FANNO

Model Details

Based on LlaMA2-7b.

Model Description

instruction-finetuning

autonomous-framework

data-annotation

FANNO is an innovative, fully autonomous, open-sourced framework designed to streamline the annotation process for instruction datasets without requiring pre-existing annotated data. Leveraging the capabilities of the Mistral-7b-instruct model, FANNO efficiently generates diverse and high-quality datasets through a structured process that includes document pre-screening, instruction generation, and response generation.

Key Features

Autonomous Annotation: Eliminates the need for manual annotations or costly API calls of proprietary LLMs, making the annotation process cost-effective and efficient.

High-Quality Data Generation: Produces datasets with diversity and complexity that are comparable to human-annotated or cleaned datasets, such as Alpaca-GPT4-Cleaned.

Open-Sourced Framework: Fully open-sourced, allowing the community to leverage and contribute to the ongoing improvement of the framework.