-
Notifications
You must be signed in to change notification settings - Fork 0
Data Version History
Shakleen Ishfar edited this page May 21, 2024
·
5 revisions
| Version | Data Source | Folds | Split Strategy | Negative Sampling | Sliding Window | Max Length | Processing |
|---|---|---|---|---|---|---|---|
| 1 | Competition | 7 | StratefiedGroupKFold | No | No | 512 | Strip extra space |
| 2 | Competition | 7 | StratefiedGroupKFold | No | Yes | 512 | New line to space and multiple space to single space |
| 3 | Competition | 7 | StratefiedGroupKFold | Yes | Yes | 512 | Same as 2 |
| 4 | Competition + Persuade 2.0 | 7 | StratefiedGroupKFold | No | Yes | 512 | Same as 2 |
| 5 | Competition + Persuade 2.0 | 7 | StratefiedGroupKFold | Yes | Yes | 512 | Same as 2 |
Base tokenizer with no added tokens.
Base tokenizer with two added tokens
- New line token (\n)
- Double space token ( )
DeBERTA-V3 model with Mean pooling for classification