Multi-task Language Evaluation on FineTasks average across languages
4.35Average RankMLP MKC+
Evaluation Results
| Method | Links | |
|---|---|---|
| MLP MKC+Filtering Method=MLP, Classifier Training Dataset=MKC+, Model Parameters=1B, Evaluation Tokens=Average of 70B and 119B2025.02 | 4.35 | |
| MLP MKCFiltering Method=MLP, Classifier Training Dataset=MKC, Model Parameters=1B, Evaluation Tokens=Average of 70B and 119B2025.02 | 6.11 | |
| FT MKC+Filtering Method=FastText (FT), Classifier Training Dataset=MKC+, Model Parameters=1B, Evaluation Tokens=Average of 70B and 119B2025.02 | 7.17 | |
| FT MKCFiltering Method=FastText (FT), Classifier Training Dataset=MKC, Model Parameters=1B, Evaluation Tokens=Average of 70B and 119B2025.02 | 8.04 | |
| CS MKCFiltering Method=Cosine Similarity (CS), Classifier Training Dataset=MKC, Model Parameters=1B, Evaluation Tokens=Average of 70B and 119B2025.02 | 8.1 | |
| BaselineDataset=FineWeb-2, Model Parameters=1B, Evaluation Tokens=Average of 70B and 119B2025.02 | 8.72 | |
| CS MKC+Filtering Method=Cosine Similarity (CS), Classifier Training Dataset=MKC+, Model Parameters=1B, Evaluation Tokens=Average of 70B and 119B2025.02 | 8.79 |