https://github.com/macabdul9/adaptil
Investigating the Domain Robustness of Distilled Models
https://github.com/macabdul9/adaptil
Last synced: over 1 year ago
JSON representation
Investigating the Domain Robustness of Distilled Models
- Host: GitHub
- URL: https://github.com/macabdul9/adaptil
- Owner: macabdul9
- License: mit
- Created: 2021-03-01T08:08:26.000Z (over 5 years ago)
- Default Branch: main
- Last Pushed: 2021-07-02T17:30:11.000Z (about 5 years ago)
- Last Synced: 2025-02-12T00:45:55.858Z (over 1 year ago)
- Language: Jupyter Notebook
- Size: 13.5 MB
- Stars: 0
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Adaptil
Investigating the Domain Robustness of Distilled Models and Pre-training models.
Most of the experiment in this repo was presented in [Pretrained Transformers Improve Out-of-Distribution Robustness](https://arxiv.org/abs/2004.06100). Did not train models for few tasks due to minimal shift in distribution.
## Tasks and Domains:
- Sentiment Analysis (IMDB, SST2)
- Natural Language Inference (MNLI Matching Domains/MNLI-HANS)
- Paraphase Indentification(QQP, PAWS)
## Representations:
### Amazon Product Reviews

### Amazon Product Reviews

### Paraphrase Indentification (QQP-PAWS)

### WILDS Toxic Comments

### Sentiment Analysis (IMDB-SST2)

## Results (IID-vs-IID and Performance Drop(PD)):
### Paraphrase (QQP-PAWS)


### MNLI (Matched Domain)


### Sentiment Analysis (IMDB-SST2)

