https://github.com/voorhs/hssa
https://github.com/voorhs/hssa
Last synced: 3 months ago
JSON representation
- Host: GitHub
- URL: https://github.com/voorhs/hssa
- Owner: voorhs
- Created: 2023-09-06T23:08:22.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2023-09-14T16:42:14.000Z (over 1 year ago)
- Last Synced: 2024-12-29T00:12:25.791Z (5 months ago)
- Language: Python
- Size: 181 KB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# hssa
Custom implementation of hierarchical segment-wise self-attention network (HHSA) from [Dialog-Post: Multi-Level Self-Supervised Objectives and Hierarchical Model for Dialogue Post-Training (Zhang et al., ACL 2023)](https://aclanthology.org/2023.acl-long.564/). Main difference is that each segment is separate utterance. Built upon [huffing face MPNet](https://github.com/huggingface/transformers/tree/31d452c68b34c2567b62924ee0df40a83cbc52d5/src/transformers/models/mpnet). You can load weights using `from_pretrained()` method. Example usage is provided [here](https://github.com/voorhs/hssa/blob/main/testbed.ipynb).
