Llama 3 Multilingual Fine-tuning

1 repo

Fine-tuned variants of Meta's Llama 3 8B base model optimized for multiple languages using modern alignment techniques like DPO (Direct Preference Optimization) and KTO (Kahneman-Tversky Optimization). These repositories represent systematic efforts to adapt the base model across language pairs including English paired with European languages (Spanish, Russian, German, French) and South Asian/African languages (Bengali, Swahili, Thai), leveraging safetensors for efficient model serialization.

Related clusters