Jul 24, 2023
Speaker · 0 followers
Speaker · 0 followers
Speaker · 0 followers
Speaker · 0 followers
Speaker · 0 followers
Speaker · 0 followers
Speaker · 0 followers
Speaker · 1 follower
Speaker · 0 followers
Currently, most machine learning models are trained by a centralized team and are rarely updated.In contrast, open-source software development involves the iterative development of a shared artifact through distributed collaboration of many individuals using version control systems.In the interest of enabling collaborative and continual improvement of machine learning models, we introduce Git-Theta, a version control system for machine learning models.Git-Theta is an extension to Git, the most widely used version control software, thereby allowing fine-grained tracking of changes to a model's parameters alongside code and other artifacts.Existing version control systems for model checkpoints treat the model as a blob of data and therefore do not support cheaply communicable updates or merging, which are fundamentally important operations for receiving and incorporating changes from contributors.Git-Theta supports various existing approaches for parameter-efficient training and model merging and provides functionality for displaying meaningful information about thedifference between two versions of a model.In addition, Git-Theta includes a plug-in system that enables downstream users to easily add support for new functionality.In this paper, we introduce Git-Theta's key features and design choices.We also include an example use-case of Git-Theta where a pre-trained model is adapted to new tasks through parameter-efficient fine-tuning, merging, and manual removal of parameters.We publicly release Git-Theta in hopes of kickstarting a new era of collaborative model development.Currently, most machine learning models are trained by a centralized team and are rarely updated.In contrast, open-source software development involves the iterative development of a shared artifact through distributed collaboration of many individuals using version control systems.In the interest of enabling collaborative and continual improvement of machine learning models, we introduce Git-Theta, a version control system for machine learning models.Git-Theta is an extension to Git, the most w…
Professional recording and live streaming, delivered globally.
Presentations on similar topic, category or speaker
Xingang Peng, …
Zheng Xiong, …
Yuval Atzmon, …
Yunlong Hou, …