Welcome to the llm4qual repository, your go-to toolkit for enhancing qualitative research with Large Language Models (LLMs). Our mission is to harness the power of LLMs to streamline and support various tasks in qualitative research.
As of Jan 2024, LLM4Qual supports a single workflow called llm-proxy, which uses LLMs to simulate human annotators in text labeling tasks. But we plan to add support for additional workflows.
The llm-proxy workflow is designed to act as a stand-in for human annotators in text-annotation tasks. By leveraging LLMs, llm-proxy provides a sophisticated and automated approach to evaluating qualitative data across multiple dimensions.
- Dataset Loading: Integrate seamlessly with Langchain + HuggingFace to import your qualitative datasets.
- Prompt Templating: Organize your annotation prompts with ease using our structured YAML file templates, compatible with Langchain.
- Few-Shot Examples: Incorporate few-shot learning by providing examples within your YAML files to enhance the LLM's annotation accuracy.
- Evaluation Suite Setup: Utilize helper classes and functions to establish a HuggingFace
EvaluationSuitetailored to your research needs. - Result Reporting: After analysis, neatly save and report the evaluation outcomes for further review and action.
To get started with llm4qual, follow these simple steps:
Ensure you have the latest versions of Langchain and HuggingFace libraries installed. You can install these using pip:
pip install langchain huggingface_hub- Clone the LLM4Qual repository to your local machine:
git clone https://github.com/msamogh/llm4qual.git
cd llm4qual- Navigate to the llm-proxy directory and install any necessary dependencies:
cd llm-proxy
poetry shell
poetry install- Install Transformers4Qual:
git clone https://github.com/msamogh/transformers4qual.git
pip install -e transformers4qual- Install Evaluate4Qual:
git clone https://github.com/msamogh/evaluate4qual.git
pip install -e evaluate4qual- Load Your Dataset: Use Langchain to load your dataset into the framework.
- Create Prompt Templates: Define your annotation dimensions and write prompt templates in YAML format.
- Few-Shot Examples: Add few-shot examples to your YAML files to guide the LLM in annotation tasks.
- Initialize Evaluation Suite: With the provided helper classes, set up your EvaluationSuite.
- Run Evaluation: Execute the evaluation and collect results.
- Save & Report: Utilize our reporting tools to save and visualize the analysis outcomes.
Distributed under the MIT License. See LICENSE for more information.
For any additional questions or comments, please reach out to [email protected] or submit an issue.