Today's Open Source (2024-07-18): Mem0, Providing Long-Term Memory for LLM and Agents
Discover top AI open-source projects: mem0, React Flow, OpenCompass, FasterLivePortrait, EfficientQAT, and PR Pilot. Enhance workflows, LLMs, and real-time animations.
I'll share some interesting AI open-source models and frameworks today.
Project: mem0
Mem0 is an open-source project for developing memory layers for LLMs and AI Agents, enabling personalized interactions.
The project stores and manages user-specific information, including conversation history, preferences, and metadata, helping AI applications provide tailored results.
For example, in a personalized learning assistant, long-term memory allows it to remember user preferences, past interactions, and progress, offering a more customized and effective learning experience.
https://github.com/mem0ai/mem0
Project: React Flow
xyflow is an open-source project for node-based workflow development, using low code to help users create complex workflows and improve efficiency.
Built on React and Svelte frameworks, it is ready to use out of the box and has accumulated 20k stars.
https://github.com/xyflow/xyflow
Project: OpenCompass
OpenCompass is an LLM evaluation platform supporting various models (like Llama3, Mistral, InternLM2, GPT-4, LLaMa2, Qwen, GLM, Claude, etc.) across 100+ datasets.
The platform includes CompassKit, CompassHub, and CompassRank components, aimed at simplifying and speeding up the exploration and utilization of various benchmarks for researchers and practitioners.
https://github.com/open-compass/opencompass
Project: FasterLivePortrait
FasterLivePortrait is a real-time portrait animation project supporting onnx and TensorRT.
It runs LivePortrait in real-time on an RTX 3090 GPU, achieving 30+ FPS.
The project also supports Docker, providing a one-click integration package for Windows and MacOS.
https://github.com/warmshao/FasterLivePortrait
Project: EfficientQAT
EfficientQAT is a project for efficient quantization-aware training (QAT) for LLMs.
The project aims to achieve end-to-end inference acceleration using BitBLAS technology. Currently, it reduces memory usage for training and inference, with future plans for inference acceleration.
EfficientQAT pushes the limits of uniform (INT) quantization efficiently.
https://github.com/OpenGVLab/EfficientQAT
https://arxiv.org/abs/2407.11062
Project: PR Pilot
PR Pilot is an AI Agent designed to streamline development workflows. It can search and manipulate codebases, browse the internet, and interact with GitHub issues and pull requests.
PR Pilot supports various interaction methods, including a command-line interface, Python SDK, and REST API, helping developers save time and increase productivity.
