mzbac/Phi-3-mini-4k-instruct-function-calling

TEXT GENERATIONConcurrent Unit Cost:1Model Size:4BQuant:BF16Context Size:4kPublished:May 7, 2024Architecture:Transformer0.0K Featherless Exclusive Cold

The mzbac/Phi-3-mini-4k-instruct-function-calling model is a 4 billion parameter instruction-tuned language model, fine-tuned from Microsoft's Phi-3-mini-4k-instruct. This model is specifically optimized for function calling, enabling it to generate structured outputs for tool use. It leverages a 4096-token context length and is designed for applications requiring precise interaction with external tools and APIs.

Loading preview...

Model Overview

The mzbac/Phi-3-mini-4k-instruct-function-calling model is a specialized variant of Microsoft's Phi-3-mini-4k-instruct, fine-tuned for robust function calling capabilities. With 4 billion parameters and a 4096-token context length, this model is engineered to interpret user requests and generate structured function calls, facilitating interaction with external tools and APIs.

Key Capabilities

  • Function Calling: The primary capability is generating precise JSON-formatted function calls based on natural language prompts and provided tool definitions.
  • Instruction Following: Inherits strong instruction-following abilities from its base Phi-3-mini-4k-instruct model.
  • Tool Integration: Designed to be easily integrated into systems requiring automated tool use, such as web search, data retrieval, or complex workflow automation.

Training Details

The model was fine-tuned using MLX-LM on the mzbac/function-calling-phi-3-format-v1.1 dataset. The training involved LoRA (Low-Rank Adaptation) with 32 layers, a rank of 128, and an alpha of 256, over 111,000 iterations. This targeted fine-tuning process specifically enhanced its ability to parse user intent and map it to predefined tool schemas.

Good For

  • Automated Assistants: Building intelligent agents that can interact with external services.
  • Workflow Automation: Automating tasks by translating natural language commands into executable function calls.
  • API Interaction: Generating structured requests for APIs based on user input.