Maestrale Chat v0.3-beta: Italian-Optimized Mistral Model

Maestrale chat v0.3-beta is a 7 billion parameter language model built upon the Mistral architecture, developed by @efederici and @mferraretto. Its core distinction lies in its extensive optimization for the Italian language, achieved through continued pre-training on a large-scale, high-quality Italian corpus.

Key Capabilities & Features

• Italian Language Focus: Specifically pre-trained and fine-tuned for robust performance in Italian conversations and instructions.
• KTO Alignment: Incorporates KTO (Kahneman-Tversky Optimization) for improved alignment.
• Function Calling: Version 0.3 introduces function calling capabilities, enhancing its utility for integrated applications.
• Optimized System Prompt: Features a reduced default system prompt to conserve tokens and improve efficiency.
• ChatML Format: Utilizes the ChatML prompt format for structured conversational interactions.

Intended Use Cases

This model is primarily intended for chat-based applications requiring strong Italian language understanding and generation. While it is a beta version, it offers specialized capabilities for developers building Italian-centric AI solutions.