Model Overview

The dphn/fc-dolphin-2.6-mistral-7b-dpo-laser is a 7 billion parameter model built upon the Mistral architecture, developed by David, Fernando, and Eric. It is a specialized version of cognitivecomputations/dolphin-2.6-mistral-7b-dpo-laser, with a primary focus on enhancing function calling capabilities.

Key Capabilities

• Function Calling: The model is specifically trained on a variation of the glaive function calling v2 dataset, enabling it to understand and invoke external functions. It uses a distinct <functioncall> token and JSON format for function invocation.
• Catastrophic Forgetting Prevention: It incorporates a novel training technique, inspired by laserRMT, which involves partially freezing the model. This method is designed to prevent the loss of previously acquired knowledge, a critical advantage when teaching specific skills like function calling.
• ChatML Syntax: The model is designed to be used with ChatML syntax for representing conversational turns, aligning with other Dolphin models.

Good For

• Tool Integration: Ideal for applications that require the language model to interact with external tools, APIs, or databases by generating structured function calls.
• Automated Workflows: Suitable for building intelligent agents that can perform actions based on user prompts by invoking predefined functions.
• Developers requiring robust function calling: Offers a specialized solution for integrating LLMs into systems where precise and reliable function invocation is crucial.