Meltemi: The first open Large Language Model for Greek

TL;DR

This work addresses the shortage of open, high-quality LLMs for Greek by extending Mistral-7B through continual pretraining on a Greek-rich corpus and expanding its tokenizer and embeddings to better support Greek text. It couples this foundation with instruction tuning via ORPO on a translated Greek preference dataset to yield Meltemi 7B Instruct, and introduces a Greek benchmark suite derived from translated English tests. The results show substantial gains in Greek language tasks (+20.2% average) while highlighting challenges in English capabilities after continual pretraining, underscoring language-specific adaptation dynamics. The work emphasizes sustainability, open-access availability, and future directions for larger, multimodal Greek-language AI systems with ongoing data updates.

Abstract

We describe the development and capabilities of Meltemi 7B, the first open Large Language Model for the Greek language. Meltemi 7B has 7 billion parameters and is trained on a 40 billion token Greek corpus. For the development of Meltemi 7B, we adapt Mistral, by continuous pretraining on the Greek Corpus. Meltemi 7B contains up-to-date information up to September 2023. Furthermore, we have translated and curated a Greek instruction corpus, which has been used for the instruction-tuning of a chat model, named Meltemi 7B Instruct. Special care has been given to the alignment and the removal of toxic content for the Meltemi 7B Instruct. The developed models are evaluated on a broad set of collected evaluation corpora, and examples of prompts and responses are presented. Both Meltemi 7B and Meltemi 7B Instruct are available at https://huggingface.co/ilsp under the Apache 2.0 license.