30 Latest AI Open Source Projects of the Week(2024.12.9-2024.12.13)
Explore 30 top AI open-source projects this week (2024.12.9-2024.12.13), including Llama 3.3, EXAONE 3.5, and innovative AI tools for research and creation!
I’m sharing some interesting AI open-source models and frameworks for this week (2024.12.9-2024.12.13).
There are a total of 30 AI open-source projects.
Project: Llama 3.3
Llama-3.3-70B-Instruct is a large language model developed by Meta, focused on text generation tasks with comprehensive support for 8 languages. Llama 3.3 employs an optimized transformer architecture, incorporating advanced techniques such as supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF). It supports a context length of up to 128K tokens, approximately equivalent to 400 pages of text.
https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct
Project: EXAONE 3.5
EXAONE 3.5 is a series of bilingual (English and Korean) instruction-tuned generative models developed and released by LG AI Research, with parameter scales ranging from 2.4B to 32B. The models support long-context processing of up to 32K tokens, excelling in real-world applications and long-context understanding, offering competitive performance compared to recently released models.
https://github.com/LG-AI-EXAONE/EXAONE-3.5
Project: Director
Director is a framework for building video agents capable of handling complex video tasks such as searching, editing, compiling, and generating videos while streaming results in real time. Built on the "video-as-data" infrastructure from VideoDB, it simplifies media workflows for developers, creators, and teams, unlocking new possibilities. Users can execute tasks with simple natural language commands.
https://github.com/video-db/Director
Project: Amurex
Amurex is a simple yet powerful AI meeting assistant designed to integrate seamlessly into your workflow. Using cutting-edge AI technology, Amurex ensures users never miss critical details, stay on top of action items, and make every meeting more productive. As an open-source tool, Amurex is transparent, secure, and privacy-focused, delivering a seamless AI-powered experience.
https://github.com/thepersonalaicompany/amurex
Project: Lift3D
Lift3D is a 3D robotic representation framework designed to develop 3D manipulation strategies by enhancing 2D foundational models. The project improves implicit and explicit 3D robotic representations, extending the functionality of 2D models to support robotic manipulation in complex 3D environments.
https://github.com/PKU-HMI-Lab/LIFT3D
Project: SmolChat-Android
SmolChat-Android is an application for running small language models (SLMs) locally on Android devices. Users can add or remove GGUF models and customize system prompts or inference parameters. The project provides an easy-to-use interface, enabling users to quickly create downstream tasks and generate responses. The codebase is simple to understand and extend.
https://github.com/shubham0204/SmolChat-Android
Project: See3D
See3D is a visually conditioned multi-view diffusion model designed for open-world 3D creation using large-scale internet videos. The model acquires 3D knowledge by observing visual content from video data and utilizes an automated data filtering pipeline to exclude videos with inconsistent or insufficient multi-view observations, constructing a high-quality, diverse large-scale multi-view image dataset called WebVi3D. See3D achieves high-fidelity 3D generation without explicit 3D geometry or camera pose annotations by introducing innovative visual conditioning and deformation-based 3D generation frameworks.
https://github.com/baaivision/See3D
Project: Florence-VL
Florence-VL is a project enhancing vision-language models by combining generative vision encoders with deep and broad fusion techniques. It aims to improve the interaction between vision and language, offering more accurate multimodal understanding and generation capabilities. The project provides multiple resource versions, including pre-trained and instruction-tuned checkpoints, supporting various evaluation tasks.
https://github.com/JiuhaiChen/Florence-VL
Project: RAG Research Agent Template
RAG Research Agent Template is a starter project designed to help developers use LangGraph to create RAG research agents in LangGraph Studio. The project includes three main graphs: the Index Graph for document indexing, the Retrieval Graph for managing chat history and generating responses based on retrieved documents, and the Research Subgraph for executing research plans related to user queries with LangChain.
https://github.com/langchain-ai/rag-research-agent-template
Project: ZhiLight
ZhiLight is a highly optimized large-model inference engine developed by Zhihu and ModelBest Inc. The engine is designed for accelerating inference of Llama and its variants, particularly on PCIe-based GPUs. Compared to mainstream open-source inference engines, ZhiLight offers significant performance advantages.