Mark RSE checkpoint as historical research; preserve original configuration
Browse files
README.md
CHANGED
|
@@ -1,7 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
The RSE-BERT-large-STS is trained with 2 relations including:
|
| 2 |
1) entailment
|
| 3 |
2) duplicate_question
|
| 4 |
|
|
|
|
| 5 |
The BERT-large-uncased model is used as initialization.
|
| 6 |
|
| 7 |
-
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags:
|
| 3 |
+
- historical-research
|
| 4 |
+
- rse
|
| 5 |
+
---
|
| 6 |
+
|
| 7 |
+
# Historical research checkpoint — RSE
|
| 8 |
+
|
| 9 |
+
**No longer actively maintained. Retained for reproducibility of the original work.**
|
| 10 |
+
|
| 11 |
+
This repository contains one variant of the RSE research models. The repository name identifies the backbone (BERT or RoBERTa), model size (base or large), and experiment variant (STS, USEB, Transfer, or 10-relations). Refer to the original description below for this checkpoint's configuration.
|
| 12 |
+
|
| 13 |
+
## Original model description
|
| 14 |
+
|
| 15 |
The RSE-BERT-large-STS is trained with 2 relations including:
|
| 16 |
1) entailment
|
| 17 |
2) duplicate_question
|
| 18 |
|
| 19 |
+
|
| 20 |
The BERT-large-uncased model is used as initialization.
|
| 21 |
|
| 22 |
+
|
| 23 |
+
It can be used ideally for STS datasets.
|