Research & Education

MARLlib is a comprehensive, open-source library designed for Multi-agent Reinforcement Learning (MARL), leveraging Ray and its RLlib toolkit. It offers a unified platform for researchers and developers to create, train, and evaluate MARL algorithms across a wide array of tasks and environments. Key features include support for all task modes (cooperative, collaborative, competitive, mixed), a Gym-like interface for multi-agent environments, and flexible parameter-sharing strategies. MARLlib provides 18 pre-built algorithms with an intuitive API, making it accessible even for those new to MARL. Users can customize model architectures, policy sharing, and access over a thousand released experiments. It is compatible with Linux operating systems and offers step-by-step installation or Docker-based usage.

Translation-Agent-WebUI is an AI-powered translation tool accessible via a web user interface. It is designed to facilitate text translation between various languages, making it a convenient option for users needing quick and accessible translation services. The tool is available for free on Hugging Face, indicating its open-source or community-driven nature. While the specific features beyond basic text translation are not detailed, its web-based interface suggests ease of access without requiring complex installations. The project is hosted on Hugging Face Spaces, which often provides a platform for experimental or community-developed AI applications.

Cosmo Tech offers an AI-Simulation platform designed to tackle complex industrial challenges and enhance enterprise decision-making. The platform enables organizations to simulate intricate scenarios, capture interdependencies across the value chain, and generate high-fidelity synthetic data. This approach helps in developing resilient strategies, accelerating time-to-value, and unifying organizations around a single model of truth. It's particularly useful for asset management and supply chain optimization, allowing users to anticipate disruptions, improve decision-making, and balance sustainability with profitability. The cloud-native platform supports deep-system modeling and simulation, providing a scalable and open framework for visualizing operations and projecting future outcomes.

PySlowFast is an open-source video understanding codebase developed by FAIR, designed to provide high-performance, lightweight PyTorch implementations of state-of-the-art video backbones. It supports various video understanding research tasks, including classification and detection, and is built for rapid implementation and evaluation of novel video research ideas. The repository features implementations of methods like SlowFast Networks, Non-local Neural Networks, X3D, Multiscale Vision Transformers (MViTv1 and MViTv2), Reversible Vision Transformers (Rev-ViT and Rev-MViT), and supports advanced techniques such as Multigrid Training, MAE for Video, and MaskFeat. It also includes a comprehensive model zoo with pre-trained models and baselines, along with visualization tools for analysis and inference.

camel_tools is a comprehensive, open-source Python toolkit developed by the CAMeL Lab at New York University Abu Dhabi, specifically designed for Arabic natural language processing. It offers a wide array of functionalities including text pre-processing, advanced morphological modeling, and specialized components for Dialect Identification, Named Entity Recognition, and Sentiment Analysis. The tool is built to be accessible for researchers and developers, with clear installation instructions for various operating systems like Linux, macOS, and Windows. It also provides options for installing necessary data packages, making it a robust solution for anyone working with the complexities of the Arabic language in NLP tasks.

Suiron is an open-source project dedicated to applying machine learning principles to RC cars, offering a platform for developing and testing autonomous navigation and control systems. The project provides a comprehensive set of tools and scripts for collecting data, training neural networks, and visualizing predictions. It supports Python 2.7 and integrates with libraries like TensorFlow for model training. Users can collect data from their RC cars, train models based on this data, and then visualize how the trained models predict car behavior. This makes Suiron an excellent resource for robotics enthusiasts, machine learning students, and researchers interested in practical applications of AI in autonomous systems.

siggraph2016_colorization is an open-source tool offering code for automatic image colorization, leveraging deep learning techniques. It specifically implements a method for joint end-to-end learning of global and local image priors, allowing for nuanced and context-aware colorization. A key feature is its ability to perform simultaneous classification during the colorization process of grayscale images, which can enhance the accuracy and quality of the output. This tool is ideal for researchers, developers, and enthusiasts interested in computer vision and image processing, providing a foundational codebase for further experimentation and application in image restoration and enhancement.

TensorFlow-VAE-GAN-DRAW is an open-source collection of generative methods implemented using TensorFlow. This repository offers implementations of Deep Convolutional Generative Adversarial Networks (DCGAN), Variational Autoencoders (VAE), and DRAW: A Recurrent Neural Network For Image Generation. It allows users to experiment with and run these different generative models, providing a foundation for research and development in image generation. The project highlights that DCGANs produce decent results after 10 epochs with default parameters and outlines future enhancements like more complex data integration and replacing the current attention mechanism with a Spatial Transformer Layer.

RoseTTAFold is a deep learning model and script package designed for the accurate prediction of protein structures and interactions. This tool is an official implementation of the RoseTTAFold architecture, which employs a 3-track neural network to achieve its predictions. It is primarily intended for research in computational biology, enabling scientists to model complex protein structures and protein-protein interactions (PPIs). The package includes scripts for installation, dependency management, and running predictions for both monomer structures and complex modeling. It also features a faster 2-track version for PPI screening, making it a versatile tool for advanced biological research.

This tutorial provides a comprehensive guide for integrating DeepSpeed with the Megatron-LM GPT2 model, a powerful transformer designed for large-scale language model training. It covers essential steps from data setup and running the original Megatron-LM GPT2 model to enabling DeepSpeed through argument parsing, initialization, and training API usage. The tutorial also delves into advanced topics such as backward propagation, updating model parameters, loss scaling, and checkpoint saving/loading. Furthermore, it explains how to leverage DeepSpeed Activation Checkpoints for memory optimization and highlights the performance gains and memory footprint reduction achieved with DeepSpeed's ZeRO-2 optimizer, enabling the training of models up to 170 billion parameters with significantly improved speed and scalability.

PaLM-E is an embodied multimodal language model designed to bridge the gap between large language models and real-world physical interaction, particularly for robotics. It achieves this by directly incorporating continuous sensor modalities, such as visual and state estimation inputs, into the language embedding space of a pre-trained language model like PaLM. This allows PaLM-E to process multi-modal sentences that interleave visual, continuous state, and textual input encodings. The model is trained end-to-end for various embodied tasks, including sequential robotic manipulation planning, visual question answering, and captioning. PaLM-E demonstrates the ability to address diverse embodied reasoning tasks across multiple observation modalities and embodiments, exhibiting positive transfer from diverse joint training across internet-scale language, vision, and visual-language domains. The largest version, PaLM-E-562B, also achieves state-of-the-art performance on OK-VQA while retaining generalist language capabilities.

LLaMA (Large Language Model Meta AI) is a groundbreaking collection of foundation language models developed by Meta Research. These models, available in sizes from 7B to 65B parameters, are trained on trillions of tokens exclusively from publicly available datasets. This approach demonstrates that state-of-the-art performance can be achieved without relying on proprietary or inaccessible data. Notably, the LLaMA-13B model surpasses GPT-3 (175B) on most benchmarks, and the LLaMA-65B model competes effectively with leading models like Chinchilla70B and PaLM-540B. All LLaMA models are released to the research community, fostering innovation and accessibility in the field of large language models.

bili-hardcore is an AI-powered tool designed to automate the process of answering questions for Bilibili's hardcore member exams. Unlike OCR-based solutions, it directly interacts with the Bilibili API, ensuring higher accuracy and efficiency. The tool supports various large language models, including DeepSeek (V3.1) and Gemini (gemini-2.5-flash), with options for custom OpenAI-style APIs like those from Volcengine and SiliconFlow. Users can configure their preferred model and API key, and the tool handles the login via QR code and automatic question answering. It's crucial for users to have a Bilibili account at level 6 or above to participate in the hardcore member trials. The tool also provides guidance on troubleshooting common issues like QR code display problems, low accuracy, or API errors, and emphasizes responsible use in compliance with Bilibili's rules.

VividTalk is an open-source project designed for one-shot audio-driven talking head generation. It leverages a 3D hybrid prior to produce realistic facial animations directly from audio input. This tool is particularly suitable for researchers and developers working in AI-driven video synthesis and deepfake creation, offering a foundation for exploring advanced animation techniques. As a GitHub repository, it provides the code and resources for users to implement and experiment with the technology, making it a valuable asset for those interested in the technical aspects of generating dynamic talking head videos.

MM-EUREKA is a cutting-edge project exploring the frontiers of multimodal reasoning through rule-based reinforcement learning. It introduces powerful models such as MM-Eureka-Qwen-7B and MM-Eureka-Qwen-32B, which significantly advance performance in multidisciplinary K12 and mathematical reasoning tasks. The project has iterated on model architecture, algorithms, and data, moving from InternVL to the more robust Qwen2.5-VL base models. Key improvements include enhanced online filtering, adaptive online rollout adjustment (ADORA), and novel RL algorithms like Clipped Policy Gradient Optimization with Policy Drift (CPGD). MM-EUREKA also open-sources a comprehensive pipeline, including self-collected MMK12 datasets, to foster further research and development in multimodal AI.

rl4co is a comprehensive PyTorch library dedicated to Reinforcement Learning (RL) for Combinatorial Optimization (CO). It offers a unified and flexible framework for developing and benchmarking RL-based CO algorithms, aiming to decouple scientific research from engineering complexities. Built upon TorchRL, TensorDict, PyTorch Lightning, and Hydra, rl4co provides efficient implementations of various policies including constructive (autoregressive and non-autoregressive) and improvement methods. The library also features modular components like environment embeddings, allowing for easy adaptation to new problems. It supports installation via pip and offers clear examples for training models with default or custom configurations, making it accessible for researchers and developers in the field.

DeePMD-kit is a Python/C++ package designed to facilitate the creation of deep learning-based models for interatomic potential energy and force fields, and to perform molecular dynamics simulations. It addresses the accuracy-versus-efficiency dilemma in molecular simulations by leveraging deep learning. The package is highly modularized and interfaces with popular deep learning frameworks like TensorFlow, PyTorch, JAX, and Paddle, as well as high-performance classical and quantum MD packages such as LAMMPS, i-PI, and GROMACS. It implements the Deep Potential series models, which have been successfully applied to various systems, including organic molecules, metals, and semiconductors. DeePMD-kit also supports MPI and GPU for efficient parallel and distributed computing, making it suitable for complex scientific research.

Snipo is a comprehensive note-taking tool designed to enhance video-based learning by integrating seamlessly with Notion. It allows users to capture timestamped notes directly from videos, take screenshots of important content like charts or slides, and access video transcripts for easy reference. A standout feature is its AI Flashcards maker, which automatically generates flashcards from learning videos or any webpage, with the option to export them to Anki. Snipo supports popular learning platforms such as YouTube, Udemy, Coursera, Skillshare, and LinkedIn Learning, making it an invaluable asset for students and lifelong learners looking to streamline their study process and organize their notes efficiently.

Elia is an AI-powered tool designed to significantly enhance English vocabulary and language skills directly within the user's browsing experience. It enables users to translate English words on any webpage with a single click and save them to a personalized wordlist for future practice. A key feature is Elia's ability to highlight saved words on other websites, reinforcing learning through repeated exposure. Furthermore, it identifies and highlights new words tailored to the user's proficiency level, facilitating the acquisition of up to 300 new words monthly from their favorite online content. Elia aims to boost productivity and job performance by making language learning an integrated and effortless part of daily web browsing.

Lateron.email acts as an inbox buffer for newsletter subscriptions, designed to help users manage and prioritize their incoming content. It aggregates newsletters and offers AI summarization, allowing users to quickly grasp the main points without having to read every email in full. The tool aims to reduce inbox overload by scheduling when summarized content is delivered, ensuring users stay inspired and informed without constant interruptions. It's ideal for anyone looking to declutter their email and maintain productivity by streamlining their newsletter consumption.

Stenote is an AI-powered meeting transcription and audio recording software designed to transform spoken conversations into accurate, searchable, and actionable notes. It offers instant AI transcripts complete with speaker identification, precise timestamps, and professional summaries, making it an invaluable tool for consultants, coaches, and professional teams. The platform ensures that critical details from meetings are captured and organized, allowing users to focus on discussions without missing important information. Stenote supports multi-platform use, team collaboration, and enterprise-grade security, enhancing productivity and streamlining documentation processes for various professional settings.

Tetris-deep-Q-learning-pytorch is an open-source Python project that demonstrates the application of Deep Q-learning for training an AI agent to play the classic game Tetris. Developed with PyTorch, this tool serves as a foundational example of reinforcement learning in action. Users can leverage the provided source code to train their own Tetris-playing models from scratch or test pre-trained models. The project includes all necessary scripts for training and testing, making it accessible for those interested in understanding and experimenting with AI agents and deep learning techniques in a practical gaming context. It's an excellent resource for students and developers exploring the basics of reinforcement learning.

Sentiment-Analysis-in-Event-Driven-Stock-Price-Movement-Prediction is an open-source project designed to predict stock price movements using natural language processing (NLP) on news headlines. Specifically, it leverages Reuters news data to build a connection between Bayesian Deep Neural Networks (DNN) and stock price prediction. The methodology involves collecting and preprocessing data, including crawling ticker lists, news from Reuters, and stock prices. It then performs feature engineering through tokenization, unifying word formats, and implementing one-hot encoding. The tool trains Bayesian Convolutional Neural Networks using Stochastic Gradient Langevin Dynamics for robust predictions, which can then be used to forecast stock reactions to news events. It provides scripts for data collection, tokenization, model training, and prediction, making it a comprehensive solution for event-driven stock analysis.

ms-swift is a comprehensive, open-source framework developed by the ModelScope community, designed for fine-tuning and deploying large language models (LLMs) and multimodal large models (MLLMs). It supports over 600 text-only LLMs and 400 MLLMs, offering full-pipeline capabilities from training to inference, evaluation, quantization, and deployment. The framework integrates advanced training technologies, including Megatron parallelism (TP, PP, CP, EP) for acceleration and a rich family of GRPO reinforcement learning algorithms. ms-swift also supports various fine-tuning methods like LoRA, QLoRA, and DoRA, and provides memory optimization techniques such as Flash-Attention 2/3. It offers a Web-UI interface for simplified training, inference, evaluation, and quantization workflows, making it accessible for a wide range of users.