ValiantLabs/Llama3-70B-Fireplace Overview

ValiantLabs/Llama3-70B-Fireplace is a 70 billion parameter model developed by Valiant Labs, based on the Llama 3 70B Instruct architecture. This model is specifically engineered to combine high-performance chat capabilities with system-message function-calling, leveraging the Llama 3 Instruct format.

Key Capabilities

• Function Calling: Integrates function-calling abilities directly into a powerful chat model, enabling the AI to interact with external tools and APIs.
• Llama 3 Instruct Format: Utilizes the Llama 3 Instruct prompt format for both chat and system-message function-calling, ensuring compatibility and leveraging the base model's strengths.
• Versatile Chat: Functions as a high-performance chat model alongside its function-calling features.

Training and Performance

Fireplace was trained using the glaiveai/glaive-function-calling-v2 dataset, converted to the Llama 3 Instruct format. On the Open LLM Leaderboard, it achieves an average score of 36.82, with notable performance in IFEval (0-Shot) at 77.74 and BBH (3-Shot) at 49.56.

Good For

• Applications requiring a large, capable language model that can also execute functions or interact with external systems.
• Developers looking for a Llama 3-based model with integrated function-calling for building intelligent agents or conversational interfaces that need to perform actions.
• Use cases where both general conversational ability and specific tool-use capabilities are essential.