarwal, S., Palangi, H., and Awadallah, A. Orca: Progressive learning from complex explanation traces of gpt-4, 2023. [find](https://scholar.google.com/scholar?q=Mukherjee%2C+S%2E%2C+Mitra%2C+A%2E%2C+Jawahar%2C+G%2E%2C+Agarwal%2C+S%2E%2C+Palangi%2C+H%2E%2C+and+Awadallah%2C+A%2E+Orca%3A+Progressive+learning+from+complex+explanation+traces+of+gpt%2D4%2C+2023%2E) [our card](https://md.research.elamur.ai/free/en/papers/mukherjee-2023-orca-progressive-learning-complex-explanation/) 25. Olson, M., Wyner, A., and Berk, R. Modern neural networks generalize on small data sets. Advances in neural information processing systems, 31, 2018. [find](https://scholar.google.com/scholar?q=Olson%2C+M%2E%2C+Wyner%2C+A%2E%2C+and+Berk%2C+R%2E+Modern+neural+networks+generalize+on+small+data+sets%2E+Advances+in+neural+information+processing+systems%2C+31%2C+2018%2E) 26. OpenAI. Gpt-4 technical report, 2023. [find](https://scholar.google.com/scholar?q=OpenAI%2E+Gpt%2D4+technical+report%2C+2023%2E) [our card](https://md.research.elamur.ai/free/en/papers/openai-2023-gpt-4-technical-report/) 27. Oren, Y., Meister, N., Chatterji, N., Ladhak, F., and Hashimoto, T. B. Proving test set contamination in black box language models, 2023. [find](https://scholar.google.com/scholar?q=Oren%2C+Y%2E%2C+Meister%2C+N%2E%2C+Chatterji%2C+N%2E%2C+Ladhak%2C+F%2E%2C+and+Hashimoto%2C+T%2E+B%2E+Proving+test+set+contamination+in+black+box+language+models%2C+2023%2E) 28. Pan, X., Zhang, M., Ji, S., and Yang, M. Privacy risks of general-purpose language models. In 2020 IEEE Symposium on Security and Privacy (SP), pp. 1314–1331. [find](https://scholar.google.com/scholar?q=Pan%2C+X%2E%2C+Zhang%2C+M%2E%2C+Ji%2C+S%2E%2C+and+Yang%2C+M%2E+Privacy+risks+of+general%2Dpurpose+language+models%2E+In+2020+IEEE+Symposium+on+Security+and+Privacy+%28SP%29%2C+pp%2E+1314%E2%80%931331%2E) 29. Recht, B., Roelofs, R., Schmidt, L., and Shankar, V. Do imagenet classifiers generalize to imagenet?, 2019. [find](https://scholar.google.com/scholar?q=Recht%2C+B%2E%2C+Roelofs%2C+R%2E%2C+Schmidt%2C+L%2E%2C+and+Shankar%2C+V%2E+Do+imagenet+classifiers+generalize+to+imagenet%3F%2C+2019%2E) 30. Reimers, N. and Gurevych, I. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics, 11 2019. URL https://arxiv.org/abs/1908.10084. [↗](https://arxiv.org/abs/1908.10084) 31. Shi, W., Ajith, A., Xia, M., Huang, Y., Liu, D., Blevins, T., Chen, D., and Zettlemoyer, L. Detecting pretraining data from large language models, 2023. [find](https://scholar.google.com/scholar?q=Shi%2C+W%2E%2C+Ajith%2C+A%2E%2C+Xia%2C+M%2E%2C+Huang%2C+Y%2E%2C+Liu%2C+D%2E%2C+Blevins%2C+T%2E%2C+Chen%2C+D%2E%2C+and+Zettlemoyer%2C+L%2E+Detecting+pretraining+data+from+large+language+models%2C+2023%2E) 32. Taori, R., Gulrajani, I., Zhang, T., Dubois, Y., Li, X., Guestrin, C., Liang, P., and Hashimoto, T. B. Stanford alpaca: An instruction-following llama model. [https://github.com/tatsu-lab/stanford_alpaca](https://github.com/tatsu-lab/stanford_alpaca), 2023. [↗](https://github.com/tatsu-lab/stanford_alpaca) 33. Team, M. N. Introducing mpt-7b: A new standard for open-source, commercially usable llms, 2023. URL www. mosaicml.com/blog/mpt-7b. Accessed: 2023-05-05. [find](https://scholar.google.com/scholar?q=Team%2C+M%2E+N%2E+Introducing+mpt%2D7b%3A+A+new+standard+for+open%2Dsource%2C+commercially+usable+llms%2C+2023%2E+URL+www%2E+mosaicml%2Ecom%2Fblog%2Fmpt%2D7b%2E+Accessed%3A+2023%2D05%2D05%2E) 34. Touvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y., Bashlykov, N., Batra, S., Bhargava, P., Bhosale, S., Bikel, D., Blecher, L., Ferrer, C. C., Chen, M., Cucurull, G., Esiobu, D., Fernandes, J., Fu, J., Fu, W., Fuller, B., Gao, C., Goswami, V., Goyal, N., Hartshorn, A., Hossei… [find](https://scholar.google.com/scholar?q=Touvron%2C+H%2E%2C+Martin%2C+L%2E%2C+Stone%2C+K%2E%2C+Albert%2C+P%2E%2C+Almahairi%2C+A%2E%2C+Babaei%2C+Y%2E%2C+Bashlykov%2C+N%2E%2C+Batra%2C+S%2E%2C+Bhargava%2C+P%2E%2C+Bhosale%2C+S%2E%2C+Bikel%2C+D%2E%2C+Blecher%2C+L%2E%2C+Ferrer%2C+C%2E+C%2E%2C+Chen%2C+M%2E%2C+Cucurull%2C+G%2E%2C+Esiobu%2C+D) 35. Wang, Y., Ivison, H., Dasigi, P., Hessel, J., Khot, T., Chandu, K. R., Wadden, D., MacMillan, K., Smith, N. A., Beltagy, I., and Hajishirzi, H. How far can camels go? exploring the state of instruction tuning on open resources, 2023a. [find](https://scholar.google.com/scholar?q=Wang%2C+Y%2E%2C+Ivison%2C+H%2E%2C+Dasigi%2C+P%2E%2C+Hessel%2C+J%2E%2C+Khot%2C+T%2E%2C+Chandu%2C+K%2E+R%2E%2C+Wadden%2C+D%2E%2C+MacMillan%2C+K%2E%2C+Smith%2C+N%2E+A%2E%2C+Beltagy%2C+I%2E%2C+and+Hajishirzi%2C+H%2E+How+far+can+camels+go%3F+exploring+the+state+of+instructio) [our card](https://md.research.elamur.ai/free/en/papers/wang-2023-far-camels-go-exploring-state/) 36. Wang, Y., Kordi, Y., Mishra, S., Liu, A., Smith, N. A., Khashabi, D., and Hajishirzi, H. Self-instruct: Aligning language models with self-generated instructions, 2023b. [find](https://scholar.google.com/scholar?q=Wang%2C+Y%2E%2C+Kordi%2C+Y%2E%2C+Mishra%2C+S%2E%2C+Liu%2C+A%2E%2C+Smith%2C+N%2E+A%2E%2C+Khashabi%2C+D%2E%2C+and+Hajishirzi%2C+H%2E+Self%2Dinstruct%3A+Aligning+language+models+with+self%2Dgenerated+instructions%2C+2023b%2E) 37. Xu, C., Sun, Q., Zheng, K., Geng, X., Zhao, P., Feng, J., Tao, C., and Jiang, D. Wizardlm: Empowering large language models to follow complex instructions, 2023. [find](https://scholar.google.com/scholar?q=Xu%2C+C%2E%2C+Sun%2C+Q%2E%2C+Zheng%2C+K%2E%2C+Geng%2C+X%2E%2C+Zhao%2C+P%2E%2C+Feng%2C+J%2E%2C+Tao%2C+C%2E%2C+and+Jiang%2C+D%2E+Wizardlm%3A+Empowering+large+language+models+to+follow+complex+instructions%2C+2023%2E) 38. Xu, H. and Koehn, P. Zipporah: a fast and scalable data cleaning system for noisy web-crawled parallel corpora. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pp. 2945– 2950, 2017. [find](https://scholar.google.com/scholar?q=Xu%2C+H%2E+and+Koehn%2C+P%2E+Zipporah%3A+a+fast+and+scalable+data+cleaning+system+for+noisy+web%2Dcrawled+parallel+corpora%2E+In+Proceedings+of+the+2017+Conference+on+Empirical+Methods+in+Natural+Language+Processin) 39. Yue, X., Qu, X., Zhang, G., Fu, Y., Huang, W., Sun, H., Su, Y., and Chen, W. Mammoth: Building math generalist models through hybrid instruction tuning, 2023. [find](https://scholar.google.com/scholar?q=Yue%2C+X%2E%2C+Qu%2C+X%2E%2C+Zhang%2C+G%2E%2C+Fu%2C+Y%2E%2C+Huang%2C+W%2E%2C+Sun%2C+H%2E%2C+Su%2C+Y%2E%2C+and+Chen%2C+W%2E+Mammoth%3A+Building+math+generalist+models+through+hybrid+instruction+tuning%2C+2023%2E) [our card](https://md.research.elamur.ai/free/en/papers/yue-2023-mammoth-building-math-generalist-models/) 40. Zanella-Béguelin, S., Wutschitz, L., Tople, S., Rühle, V., Paverd, A., Ohrimenko, O., Köpf, B., and Brockschmidt, M. Analyzing information leakage of updates to natural language models. In Proceedings of the 2020 ACM SIGSAC conference on computer and communications security, pp. 363–375, 2020. [find](https://scholar.google.com/scholar?q=Zanella%2DBe%CC%81guelin%2C+S%2E%2C+Wutschitz%2C+L%2E%2C+Tople%2C+S%2E%2C+Ru%CC%88hle%2C+V%2E%2C+Paverd%2C+A%2E%2C+Ohrimenko%2C+O%2E%2C+Ko%CC%88pf%2C+B%2E%2C+and+Brockschmidt%2C+M%2E+Analyzing+information+leakage+of+updates+to+natural+language+models%2E+In+Proceedi) 41. Zhang, C., Bengio, S., Hardt, M., Recht, B., and Vinyals, O. Understanding deep learning requires rethinking generalization, 2017. [find](https://scholar.google.com/scholar?q=Zhang%2C+C%2E%2C+Bengio%2C+S%2E%2C+Hardt%2C+M%2E%2C+Recht%2C+B%2E%2C+and+Vinyals%2C+O%2E+Understanding+deep+learning+requires+rethinking+generalization%2C+2017%2E) 42. Zong, Y., Yu, T., Zhao, B., Chavhan, R., and Hospedales, T. Fool your (vision and) language model with embarrassingly simple permutations, 2023. [find](https://scholar.google.com/scholar?q=Zong%2C+Y%2E%2C+Yu%2C+T%2E%2C+Zhao%2C+B%2E%2C+Chavhan%2C+R%2E%2C+and+Hospedales%2C+T%2E+Fool+your+%28vision+and%29+language+model+with+embarrassingly+simple+permutations%2C+2023%2E)