Occiglot-7B-EU5-Instruct

Occiglot-7B-EU5-Instruct is the instruction-tuned version of Occiglot's multilingual 7B model covering English, German, French, Spanish, and Italian. It was produced by the academic, non-profit Occiglot Research Collective through bilingual continual pre-training of Mistral 7B plus multilingual instruction tuning.

The project aims at European AI sovereignty and language diversity, using 700B additional multilingual tokens for continual pre-training. Weights are released under Apache 2.0. The collective notes the model was not safety aligned and may produce problematic outputs.