Beating Full Fine-Tuning with Just 0.2% of Parameters

Table of Links

Abstract and 1. Introduction

2. Background

   2.1 Mixture-of-Experts

   2.2 Adapters

3. Mixture-of-Adaptations

   3.1 Routing Policy

   3.2 Consistency regularization

   3.3 Adaptation module merging and 3.4 Adaptation module sharing

   3.5 Connection to Bayesian Neural Networks and Model Ensembling

4. Experiments

   4.1 Experimental Setup

   4.2 Key Results

   4.3 Ablation Study

5. Related Work

6. Conclusions

7. Limitations

8. Acknowledgment and References

Appendix

A. Few-shot NLU Datasets B. Ablation Study C. Detailed Results on NLU Tasks D. Hyper-parameter

5 Related Work

Parameter-efficient fine-tuning of PLMs. Recent works on parameter-efficient fine-tuning (PEFT) can be roughly categorized into two

categories: (1) tuning a subset of existing parameters including head fine-tuning (Lee et al., 2019), bias term tuning (Zaken et al., 2021), (2) tuning newly-introduced parameters including adapters (Houlsby et al., 2019; Pfeiffer et al., 2020), prompt-tuning (Lester et al., 2021), prefixtuning (Li and Liang, 2021) and low-rank adaptation (Hu et al., 2021). As opposed to prior works operating on a single adaptation module, AdaMix introduces a mixture of adaptation modules with stochastic routing during training and adaptation module merging during inference to keep the same computational cost as with a single module. Further, AdaMix can be used on top of any PEFT method to further boost its performance.

Mixture-of-Expert (MoE). Shazeer et al., 2017 introduced the MoE model with a single gating network with T op-k routing and load balancing across experts. Fedus et al., 2021 propose initialization and training schemes for T op-1 routing. Zuo et al., 2021 propose consistency regularization for random routing; Yang et al., 2021 propose k T op-1 routing with expert-prototypes, and Roller et al., 2021; Lewis et al., 2021 address other load balancing issues. All the above works study sparse MoE with pre-training the entire model from scratch. In contrast, we study parameter-efficient adaptation of pre-trained language models by tuning only a very small number of sparse adapter parameters.

Averaging model weights. Recent explorations (Szegedy et al., 2016; Matena and Raffel, 2021; Wortsman et al., 2022; Izmailov et al., 2018) study model aggregation by averaging all the model weights. (Matena and Raffel, 2021) propose to merge pre-trained language models which are fine-tuned on various text classification tasks. (Wortsman et al., 2022) explores averaging model weights from various independent runs on the same task with different hyper-parameter configurations. In contrast to the above works on full model finetuning, we focus on parameter-efficient fine-tuning. We explore weight averaging for merging weights of adaptation modules consisting of small tunable parameters that are updated during model tuning while keeping the large model parameters fixed.

6 Conclusions

We develop a new framework AdaMix for parameter-efficient fine-tuning (PEFT) of large pretrained language models (PLM). AdaMix leverages a mixture of adaptation modules to improve downstream task performance without increasing the computational cost (e.g., FLOPs, parameters) of the underlying adaptation method. We demonstrate AdaMix to work with and improve over different PEFT methods like adapters and low rank decompositions across NLU and NLG tasks.

By tuning only 0.1 − 0.2% of PLM parameters, AdaMix outperforms full model fine-tuning that updates all the model parameters as well as other state-of-the-art PEFT methods.

7 Limitations

The proposed AdaMix method is somewhat compute-intensive as it involves fine-tuning largescale language models. The training cost of the proposed AdaMix is higher than standard PEFT methods since the training procedure involves multiple copies of adapters. Based on our empirical observation, the number of training iterations for AdaMix is usually between 1∼2 times the training for standard PEFT methods. This imposes negative impact on carbon footprint from training the described models.

AdaMix is orthogonal to most of the existing parameter-efficient fine-tuning (PEFT) studies and is able to potentially improve the performance of any PEFT method. In this work, we explore two representative PEFT methods like adapter and LoRA but we did not experiment with other combinations like prompt-tuning and prefix-tuning. We leave those studies to future work.

8 Acknowledgment

The authors would like to thank the anonymous referees for their valuable comments and helpful suggestions and would like to thank Guoqing Zheng and Ruya Kang for their insightful comments on the project. This work is supported in part by the US National Science Foundation under grants NSFIIS 1747614 and NSF-IIS-2141037. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the National Science Foundation.

References

Armen Aghajanyan, Sonal Gupta, and Luke Zettlemoyer. 2021. Intrinsic dimensionality explains the effectiveness of language model fine-tuning. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 7319– 7328, Online. Association for Computational Linguistics.

Roy Bar Haim, Ido Dagan, Bill Dolan, Lisa Ferro, Danilo Giampiccolo, Bernardo Magnini, and Idan Szpektor. 2006. The second PASCAL recognising textual entailment challenge.

Luisa Bentivogli, Peter Clark, Ido Dagan, and Danilo Giampiccolo. 2009. The fifth PASCAL recognizing textual entailment challenge. In TAC.

Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel HerbertVoss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, Chris Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott Gray, Benjamin Chess, Jack Clark, Christopher Berner, Sam McCandlish, Alec Radford, Ilya Sutskever, and Dario Amodei. 2020. Language models are few-shot learners. In Advances in Neural Information Processing Systems, volume 33, pages 1877–1901. Curran Associates, Inc.

Ido Dagan, Oren Glickman, and Bernardo Magnini. 2005. The PASCAL recognising textual entailment challenge. In the First International Conference on Machine Learning Challenges: Evaluating Predictive Uncertainty Visual Object Classification, and Recognizing Textual Entailment.

Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL-HLT 2019, Volume 1 (Long and Short Papers), pages 4171–4186.

William Fedus, Barret Zoph, and Noam Shazeer. 2021. Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity. arXiv preprint arXiv:2101.03961.

Jonathan Frankle, Gintare Karolina Dziugaite, Daniel Roy, and Michael Carbin. 2020. Linear mode connectivity and the lottery ticket hypothesis. In International Conference on Machine Learning, pages 3259–3269. PMLR.

Yarin Gal and Zoubin Ghahramani. 2015. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. CoRR, abs/1506.02142.

Yarin Gal, Riashat Islam, and Zoubin Ghahramani. 2017. Deep Bayesian active learning with image data. In Proceedings of the 34th International Conference on Machine Learning, volume 70 of Proceedings of Machine Learning Research, pages 1183–1192. PMLR.

Tianyu Gao, Adam Fisch, and Danqi Chen. 2021. Making pre-trained language models better few-shot learners. In Association for Computational Linguistics (ACL).

Claire Gardent, Anastasia Shimorina, Shashi Narayan, and Laura Perez-Beltrachini. 2017. The webnlg challenge: Generating text from rdf data. In Proceedings of the 10th International Conference on Natural Language Generation, pages 124–133.

Danilo Giampiccolo, Bernardo Magnini, Ido Dagan, and Bill Dolan. 2007. The third PASCAL recognizing textual entailment challenge. In the ACLPASCAL Workshop on Textual Entailment and Paraphrasing.

Neil Houlsby, Andrei Giurgiu, Stanislaw Jastrzebski, Bruna Morrone, Quentin De Laroussilhe, Andrea Gesmundo, Mona Attariyan, and Sylvain Gelly. 2019. Parameter-efficient transfer learning for nlp. In International Conference on Machine Learning, pages 2790–2799. PMLR.

Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2021. Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685.

Pavel Izmailov, Dmitrii Podoprikhin, Timur Garipov, Dmitry Vetrov, and Andrew Gordon Wilson. 2018. Averaging weights leads to wider optima and better generalization. arXiv preprint arXiv:1803.05407.

Jaejun Lee, Raphael Tang, and Jimmy Lin. 2019. What would elsa do? freezing layers during transformer fine-tuning. arXiv preprint arXiv:1911.03090.

Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, Dehao Chen, Orhan Firat, Yanping Huang, Maxim Krikun, Noam Shazeer, and Zhifeng Chen. 2020. Gshard: Scaling giant models with conditional computation and automatic sharding. arXiv preprint arXiv:2006.16668.

Brian Lester, Rami Al-Rfou, and Noah Constant. 2021. The power of scale for parameter-efficient prompt tuning. CoRR, abs/2104.08691.

Mike Lewis, Shruti Bhosale, Tim Dettmers, Naman Goyal, and Luke Zettlemoyer. 2021. Base layers: Simplifying training of large, sparse models. In ICML.

Xiang Lisa Li and Percy Liang. 2021. Prefixtuning: Optimizing continuous prompts for generation. CoRR, abs/2101.00190.

Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Mandar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized BERT pretraining approach. CoRR, abs/1907.11692.

Yuning Mao, Lambert Mathias, Rui Hou, Amjad Almahairi, Hao Ma, Jiawei Han, Wen-tau Yih, and Madian Khabsa. 2021. Unipelt: A unified framework for parameter-efficient language model tuning. arXiv preprint arXiv:2110.07577.

Michael Matena and Colin Raffel. 2021. Merging models with fisher-weighted averaging. arXiv preprint arXiv:2111.09832.

Linyong Nan, Dragomir Radev, Rui Zhang, Amrit Rau, Abhinand Sivaprasad, Chiachun Hsieh, Xiangru Tang, Aadit Vyas, Neha Verma, Pranav Krishna, et al. 2020. Dart: Open-domain structured data record to text generation. arXiv preprint arXiv:2007.02871.

Behnam Neyshabur, Hanie Sedghi, and Chiyuan Zhang. 2020. What is being transferred in transfer learning? Advances in neural information processing systems, 33:512–523.

Jekaterina Novikova, Ondˇrej Dušek, and Verena Rieser. 2017. The e2e dataset: New challenges for end-toend generation. arXiv preprint arXiv:1706.09254.

Bo Pang and Lillian Lee. 2004. A sentimental education: Sentiment analysis using subjectivity summarization based on minimum cuts.

Ethan Perez, Douwe Kiela, and Kyunghyun Cho. 2021. True few-shot learning with language models. arXiv preprint arXiv:2105.11447.

Jonas Pfeiffer, Aishwarya Kamath, Andreas Rücklé, Kyunghyun Cho, and Iryna Gurevych. 2021. Adapterfusion: Non-destructive task composition for transfer learning. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, pages 487–503.

Jonas Pfeiffer, Andreas Rücklé, Clifton Poth, Aishwarya Kamath, Ivan Vulic, Sebastian Ruder, ´ Kyunghyun Cho, and Iryna Gurevych. 2020. Adapterhub: A framework for adapting transformers. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP 2020): Systems Demonstrations, pages 46– 54, Online. Association for Computational Linguistics.

Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. 2019. Exploring the limits of transfer learning with a unified text-to-text transformer. arXiv preprint arXiv:1910.10683.

Stephen Roller, Sainbayar Sukhbaatar, Arthur D. Szlam, and Jason Weston. 2021. Hash layers for large sparse models. ArXiv, abs/2106.04426.

Thibault Sellam, Steve Yadlowsky, Ian Tenney, Jason Wei, Naomi Saphra, Alexander D’Amour, Tal Linzen, Jasmijn Bastings, Iulia Raluca Turc, Jacob Eisenstein, Dipanjan Das, and Ellie Pavlick. 2022. The multiBERTs: BERT reproductions for robustness analysis. In International Conference on Learning Representations.

Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. 2017. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. arXiv preprint arXiv:1701.06538.

Shaden Smith, Mostofa Patwary, Brandon Norick, Patrick LeGresley, Samyam Rajbhandari, Jared Casper, Zhun Liu, Shrimai Prabhumoye, George Zerveas, Vijay Korthikanti, et al. 2022. Using deepspeed and megatron to train megatron-turing nlg 530b, a large-scale generative language model. arXiv preprint arXiv:2201.11990.

Richard Socher, Alex Perelygin, Jean Wu, Jason Chuang, Christopher D. Manning, Andrew Ng, and Christopher Potts. Recursive deep models for semantic compositionality over a sentiment treebank.

Christian Szegedy, Vincent Vanhoucke, Sergey Ioffe, Jon Shlens, and Zbigniew Wojna. 2016. Rethinking the inception architecture for computer vision. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 2818–2826.

Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. Advances in neural information processing systems, 30.

Alex Wang, Amanpreet Singh, Julian Michael, Felix Hill, Omer Levy, and Samuel R Bowman. 2019. GLUE: A multi-task benchmark and analysis platform for natural language understanding.

Yaqing Wang, Subhabrata Mukherjee, Xiaodong Liu, Jing Gao, Ahmed Hassan Awadallah, and Jianfeng Gao. 2021. List: Lite self-training makes efficient few-shot learners. arXiv preprint arXiv:2110.06274.

Janyce Wiebe, Theresa Wilson, and Claire Cardie. 2005. Annotating expressions of opinions and emotions in language. Language resources and evaluation, 39(2):165–210.

Adina Williams, Nikita Nangia, and Samuel Bowman. 2018. A broad-coverage challenge corpus for sentence understanding through inference.

Mitchell Wortsman, Gabriel Ilharco, Samir Yitzhak Gadre, Rebecca Roelofs, Raphael Gontijo-Lopes, Ari S Morcos, Hongseok Namkoong, Ali Farhadi, Yair Carmon, Simon Kornblith, et al. 2022. Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time. arXiv preprint arXiv:2203.05482.

An Yang, Junyang Lin, Rui Men, Chang Zhou, Le Jiang, Xianyan Jia, Ang Wang, Jie Zhang, Jiamang Wang, Yong Li, et al. 2021. M6-t: Exploring sparse expert models and beyond. arXiv preprint arXiv:2105.15082.

Elad Ben Zaken, Shauli Ravfogel, and Yoav Goldberg. 2021. Bitfit: Simple parameter-efficient fine-tuning for transformer-based masked languagemodels. arXiv preprint arXiv:2106.10199.

Tianyi Zhang, Felix Wu, Arzoo Katiyar, Kilian Q Weinberger, and Yoav Artzi. 2021. Revisiting fewsample BERT fine-tuning.

Simiao Zuo, Xiaodong Liu, Jian Jiao, Young Jin Kim, Hany Hassan, Ruofei Zhang, Tuo Zhao, and Jianfeng Gao. 2021. Taming sparsely activated transformer with stochastic experts. arXiv preprint arXiv:2110.04260.