SimplyRetrieve

Lightweight chat AI platform featuring custom knowledge, open-source LLMs, prompt-engineering, retrieval analysis. Highly customizable. For Retrieval-Centric &

What is SimplyRetrieve?

Lightweight chat AI platform featuring custom knowledge, open-source LLMs, prompt-engineering, retrieval analysis. Highly customizable. For Retrieval-Centric & Retrieval-Augmented Generation.

Overview of SimplyRetrieve

What it is

SimplyRetrieve is an open-source tool that provides a fully localized, lightweight, and user-friendly GUI and API platform for Retrieval-Centric Generation (RCG). It allows users to create chat tools with their own documents and language models, and is highly customizable. The tool is designed to be run on a single Nvidia GPU such as T4, V100, or A100. It is built on Hugging Face, Gradio, PyTorch, and Faiss. The default LLM is Wizard-Vicuna-13B-Uncensored, and the default embedding model is multilingual-e5-base. A technical report is available on arXiv, and a short video on YouTube. The tool is affiliated with the Institute of Memory Technology R&D, Kioxia Corporation, Japan.

Key features

  • GUI and API based Retrieval-Centric Generation platform - Retrieval Tuning Module with Prompts Engineering, Tool Configuration, and Retrieval Analysis - Private Knowledge Base Constructor - Fully localized, private, and lightweight access to various sizes of open-source Large Language Models (LLMs)
  • Multi-user concurrent access to UI using Gradio's queue function - Knowledge Tab for on-the-fly knowledge creation and appending

Use cases

  • Examining the effectiveness of retrieval-centric generation in developing safer, more interpretable, and responsible AI systems - Optimizing the efficiency of separating context interpretation and knowledge memorization within retrieval-centric generation - Improving prompt engineering techniques for retrieval-centric generation - Implementing Chat AI for fully local and edge computing - Creating personalized AI assistants tailored to individual users

Getting started

First, git clone the repository. In a GPU-based Linux machine, activate a Python virtual environment and install the necessary packages: pip install -r requirements.txt Optionally, prepare your own knowledge source by placing documents (pdf, txt, doc, etc.) into chat/data/ and running the data preparation script: cd chat/ CUDA_VISIBLE_DEVICES=0 python prepare.py --input data/ --output knowledge/ --config configs/default_release.json Alternatively, use the Knowledge Tab in the GUI to add knowledge on-the-fly. To run the tool, set the current path to chat/ and execute: CUDA_VISIBLE_DEVICES=0 python chat.py --config configs/default_release.json Then access the web-based GUI at http://:7860. For GUI operation manual, see the docs/ directory. For API access, see the examples/ directory.

License and pricing

The tool is licensed under the MIT License and is open source (free). The license is stated in the metadata as MIT, and the pricing type is open_source.

Based on the project's public README and repository metadata.

Add this badge to your README

Featured on YunoDB Badge styles & help

Maintain this project? Show that it is featured on YunoDB.

Markdown[![Featured on YunoDB](https://yunodb.com/badge/rcgaisimplyretrieve.svg)](https://yunodb.com/tools/rcgaisimplyretrieve?utm_source=github&utm_medium=badge)
HTML<a href="https://yunodb.com/tools/rcgaisimplyretrieve?utm_source=github&utm_medium=badge"><img src="https://yunodb.com/badge/rcgaisimplyretrieve.svg" alt="Featured on YunoDB" height="20"></a>