Part-of-speech tagging on MasakhaPOS Setswana
83Token AccuracyAfroXLMR-large
Evaluation results
| Method | Links | |
|---|---|---|
| AfroXLMR-largeArchitecture=Encoder-Only, Size=0.550B2026.03 | 83 | |
| AfroXLMR-baseArchitecture=Encoder-Only, Size=0.270B2026.03 | 82.7 | |
| AfriBERTaArchitecture=Encoder-Only, Size=0.126B2026.03 | 82.5 | |
| Llama-3.1-70B-InstructArchitecture=Decoder-Only, Size=70B2026.03 | 52.2 | |
| Llama-3.1-8B-InstructArchitecture=Decoder-Only, Size=8B2026.03 | 49.5 | |
| MzansiLM mono-masakhapos-ftArchitecture=Decoder-Only, Size=0.125B2026.03 | 5.6 | |
| MzansiLM multi-masakhapos-ftArchitecture=Decoder-Only, Size=0.125B2026.03 | 0.5 | |
| MzansiLM Base 0-shotArchitecture=Decoder-Only, Size=0.125B2026.03 | 0 | |
| MzansiLM general-ftArchitecture=Decoder-Only, Size=0.125B2026.03 | 0 | |
| InkubaLM-0.4BArchitecture=Decoder-Only, Size=0.4B2026.03 | 0 | |
| AfroLlama-V1Architecture=Decoder-Only, Size=8B2026.03 | 0 | |
| Aya-101Architecture=Encoder-Decoder, Size=13B2026.03 | 0 |