Search

Your search keyword '"Gao, Leo"' showing total 16 results
Start Over Author "Gao, Leo" Search Limiters Available in Library Collection
16 results on '"Gao, Leo"'

1. Scaling and evaluating sparse autoencoders

Author

Gao, Leo, la Tour, Tom Dupré, Tillman, Henk, Goh, Gabriel, Troll, Rajan, Radford, Alec, Sutskever, Ilya, Leike, Jan, and Wu, Jeffrey

Subjects
Computer Science - Machine Learning, Computer Science - Artificial Intelligence
Abstract

Sparse autoencoders provide a promising unsupervised approach for extracting interpretable features from a language model by reconstructing activations from a sparse bottleneck layer. Since language models learn many concepts, autoencoders need to be very large to recover all relevant features. However, studying the properties of autoencoder scaling is difficult due to the need to balance reconstruction and sparsity objectives and the presence of dead latents. We propose using k-sparse autoencoders [Makhzani and Frey, 2013] to directly control sparsity, simplifying tuning and improving the reconstruction-sparsity frontier. Additionally, we find modifications that result in few dead latents, even at the largest scales we tried. Using these techniques, we find clean scaling laws with respect to autoencoder size and sparsity. We also introduce several new metrics for evaluating feature quality based on the recovery of hypothesized features, the explainability of activation patterns, and the sparsity of downstream effects. These metrics all generally improve with autoencoder size. To demonstrate the scalability of our approach, we train a 16 million latent autoencoder on GPT-4 activations for 40 billion tokens. We release training code and autoencoders for open-source models, as well as a visualizer.

Published
2024

2. Lessons from the Trenches on Reproducible Evaluation of Language Models

Author

Biderman, Stella, Schoelkopf, Hailey, Sutawika, Lintang, Gao, Leo, Tow, Jonathan, Abbasi, Baber, Aji, Alham Fikri, Ammanamanchi, Pawan Sasanka, Black, Sidney, Clive, Jordan, DiPofi, Anthony, Etxaniz, Julen, Fattori, Benjamin, Forde, Jessica Zosa, Foster, Charles, Hsu, Jeffrey, Jaiswal, Mimansa, Lee, Wilson Y., Li, Haonan, Lovering, Charles, Muennighoff, Niklas, Pavlick, Ellie, Phang, Jason, Skowron, Aviya, Tan, Samson, Tang, Xiangru, Wang, Kevin A., Winata, Genta Indra, Yvon, François, and Zou, Andy

Subjects
Computer Science - Computation and Language
Abstract

Effective evaluation of language models remains an open challenge in NLP. Researchers and engineers face methodological issues such as the sensitivity of models to evaluation setup, difficulty of proper comparisons across methods, and the lack of reproducibility and transparency. In this paper we draw on three years of experience in evaluating large language models to provide guidance and lessons for researchers. First, we provide an overview of common challenges faced in language model evaluation. Second, we delineate best practices for addressing or lessening the impact of these challenges on research. Third, we present the Language Model Evaluation Harness (lm-eval): an open source library for independent, reproducible, and extensible evaluation of language models that seeks to address these issues. We describe the features of the library as well as case studies in which the library has been used to alleviate these methodological concerns.

Published
2024

3. Weak-to-Strong Generalization: Eliciting Strong Capabilities With Weak Supervision

Author

Burns, Collin, Izmailov, Pavel, Kirchner, Jan Hendrik, Baker, Bowen, Gao, Leo, Aschenbrenner, Leopold, Chen, Yining, Ecoffet, Adrien, Joglekar, Manas, Leike, Jan, Sutskever, Ilya, and Wu, Jeff

Subjects
Computer Science - Computation and Language
Abstract

Widely used alignment techniques, such as reinforcement learning from human feedback (RLHF), rely on the ability of humans to supervise model behavior - for example, to evaluate whether a model faithfully followed instructions or generated safe outputs. However, future superhuman models will behave in complex ways too difficult for humans to reliably evaluate; humans will only be able to weakly supervise superhuman models. We study an analogy to this problem: can weak model supervision elicit the full capabilities of a much stronger model? We test this using a range of pretrained language models in the GPT-4 family on natural language processing (NLP), chess, and reward modeling tasks. We find that when we naively finetune strong pretrained models on labels generated by a weak model, they consistently perform better than their weak supervisors, a phenomenon we call weak-to-strong generalization. However, we are still far from recovering the full capabilities of strong models with naive finetuning alone, suggesting that techniques like RLHF may scale poorly to superhuman models without further work. We find that simple methods can often significantly improve weak-to-strong generalization: for example, when finetuning GPT-4 with a GPT-2-level supervisor and an auxiliary confidence loss, we can recover close to GPT-3.5-level performance on NLP tasks. Our results suggest that it is feasible to make empirical progress today on a fundamental challenge of aligning superhuman models.

Published
2023

4. GPT-4 Technical Report

Author

OpenAI, Achiam, Josh, Adler, Steven, Agarwal, Sandhini, Ahmad, Lama, Akkaya, Ilge, Aleman, Florencia Leoni, Almeida, Diogo, Altenschmidt, Janko, Altman, Sam, Anadkat, Shyamal, Avila, Red, Babuschkin, Igor, Balaji, Suchir, Balcom, Valerie, Baltescu, Paul, Bao, Haiming, Bavarian, Mohammad, Belgum, Jeff, Bello, Irwan, Berdine, Jake, Bernadett-Shapiro, Gabriel, Berner, Christopher, Bogdonoff, Lenny, Boiko, Oleg, Boyd, Madelaine, Brakman, Anna-Luisa, Brockman, Greg, Brooks, Tim, Brundage, Miles, Button, Kevin, Cai, Trevor, Campbell, Rosie, Cann, Andrew, Carey, Brittany, Carlson, Chelsea, Carmichael, Rory, Chan, Brooke, Chang, Che, Chantzis, Fotis, Chen, Derek, Chen, Sully, Chen, Ruby, Chen, Jason, Chen, Mark, Chess, Ben, Cho, Chester, Chu, Casey, Chung, Hyung Won, Cummings, Dave, Currier, Jeremiah, Dai, Yunxing, Decareaux, Cory, Degry, Thomas, Deutsch, Noah, Deville, Damien, Dhar, Arka, Dohan, David, Dowling, Steve, Dunning, Sheila, Ecoffet, Adrien, Eleti, Atty, Eloundou, Tyna, Farhi, David, Fedus, Liam, Felix, Niko, Fishman, Simón Posada, Forte, Juston, Fulford, Isabella, Gao, Leo, Georges, Elie, Gibson, Christian, Goel, Vik, Gogineni, Tarun, Goh, Gabriel, Gontijo-Lopes, Rapha, Gordon, Jonathan, Grafstein, Morgan, Gray, Scott, Greene, Ryan, Gross, Joshua, Gu, Shixiang Shane, Guo, Yufei, Hallacy, Chris, Han, Jesse, Harris, Jeff, He, Yuchen, Heaton, Mike, Heidecke, Johannes, Hesse, Chris, Hickey, Alan, Hickey, Wade, Hoeschele, Peter, Houghton, Brandon, Hsu, Kenny, Hu, Shengli, Hu, Xin, Huizinga, Joost, Jain, Shantanu, Jain, Shawn, Jang, Joanne, Jiang, Angela, Jiang, Roger, Jin, Haozhun, Jin, Denny, Jomoto, Shino, Jonn, Billie, Jun, Heewoo, Kaftan, Tomer, Kaiser, Łukasz, Kamali, Ali, Kanitscheider, Ingmar, Keskar, Nitish Shirish, Khan, Tabarak, Kilpatrick, Logan, Kim, Jong Wook, Kim, Christina, Kim, Yongjik, Kirchner, Jan Hendrik, Kiros, Jamie, Knight, Matt, Kokotajlo, Daniel, Kondraciuk, Łukasz, Kondrich, Andrew, Konstantinidis, Aris, Kosic, Kyle, Krueger, Gretchen, Kuo, Vishal, Lampe, Michael, Lan, Ikai, Lee, Teddy, Leike, Jan, Leung, Jade, Levy, Daniel, Li, Chak Ming, Lim, Rachel, Lin, Molly, Lin, Stephanie, Litwin, Mateusz, Lopez, Theresa, Lowe, Ryan, Lue, Patricia, Makanju, Anna, Malfacini, Kim, Manning, Sam, Markov, Todor, Markovski, Yaniv, Martin, Bianca, Mayer, Katie, Mayne, Andrew, McGrew, Bob, McKinney, Scott Mayer, McLeavey, Christine, McMillan, Paul, McNeil, Jake, Medina, David, Mehta, Aalok, Menick, Jacob, Metz, Luke, Mishchenko, Andrey, Mishkin, Pamela, Monaco, Vinnie, Morikawa, Evan, Mossing, Daniel, Mu, Tong, Murati, Mira, Murk, Oleg, Mély, David, Nair, Ashvin, Nakano, Reiichiro, Nayak, Rajeev, Neelakantan, Arvind, Ngo, Richard, Noh, Hyeonwoo, Ouyang, Long, O'Keefe, Cullen, Pachocki, Jakub, Paino, Alex, Palermo, Joe, Pantuliano, Ashley, Parascandolo, Giambattista, Parish, Joel, Parparita, Emy, Passos, Alex, Pavlov, Mikhail, Peng, Andrew, Perelman, Adam, Peres, Filipe de Avila Belbute, Petrov, Michael, Pinto, Henrique Ponde de Oliveira, Michael, Pokorny, Pokrass, Michelle, Pong, Vitchyr H., Powell, Tolly, Power, Alethea, Power, Boris, Proehl, Elizabeth, Puri, Raul, Radford, Alec, Rae, Jack, Ramesh, Aditya, Raymond, Cameron, Real, Francis, Rimbach, Kendra, Ross, Carl, Rotsted, Bob, Roussez, Henri, Ryder, Nick, Saltarelli, Mario, Sanders, Ted, Santurkar, Shibani, Sastry, Girish, Schmidt, Heather, Schnurr, David, Schulman, John, Selsam, Daniel, Sheppard, Kyla, Sherbakov, Toki, Shieh, Jessica, Shoker, Sarah, Shyam, Pranav, Sidor, Szymon, Sigler, Eric, Simens, Maddie, Sitkin, Jordan, Slama, Katarina, Sohl, Ian, Sokolowsky, Benjamin, Song, Yang, Staudacher, Natalie, Such, Felipe Petroski, Summers, Natalie, Sutskever, Ilya, Tang, Jie, Tezak, Nikolas, Thompson, Madeleine B., Tillet, Phil, Tootoonchian, Amin, Tseng, Elizabeth, Tuggle, Preston, Turley, Nick, Tworek, Jerry, Uribe, Juan Felipe Cerón, Vallone, Andrea, Vijayvergiya, Arun, Voss, Chelsea, Wainwright, Carroll, Wang, Justin Jay, Wang, Alvin, Wang, Ben, Ward, Jonathan, Wei, Jason, Weinmann, CJ, Welihinda, Akila, Welinder, Peter, Weng, Jiayi, Weng, Lilian, Wiethoff, Matt, Willner, Dave, Winter, Clemens, Wolrich, Samuel, Wong, Hannah, Workman, Lauren, Wu, Sherwin, Wu, Jeff, Wu, Michael, Xiao, Kai, Xu, Tao, Yoo, Sarah, Yu, Kevin, Yuan, Qiming, Zaremba, Wojciech, Zellers, Rowan, Zhang, Chong, Zhang, Marvin, Zhao, Shengjia, Zheng, Tianhao, Zhuang, Juntang, Zhuk, William, and Zoph, Barret

Subjects
Computer Science - Computation and Language, Computer Science - Artificial Intelligence
Abstract

We report the development of GPT-4, a large-scale, multimodal model which can accept image and text inputs and produce text outputs. While less capable than humans in many real-world scenarios, GPT-4 exhibits human-level performance on various professional and academic benchmarks, including passing a simulated bar exam with a score around the top 10% of test takers. GPT-4 is a Transformer-based model pre-trained to predict the next token in a document. The post-training alignment process results in improved performance on measures of factuality and adherence to desired behavior. A core component of this project was developing infrastructure and optimization methods that behave predictably across a wide range of scales. This allowed us to accurately predict some aspects of GPT-4's performance based on models trained with no more than 1/1,000th the compute of GPT-4., Comment: 100 pages; updated authors list; fixed author names and added citation

Published
2023

5. BLOOM: A 176B-Parameter Open-Access Multilingual Language Model

Author

Workshop, BigScience, Scao, Teven Le, Fan, Angela, Akiki, Christopher, Pavlick, Ellie, Ilić, Suzana, Hesslow, Daniel, Castagné, Roman, Luccioni, Alexandra Sasha, Yvon, François, Gallé, Matthias, Tow, Jonathan, Rush, Alexander M., Biderman, Stella, Webson, Albert, Ammanamanchi, Pawan Sasanka, Wang, Thomas, Sagot, Benoît, Muennighoff, Niklas, del Moral, Albert Villanova, Ruwase, Olatunji, Bawden, Rachel, Bekman, Stas, McMillan-Major, Angelina, Beltagy, Iz, Nguyen, Huu, Saulnier, Lucile, Tan, Samson, Suarez, Pedro Ortiz, Sanh, Victor, Laurençon, Hugo, Jernite, Yacine, Launay, Julien, Mitchell, Margaret, Raffel, Colin, Gokaslan, Aaron, Simhi, Adi, Soroa, Aitor, Aji, Alham Fikri, Alfassy, Amit, Rogers, Anna, Nitzav, Ariel Kreisberg, Xu, Canwen, Mou, Chenghao, Emezue, Chris, Klamm, Christopher, Leong, Colin, van Strien, Daniel, Adelani, David Ifeoluwa, Radev, Dragomir, Ponferrada, Eduardo González, Levkovizh, Efrat, Kim, Ethan, Natan, Eyal Bar, De Toni, Francesco, Dupont, Gérard, Kruszewski, Germán, Pistilli, Giada, Elsahar, Hady, Benyamina, Hamza, Tran, Hieu, Yu, Ian, Abdulmumin, Idris, Johnson, Isaac, Gonzalez-Dios, Itziar, de la Rosa, Javier, Chim, Jenny, Dodge, Jesse, Zhu, Jian, Chang, Jonathan, Frohberg, Jörg, Tobing, Joseph, Bhattacharjee, Joydeep, Almubarak, Khalid, Chen, Kimbo, Lo, Kyle, Von Werra, Leandro, Weber, Leon, Phan, Long, allal, Loubna Ben, Tanguy, Ludovic, Dey, Manan, Muñoz, Manuel Romero, Masoud, Maraim, Grandury, María, Šaško, Mario, Huang, Max, Coavoux, Maximin, Singh, Mayank, Jiang, Mike Tian-Jian, Vu, Minh Chien, Jauhar, Mohammad A., Ghaleb, Mustafa, Subramani, Nishant, Kassner, Nora, Khamis, Nurulaqilla, Nguyen, Olivier, Espejel, Omar, de Gibert, Ona, Villegas, Paulo, Henderson, Peter, Colombo, Pierre, Amuok, Priscilla, Lhoest, Quentin, Harliman, Rheza, Bommasani, Rishi, López, Roberto Luis, Ribeiro, Rui, Osei, Salomey, Pyysalo, Sampo, Nagel, Sebastian, Bose, Shamik, Muhammad, Shamsuddeen Hassan, Sharma, Shanya, Longpre, Shayne, Nikpoor, Somaieh, Silberberg, Stanislav, Pai, Suhas, Zink, Sydney, Torrent, Tiago Timponi, Schick, Timo, Thrush, Tristan, Danchev, Valentin, Nikoulina, Vassilina, Laippala, Veronika, Lepercq, Violette, Prabhu, Vrinda, Alyafeai, Zaid, Talat, Zeerak, Raja, Arun, Heinzerling, Benjamin, Si, Chenglei, Taşar, Davut Emre, Salesky, Elizabeth, Mielke, Sabrina J., Lee, Wilson Y., Sharma, Abheesht, Santilli, Andrea, Chaffin, Antoine, Stiegler, Arnaud, Datta, Debajyoti, Szczechla, Eliza, Chhablani, Gunjan, Wang, Han, Pandey, Harshit, Strobelt, Hendrik, Fries, Jason Alan, Rozen, Jos, Gao, Leo, Sutawika, Lintang, Bari, M Saiful, Al-shaibani, Maged S., Manica, Matteo, Nayak, Nihal, Teehan, Ryan, Albanie, Samuel, Shen, Sheng, Ben-David, Srulik, Bach, Stephen H., Kim, Taewoon, Bers, Tali, Fevry, Thibault, Neeraj, Trishala, Thakker, Urmish, Raunak, Vikas, Tang, Xiangru, Yong, Zheng-Xin, Sun, Zhiqing, Brody, Shaked, Uri, Yallow, Tojarieh, Hadar, Roberts, Adam, Chung, Hyung Won, Tae, Jaesung, Phang, Jason, Press, Ofir, Li, Conglong, Narayanan, Deepak, Bourfoune, Hatim, Casper, Jared, Rasley, Jeff, Ryabinin, Max, Mishra, Mayank, Zhang, Minjia, Shoeybi, Mohammad, Peyrounette, Myriam, Patry, Nicolas, Tazi, Nouamane, Sanseviero, Omar, von Platen, Patrick, Cornette, Pierre, Lavallée, Pierre François, Lacroix, Rémi, Rajbhandari, Samyam, Gandhi, Sanchit, Smith, Shaden, Requena, Stéphane, Patil, Suraj, Dettmers, Tim, Baruwa, Ahmed, Singh, Amanpreet, Cheveleva, Anastasia, Ligozat, Anne-Laure, Subramonian, Arjun, Névéol, Aurélie, Lovering, Charles, Garrette, Dan, Tunuguntla, Deepak, Reiter, Ehud, Taktasheva, Ekaterina, Voloshina, Ekaterina, Bogdanov, Eli, Winata, Genta Indra, Schoelkopf, Hailey, Kalo, Jan-Christoph, Novikova, Jekaterina, Forde, Jessica Zosa, Clive, Jordan, Kasai, Jungo, Kawamura, Ken, Hazan, Liam, Carpuat, Marine, Clinciu, Miruna, Kim, Najoung, Cheng, Newton, Serikov, Oleg, Antverg, Omer, van der Wal, Oskar, Zhang, Rui, Zhang, Ruochen, Gehrmann, Sebastian, Mirkin, Shachar, Pais, Shani, Shavrina, Tatiana, Scialom, Thomas, Yun, Tian, Limisiewicz, Tomasz, Rieser, Verena, Protasov, Vitaly, Mikhailov, Vladislav, Pruksachatkun, Yada, Belinkov, Yonatan, Bamberger, Zachary, Kasner, Zdeněk, Rueda, Alice, Pestana, Amanda, Feizpour, Amir, Khan, Ammar, Faranak, Amy, Santos, Ana, Hevia, Anthony, Unldreaj, Antigona, Aghagol, Arash, Abdollahi, Arezoo, Tammour, Aycha, HajiHosseini, Azadeh, Behroozi, Bahareh, Ajibade, Benjamin, Saxena, Bharat, Ferrandis, Carlos Muñoz, McDuff, Daniel, Contractor, Danish, Lansky, David, David, Davis, Kiela, Douwe, Nguyen, Duong A., Tan, Edward, Baylor, Emi, Ozoani, Ezinwanne, Mirza, Fatima, Ononiwu, Frankline, Rezanejad, Habib, Jones, Hessie, Bhattacharya, Indrani, Solaiman, Irene, Sedenko, Irina, Nejadgholi, Isar, Passmore, Jesse, Seltzer, Josh, Sanz, Julio Bonis, Dutra, Livia, Samagaio, Mairon, Elbadri, Maraim, Mieskes, Margot, Gerchick, Marissa, Akinlolu, Martha, McKenna, Michael, Qiu, Mike, Ghauri, Muhammed, Burynok, Mykola, Abrar, Nafis, Rajani, Nazneen, Elkott, Nour, Fahmy, Nour, Samuel, Olanrewaju, An, Ran, Kromann, Rasmus, Hao, Ryan, Alizadeh, Samira, Shubber, Sarmad, Wang, Silas, Roy, Sourav, Viguier, Sylvain, Le, Thanh, Oyebade, Tobi, Le, Trieu, Yang, Yoyo, Nguyen, Zach, Kashyap, Abhinav Ramesh, Palasciano, Alfredo, Callahan, Alison, Shukla, Anima, Miranda-Escalada, Antonio, Singh, Ayush, Beilharz, Benjamin, Wang, Bo, Brito, Caio, Zhou, Chenxi, Jain, Chirag, Xu, Chuxin, Fourrier, Clémentine, Periñán, Daniel León, Molano, Daniel, Yu, Dian, Manjavacas, Enrique, Barth, Fabio, Fuhrimann, Florian, Altay, Gabriel, Bayrak, Giyaseddin, Burns, Gully, Vrabec, Helena U., Bello, Imane, Dash, Ishani, Kang, Jihyun, Giorgi, John, Golde, Jonas, Posada, Jose David, Sivaraman, Karthik Rangasai, Bulchandani, Lokesh, Liu, Lu, Shinzato, Luisa, de Bykhovetz, Madeleine Hahn, Takeuchi, Maiko, Pàmies, Marc, Castillo, Maria A, Nezhurina, Marianna, Sänger, Mario, Samwald, Matthias, Cullan, Michael, Weinberg, Michael, De Wolf, Michiel, Mihaljcic, Mina, Liu, Minna, Freidank, Moritz, Kang, Myungsun, Seelam, Natasha, Dahlberg, Nathan, Broad, Nicholas Michio, Muellner, Nikolaus, Fung, Pascale, Haller, Patrick, Chandrasekhar, Ramya, Eisenberg, Renata, Martin, Robert, Canalli, Rodrigo, Su, Rosaline, Su, Ruisi, Cahyawijaya, Samuel, Garda, Samuele, Deshmukh, Shlok S, Mishra, Shubhanshu, Kiblawi, Sid, Ott, Simon, Sang-aroonsiri, Sinee, Kumar, Srishti, Schweter, Stefan, Bharati, Sushil, Laud, Tanmay, Gigant, Théo, Kainuma, Tomoya, Kusa, Wojciech, Labrak, Yanis, Bajaj, Yash Shailesh, Venkatraman, Yash, Xu, Yifan, Xu, Yingxin, Xu, Yu, Tan, Zhe, Xie, Zhongli, Ye, Zifan, Bras, Mathilde, Belkada, Younes, and Wolf, Thomas

Subjects
Computer Science - Computation and Language
Abstract

Large language models (LLMs) have been shown to be able to perform new tasks based on a few demonstrations or natural language instructions. While these capabilities have led to widespread adoption, most LLMs are developed by resource-rich organizations and are frequently kept from the public. As a step towards democratizing this powerful technology, we present BLOOM, a 176B-parameter open-access language model designed and built thanks to a collaboration of hundreds of researchers. BLOOM is a decoder-only Transformer language model that was trained on the ROOTS corpus, a dataset comprising hundreds of sources in 46 natural and 13 programming languages (59 in total). We find that BLOOM achieves competitive performance on a wide variety of benchmarks, with stronger results after undergoing multitask prompted finetuning. To facilitate future research and applications using LLMs, we publicly release our models and code under the Responsible AI License.

Published
2022

6. Scaling Laws for Reward Model Overoptimization

Author

Gao, Leo, Schulman, John, and Hilton, Jacob

Subjects
Computer Science - Machine Learning, Statistics - Machine Learning
Abstract

In reinforcement learning from human feedback, it is common to optimize against a reward model trained to predict human preferences. Because the reward model is an imperfect proxy, optimizing its value too much can hinder ground truth performance, in accordance with Goodhart's law. This effect has been frequently observed, but not carefully measured due to the expense of collecting human preference data. In this work, we use a synthetic setup in which a fixed "gold-standard" reward model plays the role of humans, providing labels used to train a proxy reward model. We study how the gold reward model score changes as we optimize against the proxy reward model using either reinforcement learning or best-of-$n$ sampling. We find that this relationship follows a different functional form depending on the method of optimization, and that in both cases its coefficients scale smoothly with the number of reward model parameters. We also study the effect on this relationship of the size of the reward model dataset, the number of reward model and policy parameters, and the coefficient of the KL penalty added to the reward in the reinforcement learning setup. We explore the implications of these empirical results for theoretical considerations in AI alignment.

Published
2022

7. EleutherAI: Going Beyond 'Open Science' to 'Science in the Open'

Author

Phang, Jason, Bradley, Herbie, Gao, Leo, Castricato, Louis, and Biderman, Stella

Subjects
Computer Science - Computation and Language
Abstract

Over the past two years, EleutherAI has established itself as a radically novel initiative aimed at both promoting open-source research and conducting research in a transparent, openly accessible and collaborative manner. EleutherAI's approach to research goes beyond transparency: by doing research entirely in public, anyone in the world can observe and contribute at every stage. Our work has been received positively and has resulted in several high-impact projects in Natural Language Processing and other fields. In this paper, we describe our experience doing public-facing machine learning research, the benefits we believe this approach brings, and the pitfalls we have encountered.

Published
2022

8. Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models

Author

Srivastava, Aarohi, Rastogi, Abhinav, Rao, Abhishek, Shoeb, Abu Awal Md, Abid, Abubakar, Fisch, Adam, Brown, Adam R., Santoro, Adam, Gupta, Aditya, Garriga-Alonso, Adrià, Kluska, Agnieszka, Lewkowycz, Aitor, Agarwal, Akshat, Power, Alethea, Ray, Alex, Warstadt, Alex, Kocurek, Alexander W., Safaya, Ali, Tazarv, Ali, Xiang, Alice, Parrish, Alicia, Nie, Allen, Hussain, Aman, Askell, Amanda, Dsouza, Amanda, Slone, Ambrose, Rahane, Ameet, Iyer, Anantharaman S., Andreassen, Anders, Madotto, Andrea, Santilli, Andrea, Stuhlmüller, Andreas, Dai, Andrew, La, Andrew, Lampinen, Andrew, Zou, Andy, Jiang, Angela, Chen, Angelica, Vuong, Anh, Gupta, Animesh, Gottardi, Anna, Norelli, Antonio, Venkatesh, Anu, Gholamidavoodi, Arash, Tabassum, Arfa, Menezes, Arul, Kirubarajan, Arun, Mullokandov, Asher, Sabharwal, Ashish, Herrick, Austin, Efrat, Avia, Erdem, Aykut, Karakaş, Ayla, Roberts, B. Ryan, Loe, Bao Sheng, Zoph, Barret, Bojanowski, Bartłomiej, Özyurt, Batuhan, Hedayatnia, Behnam, Neyshabur, Behnam, Inden, Benjamin, Stein, Benno, Ekmekci, Berk, Lin, Bill Yuchen, Howald, Blake, Orinion, Bryan, Diao, Cameron, Dour, Cameron, Stinson, Catherine, Argueta, Cedrick, Ramírez, César Ferri, Singh, Chandan, Rathkopf, Charles, Meng, Chenlin, Baral, Chitta, Wu, Chiyu, Callison-Burch, Chris, Waites, Chris, Voigt, Christian, Manning, Christopher D., Potts, Christopher, Ramirez, Cindy, Rivera, Clara E., Siro, Clemencia, Raffel, Colin, Ashcraft, Courtney, Garbacea, Cristina, Sileo, Damien, Garrette, Dan, Hendrycks, Dan, Kilman, Dan, Roth, Dan, Freeman, Daniel, Khashabi, Daniel, Levy, Daniel, González, Daniel Moseguí, Perszyk, Danielle, Hernandez, Danny, Chen, Danqi, Ippolito, Daphne, Gilboa, Dar, Dohan, David, Drakard, David, Jurgens, David, Datta, Debajyoti, Ganguli, Deep, Emelin, Denis, Kleyko, Denis, Yuret, Deniz, Chen, Derek, Tam, Derek, Hupkes, Dieuwke, Misra, Diganta, Buzan, Dilyar, Mollo, Dimitri Coelho, Yang, Diyi, Lee, Dong-Ho, Schrader, Dylan, Shutova, Ekaterina, Cubuk, Ekin Dogus, Segal, Elad, Hagerman, Eleanor, Barnes, Elizabeth, Donoway, Elizabeth, Pavlick, Ellie, Rodola, Emanuele, Lam, Emma, Chu, Eric, Tang, Eric, Erdem, Erkut, Chang, Ernie, Chi, Ethan A., Dyer, Ethan, Jerzak, Ethan, Kim, Ethan, Manyasi, Eunice Engefu, Zheltonozhskii, Evgenii, Xia, Fanyue, Siar, Fatemeh, Martínez-Plumed, Fernando, Happé, Francesca, Chollet, Francois, Rong, Frieda, Mishra, Gaurav, Winata, Genta Indra, de Melo, Gerard, Kruszewski, Germán, Parascandolo, Giambattista, Mariani, Giorgio, Wang, Gloria, Jaimovitch-López, Gonzalo, Betz, Gregor, Gur-Ari, Guy, Galijasevic, Hana, Kim, Hannah, Rashkin, Hannah, Hajishirzi, Hannaneh, Mehta, Harsh, Bogar, Hayden, Shevlin, Henry, Schütze, Hinrich, Yakura, Hiromu, Zhang, Hongming, Wong, Hugh Mee, Ng, Ian, Noble, Isaac, Jumelet, Jaap, Geissinger, Jack, Kernion, Jackson, Hilton, Jacob, Lee, Jaehoon, Fisac, Jaime Fernández, Simon, James B., Koppel, James, Zheng, James, Zou, James, Kocoń, Jan, Thompson, Jana, Wingfield, Janelle, Kaplan, Jared, Radom, Jarema, Sohl-Dickstein, Jascha, Phang, Jason, Wei, Jason, Yosinski, Jason, Novikova, Jekaterina, Bosscher, Jelle, Marsh, Jennifer, Kim, Jeremy, Taal, Jeroen, Engel, Jesse, Alabi, Jesujoba, Xu, Jiacheng, Song, Jiaming, Tang, Jillian, Waweru, Joan, Burden, John, Miller, John, Balis, John U., Batchelder, Jonathan, Berant, Jonathan, Frohberg, Jörg, Rozen, Jos, Hernandez-Orallo, Jose, Boudeman, Joseph, Guerr, Joseph, Jones, Joseph, Tenenbaum, Joshua B., Rule, Joshua S., Chua, Joyce, Kanclerz, Kamil, Livescu, Karen, Krauth, Karl, Gopalakrishnan, Karthik, Ignatyeva, Katerina, Markert, Katja, Dhole, Kaustubh D., Gimpel, Kevin, Omondi, Kevin, Mathewson, Kory, Chiafullo, Kristen, Shkaruta, Ksenia, Shridhar, Kumar, McDonell, Kyle, Richardson, Kyle, Reynolds, Laria, Gao, Leo, Zhang, Li, Dugan, Liam, Qin, Lianhui, Contreras-Ochando, Lidia, Morency, Louis-Philippe, Moschella, Luca, Lam, Lucas, Noble, Lucy, Schmidt, Ludwig, He, Luheng, Colón, Luis Oliveros, Metz, Luke, Şenel, Lütfi Kerem, Bosma, Maarten, Sap, Maarten, ter Hoeve, Maartje, Farooqi, Maheen, Faruqui, Manaal, Mazeika, Mantas, Baturan, Marco, Marelli, Marco, Maru, Marco, Quintana, Maria Jose Ramírez, Tolkiehn, Marie, Giulianelli, Mario, Lewis, Martha, Potthast, Martin, Leavitt, Matthew L., Hagen, Matthias, Schubert, Mátyás, Baitemirova, Medina Orduna, Arnaud, Melody, McElrath, Melvin, Yee, Michael A., Cohen, Michael, Gu, Michael, Ivanitskiy, Michael, Starritt, Michael, Strube, Michael, Swędrowski, Michał, Bevilacqua, Michele, Yasunaga, Michihiro, Kale, Mihir, Cain, Mike, Xu, Mimee, Suzgun, Mirac, Walker, Mitch, Tiwari, Mo, Bansal, Mohit, Aminnaseri, Moin, Geva, Mor, Gheini, Mozhdeh, T, Mukund Varma, Peng, Nanyun, Chi, Nathan A., Lee, Nayeon, Krakover, Neta Gur-Ari, Cameron, Nicholas, Roberts, Nicholas, Doiron, Nick, Martinez, Nicole, Nangia, Nikita, Deckers, Niklas, Muennighoff, Niklas, Keskar, Nitish Shirish, Iyer, Niveditha S., Constant, Noah, Fiedel, Noah, Wen, Nuan, Zhang, Oliver, Agha, Omar, Elbaghdadi, Omar, Levy, Omer, Evans, Owain, Casares, Pablo Antonio Moreno, Doshi, Parth, Fung, Pascale, Liang, Paul Pu, Vicol, Paul, Alipoormolabashi, Pegah, Liao, Peiyuan, Liang, Percy, Chang, Peter, Eckersley, Peter, Htut, Phu Mon, Hwang, Pinyu, Miłkowski, Piotr, Patil, Piyush, Pezeshkpour, Pouya, Oli, Priti, Mei, Qiaozhu, Lyu, Qing, Chen, Qinlang, Banjade, Rabin, Rudolph, Rachel Etta, Gabriel, Raefer, Habacker, Rahel, Risco, Ramon, Millière, Raphaël, Garg, Rhythm, Barnes, Richard, Saurous, Rif A., Arakawa, Riku, Raymaekers, Robbe, Frank, Robert, Sikand, Rohan, Novak, Roman, Sitelew, Roman, LeBras, Ronan, Liu, Rosanne, Jacobs, Rowan, Zhang, Rui, Salakhutdinov, Ruslan, Chi, Ryan, Lee, Ryan, Stovall, Ryan, Teehan, Ryan, Yang, Rylan, Singh, Sahib, Mohammad, Saif M., Anand, Sajant, Dillavou, Sam, Shleifer, Sam, Wiseman, Sam, Gruetter, Samuel, Bowman, Samuel R., Schoenholz, Samuel S., Han, Sanghyun, Kwatra, Sanjeev, Rous, Sarah A., Ghazarian, Sarik, Ghosh, Sayan, Casey, Sean, Bischoff, Sebastian, Gehrmann, Sebastian, Schuster, Sebastian, Sadeghi, Sepideh, Hamdan, Shadi, Zhou, Sharon, Srivastava, Shashank, Shi, Sherry, Singh, Shikhar, Asaadi, Shima, Gu, Shixiang Shane, Pachchigar, Shubh, Toshniwal, Shubham, Upadhyay, Shyam, Shyamolima, Debnath, Shakeri, Siamak, Thormeyer, Simon, Melzi, Simone, Reddy, Siva, Makini, Sneha Priscilla, Lee, Soo-Hwan, Torene, Spencer, Hatwar, Sriharsha, Dehaene, Stanislas, Divic, Stefan, Ermon, Stefano, Biderman, Stella, Lin, Stephanie, Prasad, Stephen, Piantadosi, Steven T., Shieber, Stuart M., Misherghi, Summer, Kiritchenko, Svetlana, Mishra, Swaroop, Linzen, Tal, Schuster, Tal, Li, Tao, Yu, Tao, Ali, Tariq, Hashimoto, Tatsu, Wu, Te-Lin, Desbordes, Théo, Rothschild, Theodore, Phan, Thomas, Wang, Tianle, Nkinyili, Tiberius, Schick, Timo, Kornev, Timofei, Tunduny, Titus, Gerstenberg, Tobias, Chang, Trenton, Neeraj, Trishala, Khot, Tushar, Shultz, Tyler, Shaham, Uri, Misra, Vedant, Demberg, Vera, Nyamai, Victoria, Raunak, Vikas, Ramasesh, Vinay, Prabhu, Vinay Uday, Padmakumar, Vishakh, Srikumar, Vivek, Fedus, William, Saunders, William, Zhang, William, Vossen, Wout, Ren, Xiang, Tong, Xiaoyu, Zhao, Xinran, Wu, Xinyi, Shen, Xudong, Yaghoobzadeh, Yadollah, Lakretz, Yair, Song, Yangqiu, Bahri, Yasaman, Choi, Yejin, Yang, Yichi, Hao, Yiding, Chen, Yifu, Belinkov, Yonatan, Hou, Yu, Hou, Yufang, Bai, Yuntao, Seid, Zachary, Zhao, Zhuoye, Wang, Zijian, Wang, Zijie J., Wang, Zirui, and Wu, Ziyi

Subjects
Computer Science - Computation and Language, Computer Science - Artificial Intelligence, Computer Science - Computers and Society, Computer Science - Machine Learning, Statistics - Machine Learning
Abstract

Language models demonstrate both quantitative improvement and new qualitative capabilities with increasing scale. Despite their potentially transformative impact, these new capabilities are as yet poorly characterized. In order to inform future research, prepare for disruptive new model capabilities, and ameliorate socially harmful effects, it is vital that we understand the present and near-future capabilities and limitations of language models. To address this challenge, we introduce the Beyond the Imitation Game benchmark (BIG-bench). BIG-bench currently consists of 204 tasks, contributed by 450 authors across 132 institutions. Task topics are diverse, drawing problems from linguistics, childhood development, math, common-sense reasoning, biology, physics, social bias, software development, and beyond. BIG-bench focuses on tasks that are believed to be beyond the capabilities of current language models. We evaluate the behavior of OpenAI's GPT models, Google-internal dense transformer architectures, and Switch-style sparse transformers on BIG-bench, across model sizes spanning millions to hundreds of billions of parameters. In addition, a team of human expert raters performed all tasks in order to provide a strong baseline. Findings include: model performance and calibration both improve with scale, but are poor in absolute terms (and when compared with rater performance); performance is remarkably similar across model classes, though with benefits from sparsity; tasks that improve gradually and predictably commonly involve a large knowledge or memorization component, whereas tasks that exhibit "breakthrough" behavior at a critical scale often involve multiple steps or components, or brittle metrics; social bias typically increases with scale in settings with ambiguous context, but this can be improved with prompting., Comment: 27 pages, 17 figures + references and appendices, repo: https://github.com/google/BIG-bench

Published
2022

9. GPT-NeoX-20B: An Open-Source Autoregressive Language Model

Author

Black, Sid, Biderman, Stella, Hallahan, Eric, Anthony, Quentin, Gao, Leo, Golding, Laurence, He, Horace, Leahy, Connor, McDonell, Kyle, Phang, Jason, Pieler, Michael, Prashanth, USVSN Sai, Purohit, Shivanshu, Reynolds, Laria, Tow, Jonathan, Wang, Ben, and Weinbach, Samuel

Subjects
Computer Science - Computation and Language
Abstract

We introduce GPT-NeoX-20B, a 20 billion parameter autoregressive language model trained on the Pile, whose weights will be made freely and openly available to the public through a permissive license. It is, to the best of our knowledge, the largest dense autoregressive model that has publicly available weights at the time of submission. In this work, we describe \model{}'s architecture and training and evaluate its performance on a range of language-understanding, mathematics, and knowledge-based tasks. We find that GPT-NeoX-20B is a particularly powerful few-shot reasoner and gains far more in performance when evaluated five-shot than similarly sized GPT-3 and FairSeq models. We open-source the training and evaluation code, as well as the model weights, at https://github.com/EleutherAI/gpt-neox., Comment: To appear in the Proceedings of the ACL Workshop on Challenges & Perspectives in Creating Large Language Models

Published
2022

10. Datasheet for the Pile

Author

Biderman, Stella, Bicheno, Kieran, and Gao, Leo

Subjects
Computer Science - Computation and Language
Abstract

This datasheet describes the Pile, a 825 GiB dataset of human-authored text compiled by EleutherAI for use in large-scale language modeling. The Pile is comprised of 22 different text sources, ranging from original scrapes done for this project, to text data made available by the data owners, to third-party scrapes available online., Comment: Accompanies "The Pile: An 800GB Dataset of Diverse Text for Language Modeling" arXiv:2101.00027

Published
2022

11. Multitask Prompted Training Enables Zero-Shot Task Generalization

Author

Sanh, Victor, Webson, Albert, Raffel, Colin, Bach, Stephen H., Sutawika, Lintang, Alyafeai, Zaid, Chaffin, Antoine, Stiegler, Arnaud, Scao, Teven Le, Raja, Arun, Dey, Manan, Bari, M Saiful, Xu, Canwen, Thakker, Urmish, Sharma, Shanya Sharma, Szczechla, Eliza, Kim, Taewoon, Chhablani, Gunjan, Nayak, Nihal, Datta, Debajyoti, Chang, Jonathan, Jiang, Mike Tian-Jian, Wang, Han, Manica, Matteo, Shen, Sheng, Yong, Zheng Xin, Pandey, Harshit, Bawden, Rachel, Wang, Thomas, Neeraj, Trishala, Rozen, Jos, Sharma, Abheesht, Santilli, Andrea, Fevry, Thibault, Fries, Jason Alan, Teehan, Ryan, Bers, Tali, Biderman, Stella, Gao, Leo, Wolf, Thomas, and Rush, Alexander M.

Subjects
Computer Science - Machine Learning, Computer Science - Computation and Language
Abstract

Large language models have recently been shown to attain reasonable zero-shot generalization on a diverse set of tasks (Brown et al., 2020). It has been hypothesized that this is a consequence of implicit multitask learning in language models' pretraining (Radford et al., 2019). Can zero-shot generalization instead be directly induced by explicit multitask learning? To test this question at scale, we develop a system for easily mapping any natural language tasks into a human-readable prompted form. We convert a large set of supervised datasets, each with multiple prompts with diverse wording. These prompted datasets allow for benchmarking the ability of a model to perform completely held-out tasks. We fine-tune a pretrained encoder-decoder model (Raffel et al., 2020; Lester et al., 2021) on this multitask mixture covering a wide variety of tasks. The model attains strong zero-shot performance on several standard datasets, often outperforming models up to 16x its size. Further, our approach attains strong performance on a subset of tasks from the BIG-bench benchmark, outperforming models up to 6x its size. All trained models are available at https://github.com/bigscience-workshop/t-zero and all prompts are available at https://github.com/bigscience-workshop/promptsource., Comment: ICLR 2022 Spotlight (with extended discussion)

Published
2021

12. Cut the CARP: Fishing for zero-shot story evaluation

Author

Matiana, Shahbuland, Smith, JR, Teehan, Ryan, Castricato, Louis, Biderman, Stella, Gao, Leo, and Frazier, Spencer

Subjects
Computer Science - Computation and Language
Abstract

Recent advances in large-scale language models (Raffel et al., 2019; Brown et al., 2020) have brought significant qualitative and quantitative improvements in machine-driven text generation. Despite this, generation and evaluation of machine-generated narrative text remains a challenging problem. Objective evaluation of computationally-generated stories may be prohibitively expensive, require meticulously annotated datasets, or may not adequately measure the logical coherence of a generated story's narratological structure. Informed by recent advances in contrastive learning (Radford et al., 2021), we present Contrastive Authoring and Reviewing Pairing (CARP): a scalable, efficient method for performing qualitatively superior, zero-shot evaluation of stories. We show a strong correlation between human evaluation of stories and those of CARP. Model outputs more significantly correlate with corresponding human input than those language-model based methods which utilize finetuning or prompt engineering approaches. We also present and analyze the Story-Critique Dataset, a new corpora composed of 1.3 million aligned story-critique pairs derived from over 80,000 stories. We expect this corpus to be of interest to NLP researchers., Comment: 9 pages, 4 figures

Published
2021

13. An Empirical Exploration in Quality Filtering of Text Data

Author

Gao, Leo

Subjects
Computer Science - Computation and Language
Abstract

While conventional wisdom suggests that more aggressively filtering data from low-quality sources like Common Crawl always monotonically improves the quality of training data, we find that aggressive filtering can in fact lead to a decrease in model quality on a wide array of downstream tasks for a GPT-like language model. We speculate that this is because optimizing sufficiently strongly for a proxy metric harms performance on the true objective, suggesting a need for more robust filtering objectives when attempting to filter more aggressively. We hope this work leads to detailed analysis of the effects of dataset filtering design choices on downstream model performance in future work., Comment: corrected typo in citation

Published
2021

14. The Pile: An 800GB Dataset of Diverse Text for Language Modeling

Author

Gao, Leo, Biderman, Stella, Black, Sid, Golding, Laurence, Hoppe, Travis, Foster, Charles, Phang, Jason, He, Horace, Thite, Anish, Nabeshima, Noa, Presser, Shawn, and Leahy, Connor

Subjects
Computer Science - Computation and Language
Abstract

Recent work has demonstrated that increased training dataset diversity improves general cross-domain knowledge and downstream generalization capability for large-scale language models. With this in mind, we present \textit{the Pile}: an 825 GiB English text corpus targeted at training large-scale language models. The Pile is constructed from 22 diverse high-quality subsets -- both existing and newly constructed -- many of which derive from academic or professional sources. Our evaluation of the untuned performance of GPT-2 and GPT-3 on the Pile shows that these models struggle on many of its components, such as academic writing. Conversely, models trained on the Pile improve significantly over both Raw CC and CC-100 on all components of the Pile, while improving performance on downstream evaluations. Through an in-depth exploratory analysis, we document potentially concerning aspects of the data for prospective users. We make publicly available the code used in its construction.

Published
2020

15. Collaborative Storytelling with Large-scale Neural Language Models

Author

Nichols, Eric, Gao, Leo, and Gomez, Randy

Subjects
Computer Science - Computation and Language, Computer Science - Artificial Intelligence, Computer Science - Machine Learning, Computer Science - Neural and Evolutionary Computing
Abstract

Storytelling plays a central role in human socializing and entertainment. However, much of the research on automatic storytelling generation assumes that stories will be generated by an agent without any human interaction. In this paper, we introduce the task of collaborative storytelling, where an artificial intelligence agent and a person collaborate to create a unique story by taking turns adding to it. We present a collaborative storytelling system which works with a human storyteller to create a story by generating new utterances based on the story so far. We constructed the storytelling system by tuning a publicly-available large scale language model on a dataset of writing prompts and their accompanying fictional works. We identify generating sufficiently human-like utterances to be an important technical issue and propose a sample-and-rank approach to improve utterance quality. Quantitative evaluation shows that our approach outperforms a baseline, and we present qualitative evaluation of our system's capabilities., Comment: To appear in Proceedings of the 13th Annual ACM SIGGRAPH Conference on Motion, Interaction and Games (MIG 2020)

Published
2020

Catalog

Books, media, physical & digital resources
See catalog results