# Two-stage training

Whisper model released by the Swedish National Library, optimized for Swedish, trained on 50,000+ hours of Swedish speech data, outperforming the original OpenAI version

Dragoman is a sentence-level English-Ukrainian translation model that employs a two-stage training process, achieving optimal performance with a BLEU score of 32.34 on the FLORES-101 English-Ukrainian development test subset.

Ahma-7B is a 7-billion-parameter decoder-only Transformer model based on Meta Llama(v1) architecture, fully pretrained from scratch using Finnish language.

Animagine XL 3.0 Base is the foundational version of an advanced anime text-to-image model, focusing on establishing core functionalities and refining key aspects.