Model Catalogue
- Repository of boutique AI models
- Still a work in progress
Pretrained base models for embedding
English models were finetuned on a subset of Zyphra/Zyda-2.
Multilingual models were aligned using agentlans/en-translations.
Text statistics models
- Input: text
- Output: number
| Base model | Language | Quality | Readability | Sentiment |
|---|---|---|---|---|
| deberta-v3-xsmall-zyda-2 | English only | Link | Link | Link |
| deberta-v3-base-zyda-2 | English only | Link | Link | Link |
| multilingual-e5-small-aligned | Multilingual | Link | Link | Link |
| mdeberta-v3-base | Multilingual | Link | Link | Link |
Note: the mdeberta-v3-base models were trained on a previous version of the dataset, not the complete dataset
Small text-to-text models
- Input: text
- Output: text
These are English only.
| Task | Model | Dataset |
|---|---|---|
| Keyword extraction | flan-t5-small-keywords | wikipedia-paragraph-keywords |
| Title generation | flan-t5-small-title | wikipedia-paragraph-titles |
Natural language inference (NLI) models
- Input: text
- Output: label (entailment, neutral, or contradiction)
These are English only.