Model Overview

This model, Mistral-7B-Instruct-v0.3, is an instruction-tuned variant of the Mistral-7B-v0.3 base model, developed by Mistral AI. It builds upon its predecessor with several key enhancements, making it suitable for a wide range of conversational and task-oriented applications.

Key Capabilities & Features

• Extended Vocabulary: Features an expanded vocabulary of 32,768 tokens, allowing for broader language representation.
• v3 Tokenizer Support: Utilizes an updated tokenizer (v3) for improved tokenization efficiency and quality.
• Function Calling: A significant addition, enabling the model to interact with external tools and APIs by generating structured function calls. This capability is demonstrated with examples for fetching weather information.
• Instruction Following: Fine-tuned to accurately follow user instructions, making it effective for chat and command-based interactions.

Performance Highlights

Evaluations on the Open LLM Leaderboard show the model achieving an average score of 65.21%. Notable scores include:

• HellaSwag (10-Shot): 84.82%
• MMLU (5-Shot): 62.58%
• AI2 Reasoning Challenge (25-Shot): 63.91%

Good For

• General Instruction Following: Excels at understanding and executing diverse user commands.
• Chatbots and Conversational AI: Its instruction-tuned nature makes it well-suited for interactive applications.
• Tool Use and Automation: The integrated function calling capability allows for advanced integrations with external systems and services.