unclecode/llama3-function-call-Q4_K_M_GGFU-240424

Function Calling and Tool Use LLaMA Models

This repository contains two main versions of LLaMA models fine-tuned for function calling and tool use capabilities:

1. Fine-tuned version of the LLama3-8b-instruct model
2. tinyllama - a smaller model version

For each version, the following variants are available:

• 16-bit quantized model
• 4-bit quantized model
• GGFU format for use with llama.cpp

Dataset

The models were fine-tuned using a modified version of the ilacai/glaive-function-calling-v2-sharegpt dataset, which can be found at unclecode/glaive-function-calling-llama3.

Usage

To learn how to use these models, refer to the Colab notebook:

This is the first version of the models, and work is in progress to further train them with multi-tool detection and native tool binding support.

Library and Tools Support

A library is being developed to manage tools and add tool support for major LLMs, regardless of their built-in capabilities. You can find examples and contribute to the library at the following repository:

https://github.com/unclecode/fllm

Please open an issue in the repository for any bugs or collaboration requests.

Other Models

Here are links to other related models:

• unclecode/llama3-function-call-lora-adapter-240424
• unclecode/llama3-function-call-16bit-240424
• unclecode/llama3-function-call-4bit-240424
• unclecode/llama3-function-call-Q4_K_M_GGFU-240424
• unclecode/tinyllama-function-call-lora-adapter-250424
• unclecode/tinyllama-function-call-16bit-250424
• unclecode/tinyllama-function-call-Q4_K_M_GGFU-250424

License

These models are released under the Apache 2.0 license.

Uploaded model

• Developed by: unclecode
• License: apache-2.0
• Finetuned from model : unsloth/llama-3-8b-Instruct-bnb-4bit

This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.