Maestrale-chat-v0.4-beta: An Italian-Optimized Mistral-7B Model

Maestrale-chat-v0.4-beta is a 7 billion parameter language model built upon the Mistral-7b architecture, developed by @efederici and @mferraretto. It has undergone extensive continued pre-training using a large-scale, high-quality Italian corpus and integrates with the occiglot-7b-eu5 model, making it highly proficient in the Italian language.

Key Capabilities & Features

• Italian Language Proficiency: Specifically optimized for Italian through continued pre-training and fine-tuning.
• Enhanced Reasoning: Features improved mathematical and reasoning capabilities.
• Agentic Behavior: Supports agent-like interactions and functionalities.
• Truthfulness: Aligned with DPO to improve factual accuracy and reduce hallucinations.
• Structured Output: Capable of generating Mermaid mindmaps and SQL queries from natural language prompts.
• Creative Text Generation: Can generate articles from titles and indices, and perform Latin translations and poem generation.

Use Cases & Strengths

This model is particularly well-suited for applications requiring strong performance in Italian, especially for:

• Italian Chatbots and Assistants: Its fine-tuning on 1.7M conversations makes it effective for interactive applications.
• Structured Data Generation: Ideal for tasks like generating SQL code from natural language or creating Mermaid mindmaps.
• Content Creation: Useful for generating articles, poems, or translations in Italian.
• Reasoning Tasks: Benefits from improved math and reasoning for problem-solving in Italian contexts.

Maestrale-chat-v0.4-beta is a beta version, designed to be safe and capable of refusing to answer toxic questions.