Machine Translation (En to XX) on Flores-200 (test)
88.9COMETX-ALMA-13B (Only SFT, MoE)
Evaluation Results
| Method | Links | |
|---|---|---|
| X-ALMA-13B (Only SFT, MoE)# Tokens (Pre-training)=110B, # Params (Base/Adapter)=13B/16B, # Params (Total)=29B2025.02 | 88.9 | |
| X-ALMA-13B-DAT (MoE)# Tokens (Pre-training)=20B, # Params (Base/Adapter)=13B/4B, # Params (Total)=17B2025.02 | 87.8 | |
| X-ALMA-13B-DATM (MoE)# Tokens (Pre-training)=20B, # Params (Base/Adapter)=13B/1B, # Params (Total)=14B2025.02 | 87.8 | |
| NLLB-3.3B# Tokens (Pre-training)=null, # Params (Base/Adapter)=3B/-, # Params (Total)=3B2025.02 | 87.4 | |
| Aya-101# Tokens (Pre-training)=null, # Params (Base/Adapter)=13B/-, # Params (Total)=13B2025.02 | 84.1 | |
| LLaMAX3-Alpaca-8B# Tokens (Pre-training)=66B, # Params (Base/Adapter)=8B/-, # Params (Total)=8B2025.02 | 84.1 | |
| Aya-23-35B# Tokens (Pre-training)=null, # Params (Base/Adapter)=35B/-, # Params (Total)=35B2025.02 | 76 | |
| Aya-23-8B# Tokens (Pre-training)=null, # Params (Base/Adapter)=8B/-, # Params (Total)=8B2025.02 | 74.4 |