Projects

### Overview: This repository is a curated list of resources for Large Language Models (LLM) and Foundation Models (LM, LM, FM) specifically designed for Time Series, Spatiotemporal, and Event Data. It aims to provide a comprehensive and systematic overview of recent advancements in the field, with a focus on the latest resources such as papers, codes, and datasets. ### Key Features: 1. **Large Models for Time Series and Spatio-Temporal Data**: Comprehensive resources for understanding how large language models can be applied to time series and spatiotemporal data. 2. **Survey Paper**: A survey paper on Large Models for Time Series and Spatio-Temporal Data, including a paper from *ICML* 2024. 3. **Time-FFM**: A paper on Time-FFM, a model that integrates federated learning with large language models for time series forecasting. 4. **Time-MMD**: A dataset for time series analysis, specifically designed for federated learning. 5. **From News to Forecast**: A paper that integrates event analysis into LLM-based time series forecasting. 6. **Autotimes**: An autoregressive time series forecasting model using large language models. 7. **S^2IP-LLM**: A model that uses semantic space informed prompt learning for time series forecasting. 8. **TimeCMA**: A model that uses cross-modality alignment for multivariate time series forecasting. 9. **Time-LLM**: A model that uses large language models for time series forecasting. 10

### Overview: ### PaddleViT is a comprehensive collection of state-of-the-art visual transformer and MLP models designed for the PaddlePaddle framework. This project aims to provide a wide variety of cutting-edge vision models with full training and validation procedures, making it easier for researchers and practitioners to use. ### Key features: ### - **State-of-the-art Models**: PaddleViT includes a range of transformer models, visual attentions, and MLPs, ensuring the latest in visual transformer and MLP research. - **Easy-to-Use Tools**: The project offers modular utilities and tools, making it accessible for educators and practitioners alike. - **Customizability**: Users can modify models to fit their specific needs, and the models' files can be used independently for quick experiments. - **High Performance**: PaddleViT supports DDP (Data Parallelism) and mixed-precision training, enhancing performance. ### Tech Stack: ### Python ### Why it Matters: ### PaddleViT simplifies the process of implementing cutting-edge vision models, making it easier for researchers and practitioners to experiment and deploy models. The project's modular design and comprehensive documentation make it a valuable resource for anyone looking to leverage the latest in visual transformer and MLP research.

### Overview: rPPG-Toolbox is an open-source platform designed for camera-based physiological sensing, specifically remote photoplethysmography (rPPG). It benchmarks the state-of-the-art neural and unsupervised methods and supports the development of your own algorithms. ### Key features: - Benchmarks of existing state-of-the-art neural and unsupervised methods. - Supports flexible and rapid development of your own algorithms. - Implements traditional unsupervised algorithms such as Remote Plethysmographic Imaging, Advancements in noncontact multiparameter physiological measurements, and others. - Implements supervised neural algorithms like DeepPhys, PhysNet, TS-CAN, EfficientPhys, BigSmall, PhysFormer, iBVPNet, and PhysMamba. ### Tech stack: Python ### Why it matters: rPPG-Toolbox provides a comprehensive solution for researchers and developers interested in remote physiological sensing, enabling the benchmarking of state-of-the-art methods and facilitating the development of new algorithms.

### Overview: Welcome to our list of SDKs for AI agents. This database aims to provide a comprehensive collection of tools, frameworks, and libraries for creating, monitoring, debugging, and deploying autonomous AI agents. The list is curated based on our best knowledge and is not exhaustive. ### Key features: - SDKs for various programming languages (e.g., Python, JavaScript, Rust) - Frameworks for building AI agents (e.g., Chidori, Fixie) - Libraries for AI agent development (e.g., AI libraries, debugging tools) - Tools for monitoring and debugging AI agents - Cloud environments for deploying AI agents ### Tech stack: - Python - JavaScript - Rust - Node.js - Python - Rust - AI libraries - Debugging tools - Cloud environments ### Why it matters: This list is essential for developers and researchers looking to build, deploy, and monitor AI agents. It provides a centralized resource for finding the right tools and technologies to support their projects. The diverse range of features ensures that users can find the tools that best suit their needs, whether they are developing in Python, JavaScript, or Rust. The inclusion of cloud environments and AI libraries ensures that users can easily integrate their AI agent projects into existing workflows and environments.

### Overview: GOD is a real-time control room designed for agent societies, allowing users to govern, observe, and direct their own virtual worlds. It features a user-friendly interface and supports browser replays, enabling users to inspect and modify language-agent interactions. ### Key features: - **Browser Replays:** Watch and interact with simulations directly in the browser. - **Built-in Experiments:** Experiment packs, maps, and agent packs for various scenarios. - **No-Code Setup:** Simplified setup for quick and easy use. - **Windows Launcher:** A dedicated Windows launcher for easy installation. ### Tech stack: Python ### Why it matters: GOD provides a powerful tool for exploring and understanding language-agent interactions in agent societies, making it an essential resource for researchers and enthusiasts alike.

### Overview: HunyuanCustom is a multimodal-driven architecture designed for customized video generation. It supports image, audio, video, and text inputs to produce videos that are consistent with specific subjects under flexible conditions. ### Key Features: - **Multimodal Video Customization**: HunyuanCustom integrates various modalities to generate videos that are consistent with specific subjects. - **Various Applications**: The framework can be applied to a wide range of scenarios, including image-based, audio-based, and video-based customization. ### Tech Stack: - Python ### Why It Matters: HunyuanCustom provides a versatile solution for generating customized videos that are consistent with user-defined conditions, making it useful in various applications such as content creation, education, and more. ### Repository: - GitHub Repository: [Tencent-Hunyuan/HunyuanCustom](https://github.com/Tencent/HunyuanCustom) - GitHub Stars: 1231 ### Primary Language: - Python ### GitHub Description: - HunyuanCustom: A Multimodal-Driven Architecture for Customized Video Generation ### Open-Source Plan: - Supports single-subject video customization, audio-driven video customization, and video-driven video customization. - Inference and checkpoints are available for ComfyUI integration. ### Contents: - **HunyuanCustom**: Single-subject video customization, audio-driven video customization, and video-driven video customization. - **Abstract**: Aims to generate videos consistent with specific subjects under flexible conditions. - **Open-source Plan**: Supports image, audio, video

### Overview: This repository provides an introduction to reinforcement learning (RL) through comprehensive step-by-step tutorials. It covers various RL techniques, including Q-learning, SARSA, Deep Q-learning, Deep Convolutional Q-learning, DDPG, and more. The tutorials are designed to help developers understand and implement RL algorithms effectively. ### Key features: - Comprehensive step-by-step tutorials covering various RL techniques. - Walkthroughs on how to code different RL techniques. - Tutorials include: - Tutorial 1: Q-learning - Tutorial 2: SARSA - Tutorial 3: Exploring OpenAI gym - Tutorial 4: Q-learning in OpenAI gym - Tutorial 5: Deep Q-learning (DQN) - Tutorial 6: Deep Convolutional Q-learning - Tutorial 7: Reinforcement Learning with ROS and Gazebo - Tutorial 8: Reinforcement Learning in DOOM (unfinished) - Tutorial 9: Deep Deterministic Policy Gradients (DDPG) - Tutorial 10: Guided Policy Search (GPS) (unfinished) - Tutorial 11: A review of different AI techniques for RL (WIP) - Tutorial 12: Reviewing Policy Gradient methods - Tutorial 13: Continuous-state spaces with DQN (merged) - Tutorial 14: Benchmarking RL techniques - Tutorial 15: Reviewing Vanilla Policy Gradient (VPG) (failed miserably) ### Tech stack: Jupyter

Overview: BricksLLM is an enterprise-grade API gateway designed for monitoring and imposing cost or rate limits per API key. It provides fine-grained access control and monitoring capabilities, supporting multiple LLMs including OpenAI, Anthropic, Azure OpenAI, vLLM, and open-source models. Key features: - PII detection and masking - Rate limit - Cost control - Cost analytics - Request analytics - Caching - Request retries - Failover - Model access control - Endpoint access control - Native support for all OpenAI endpoints - Native support for Anthropic - Native support for Azure OpenAI - Native support for vLLM - Native support for Deepinfra - Support for custom deployments - Integration with custom models - Datadog integration - Logging with privacy control Tech stack: Go Why it matters: BricksLLM offers a robust solution for integrating LLMs into production environments, ensuring secure and efficient usage of these powerful AI models. Repository: bricks-cloud/BricksLLM GitHub stars: 1230

Overview: AdvancedEAST is an algorithm designed for scene image text detection, built upon the EAST text detection model. It significantly improves long text predictions, making them more accurate. Key features: - Enhanced accuracy in long text detection - Uses a Keras framework for easy readability and execution - Based on the EAST model, offering improved performance - Simplifies the training process and model setup Tech stack: Python Why it matters: The algorithm's improvements in long text detection make it highly useful for applications requiring accurate text recognition in scenes, particularly in situations where text spans across multiple images. Repository: huoyijie/AdvancedEAST GitHub stars: 1230 GitHub description: AdvancedEAST is an algorithm used for Scene image text detect, which is primarily based on EAST, and the significant improvement was also made, which make long text predictions more accurate.

Overview: Taranis AI is an advanced Open-Source Intelligence (OSINT) tool that leverages Artificial Intelligence to revolutionize information gathering and situational analysis. Key features: - Scours multiple data sources, including websites, for unstructured news articles. - Utilizes Artificial Intelligence and Natural Language Processing to enhance and enrich collected articles. - Streamlines the process for analysts to convert unstructured news into structured reports. - Provides persistent, per-user conversations for general questions or access to authorized stories. - Generates a variety of end products, including structured reports and PDF files. - Facilitates the effortless publication of intelligence products. Tech stack: Python Why it matters: Taranis AI offers a comprehensive and efficient solution for OSINT, leveraging AI to enhance content quality and streamline the data transformation process, making it a valuable tool for analysts and researchers.

### Overview: 整理一些书籍，涵盖C++、Java、Python、Scala、Linux、Git、Keras、TensorFlow、大数据、推荐系统、数据库、数据挖掘、机器学习、深度学习、算法等。 ### Key features: - C++和指针 - C语言库函数使用大全 - Effective C++第2版 - Thinking C++ - Java开发手册 - Java程序员修炼之道 - Java编程思想(第4版) - Python参考手册(第4版) - Python源码剖析 - 流畅的Python - 用Python写网络爬虫 - 精通Python设计模式 - TensorFlow实战Google深度学习框架 - TensorFlow Machine Learning Cookbook - TensorFlow实战_黄文坚 - 面向机器智能的TensorFlow实践 (智能系统与技术丛书) - Python深度学习 - Practical Python AI Projects - Machine Learning with TensorFlow - Practical Machine Learning with Python - Machine Learning Applications Using Python - Machine Learning with TensorFlow - Practical Machine Learning with Python - 机器学习(周志华) - 机器学习实战-中文版 - 机器学习导论 - 模式识别与机器学习(中文) - 深度学习21天实战Caffe - 百面机器学习 - 统计学习方法(李航) - 集体智慧编程中文版 - 神经网络与机器学习(Simon Haykin) - 神经网络与深度

### Overview: Kubetorch is a Python interface that magically distributes and runs ML workloads on Kubernetes, making it easy to develop, iterate, and deploy ML applications at any scale directly from Python. It provides a 100x faster iteration time and 50%+ cost savings compared to traditional methods. ### Key features: - **100x Faster Iteration**: From 10+ minutes to 1-3 seconds for complex ML applications. - **50%+ Compute Cost Savings**: Intelligent resource allocation, bin-packing, and dynamic scaling. - **95% Fewer Production Faults**: Built-in fault handling with programmatic error recovery and resource adjustment. ### Tech stack: Kubetorch is built with Python, leveraging Kubernetes for orchestration and deployment. ### Why it matters: Kubetorch simplifies the process of deploying ML applications on Kubernetes, making it accessible and efficient for developers and researchers alike. It enables fast iteration, cost savings, and improved fault management, making it a valuable tool for ML infrastructures on Kubernetes. ### GitHub stars: 1,228 ### Installation: - **Python Client**: `pip install "kubetorch[client]"` - **Kubernetes Deployment**: `helm upgrade --install kubetorch oci://ghcr.io/run-house/charts/kubetorch --version 0.5.0 -n kubetorch --create-namespace` ### Source Layout: The repository now includes customer-facing OSS deployment components that were previously split across internal and OSS repos. ###

### Overview: BPEmb is a collection of pre-trained subword embeddings in 275 languages, designed for use with neural models in natural language processing. It leverages Byte-Pair Encoding (BPE) to efficiently segment words into subwords, making it suitable for tasks such as language modeling, translation, and text classification. ### Key Features: - **Subword Segmentation**: BPEmb provides pre-trained models for various languages, including English, Chinese, and many others, which can be used to segment words into subwords. - **Embeddings and Vectors**: The embeddings are wrapped in a `gensim` KeyedVectors object, allowing them to be used in neural networks. The models can be accessed through the `encode`, `encode_ids`, and `embed` methods. - **Vocabulary Sizes**: Models can be customized with different vocabulary sizes, affecting the segmentation of words. Smaller vocabulary sizes result in more frequent words being segmented, while larger sizes yield fewer subwords. ### Tech Stack: Python ### Why It Matters: BPEmb is essential for anyone working on natural language processing tasks that require efficient and accurate subword segmentation. Its pre-trained models can significantly speed up training times and improve model performance by leveraging existing language knowledge.

Overview: QPython is an Android Python engine designed for Python and AI learners, offering a range of features to enhance the Python programming experience. Key features: - Python interpreter - Runtime environment - Editor - QPYI and integrated SL4A - Free and open-source project Tech stack: Java Why it matters: QPython simplifies Python programming on Android, making it accessible and easy for beginners, while also providing advanced features for experienced users. Repository: qpython-android/qpython GitHub stars: 1227

Overview: WifiForge is a terminal-driven framework designed for security researchers and pentesters to practice wireless attacks and network penetration. It provides a fully virtual WiFi environment without the need for physical hardware, allowing for controlled and legal experimentation. Key features: - Fully virtual WiFi networks using mininet-wifi - Access points, clients, and attackers all run in software - Runs on Kali Linux - Includes a menu-driven interface for selecting labs - Provides detailed walkthroughs and lab documentation Tech stack: Python Why it matters: WifiForge is essential for security professionals looking to enhance their skills in wireless hacking and penetration testing. It offers a controlled environment to practice and learn without the risk of real-world damage, making it a valuable tool for ethical hackers and pentesters. Repository: blackhillsinfosec/WifiForge GitHub stars: 1226

### Overview: This repository is a curated list of resources dedicated to Relation Extraction, a crucial task in Natural Language Processing (NLP). It provides a comprehensive overview of various supervised, distant supervision, GNN-based, language models, and knowledge graph-based approaches, as well as datasets, videos, and systems used in relation extraction research. ### Key Features: - **Datasets**: Includes datasets for relation extraction, such as CoNLL, CoNLL-2003, and others. - **Frameworks**: Provides frameworks and libraries for implementing relation extraction models. - **Videos and Lectures**: Offers educational content on relation extraction techniques. - **Systems**: Lists systems and models that have been used in relation extraction research. ### Tech Stack: The repository uses Markdown, HTML, and GitHub Pages for its content. ### Why It Matters: Relation extraction is a fundamental task in NLP that involves identifying and classifying relationships between entities in text. It plays a crucial role in applications such as information extraction, knowledge graph construction, and natural language understanding. This repository serves as a valuable resource for researchers, practitioners, and students interested in advancing the field of relation extraction.

**Overview:** NFStream is a Python framework designed for handling and analyzing network data, making it easy and intuitive for both online and offline data processing. It aims to be a fundamental building block for practical network flow data analysis and a unifying framework for researchers. **Key Features:** - Fast and flexible data structures - Supports both online and offline network data - Designed for real-world network flow data analysis - Provides data reproducibility across experiments **Tech Stack:** Python **Why it Matters:** NFStream simplifies the process of working with network data, making it accessible and efficient for developers and researchers alike. Its flexibility and expressiveness make it a valuable tool for anyone dealing with network data analysis. **Repository:** nfstream/nfstream **GitHub Stars:** 1224 **Primary Language:** Python **Latest Release:** Available on PyPI **Supported Versions:** Python 3 and PyPy 3 **Project Website:** [nfstream.org](https://www.nfstream.org/) **Discussion Channel:** [Gitter](https://gitter.im/nfstream/community) **Continuous Integration:** Linux, macOS, Windows **Code Quality:** High coverage, fuzzing, and code factor scores NFStream is a powerful framework for anyone looking to handle and analyze network data efficiently and effectively.

Overview: Truss simplifies the deployment and serving of AI/ML models in production environments, offering a streamlined approach that supports a wide range of frameworks and models. Key features: - Supports deployment to Baseten and other cloud infrastructures - Handles containerization, dependency management, and GPU configuration - Integrates with various Python frameworks and models - Provides a fast developer loop with live reload and no need for Docker or Kubernetes Tech stack: Python Why it matters: Truss streamlines the process of deploying AI/ML models, making it easier to develop, test, and deploy models without the complexities of traditional deployment methods.

### Overview: Free4Chat is an experimental system designed for temporary capability access and collaboration, allowing Humans and independently running Agents to exchange context, media, and tasks without the need for a permanent workspace. ### Key features: - **Temporary Rooms**: Short-lived access/collaboration boundaries for temporary human-AI collaboration. - **Participant-Own Capabilities**: Humans and Agents keep their intelligence, tools, credentials, and private memory. - **Low Friction**: A link or Room ID is sufficient to start, and no accounts or organizations are required. - **Thin Core**: Focuses on participants and bounded shared context, not a central platform. - **Progressive Collaboration**: Valid interactions between Humans, Humans and Agents, and Agents, but multiple Agents are not a goal. - **Cost-Aware Realtime**: Client/participant compute is preferred; high frequency data stays on the realtime data plane. ### Tech stack: TypeScript ### Why it matters: Free4Chat is an experimental system designed for temporary capability access and collaboration, allowing Humans and independently running Agents to exchange context, media, and tasks without the need for a permanent workspace.

Overview: RAGxplorer is an open-source tool designed to visualize Retrieval Augmented Generation (RAG) results. Key features: - Supports various embedding models for efficient RAG processing. - Provides a user-friendly interface for easy querying and visualization of RAG outputs. - Integrates seamlessly with Jupyter Notebooks and Streamlit for interactive analysis. Tech stack: Python, Jupyter Notebook, Streamlit, embedding models. Why it matters: RAGxplorer simplifies the process of visualizing RAG results, making it easier for developers and researchers to understand and interpret the insights generated by their models.