| Author(s): |
Perin, Gabriel J.
;
Chen, Xuxi
;
Liu, Shusen
;
Kailkhura, Bhavya
;
Wang, Zhangyang
;
Gallagher, Brian
Total Authors: 6
|
| Document type: | Journal article |
| Source: | FINDINGS OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS: ACL 2024; v. N/A, p. 7-pg., 2024-01-01. |
| Abstract | |
Traditionally, developing new language models (LMs) capable of addressing multiple tasks involves fine-tuning pre-trained LMs using a wide collection of datasets, a process that often incurs significant computational expenses. Model merging emerges as a cost-effective alternative, allowing the integration of existing models fine-tuned on different tasks into a single model that performs well across all tasks, eliminating the need for additional training. In this paper, we propose RankMean, an algorithm for merging fine-tuned LMs without requiring any downstream data. RankMean determines merging coefficients based on the relative rankings of weight change magnitudes and applies these coefficients for module-wise integration of various fine-tuned models. Our experimental results demonstrate that RankMean outperforms existing baseline methods on multiple benchmarks. The code is available at github.com/VITA-Group/RankMean. (AU) | |
| FAPESP's process: | 23/15047-4 - Model Merging for Large Language Model |
| Grantee: | Gabriel Jacob Perin |
| Support Opportunities: | Scholarships abroad - Research Internship - Scientific Initiation |
| FAPESP's process: | 22/11645-1 - Classification of stars, galaxies, and quasars based on photometric multiband images |
| Grantee: | Gabriel Jacob Perin |
| Support Opportunities: | Scholarships in Brazil - Scientific Initiation |
| FAPESP's process: | 22/15304-4 - Learning context rich representations for computer vision |
| Grantee: | Nina Sumiko Tomita Hirata |
| Support Opportunities: | Research Projects - Thematic Grants |