Supported Languages | Bulgarian (NLP), Croatian (NLP), Czech (NLP), Danish (NLP), Dutch (NLP), English (NLP), Estonian (NLP), Finnish (NLP), French (NLP), German (NLP), Greek (NLP), Hungarian (NLP), Irish (NLP), Italian (NLP), Latvian (NLP), Lithuanian (NLP), Maltese (NLP), Polish (NLP), Portuguese (NLP), Romanian (NLP), Slovak (NLP), Slovenian (NLP), Spanish (NLP), Swedish (NLP), Arabic (NLP), Catalan (NLP), Chinese (NLP), Galician (NLP), Hindi (NLP), Japanese (NLP), Korean (NLP), Norwegian (NLP), Russian (NLP), Turkish (NLP), Ukrainian (NLP) |
|
Training Details |
Data Sources: | Web data, parallel data, high-quality datasets |
|
Data Volume: | |
Methodology: | Instruction tuning on EuroBlocks focused on general instruction-following and machine translation |
|
Hardware Used: | |
Model Architecture: | Standard dense Transformer with grouped query attention, pre-layer normalization, RMSNorm, SwiGLU activation, and RoPE positional embeddings |
|
|