mii-llm/maestrale-chat-v0.4-beta

Maestrale-chat-v0.4-beta is a 7 billion parameter Mistral-7b based language model developed by @efederici and @mferraretto, specifically continued pre-trained for the Italian language on a curated high-quality corpus and merged with occiglot-7b-eu5. It has been fine-tuned with SFT on 1.7M conversations/instructions and aligned using DPO, featuring improved truthfulness, enhanced math and reasoning capabilities, and support for Mermaid mindmaps. This model is optimized for Italian language tasks, agentic behavior, and structured output generation.