Projects

### Overview: Reinforcement Learning with Human Feedback (RLHF) is a rapidly developing area in artificial intelligence that aims to enable language models to align with complex human values through direct interaction with humans. This approach has the potential to revolutionize the field by addressing the limitations of traditional reinforcement learning methods and providing more accurate and efficient learning outcomes. ### Key Features: - **RLHF for Large Language Models (LLM)**: This feature explores the application of RLHF in large language models, such as ChatGPT, to align them with human values. - **RLHF for Video Games (e.g., Atari)**: This feature examines the use of RLHF in video games, particularly in the Atari domain, to improve the performance and alignment of AI agents with human feedback. ### Tech Stack: - Python - PyTorch - TensorFlow - OpenAI Gym - RLHF-specific libraries and frameworks ### Why It Matters: RLHF has the potential to significantly enhance the performance and alignment of AI systems with human values. By leveraging human feedback, it can help AI models learn more accurately and efficiently, leading to more effective and beneficial outcomes in various applications, such as language modeling and video game AI.

Overview: This repository is a curated list of papers and resources related to 3D reconstruction from images, focusing on various techniques such as Structure-from-Motion (SfM), Simultaneous Localization and Mapping (SLAM), Multiple View Geometry (MVS), and more. Key features: - Structure-from-Motion (SfM) tutorials and survey - Large-scale image matching for SfM - Incremental SfM, Global SfM, Hierarchical SfM, Multi-Stage SfM, Non Rigid SfM, Viewing graph optimization, Unordered feature tracking, Large scale image matching for SfM - Localization techniques in SfM reconstructions - Multiple View Geometry Library Solvers, MVS (Multiple View Stereovision), SLAM, Large Scale Image Retrieval, Minimization, Nearest Neighbor Search, Mesh storage processing - Feature detection and description - Datasets with ground truth for reproducible research - Open source software resources for SfM, MVS, SLAM, Large Scale Image Retrieval, Minimization, Nearest Neighbor Search, Mesh storage processing - Open source tutorials and survey on SLAM, SfM, and related topics Tech stack: The repository is built using Markdown, HTML, and GitHub Pages. Why it matters: This list provides a comprehensive overview of the latest research and techniques in 3D reconstruction from images, making it an essential resource for developers and researchers in the field.

Overview: This repository contains the source code for "Machine Learning with TensorFlow," a book that provides step-by-step explanations on using TensorFlow for machine learning. Key features: - Code for TensorFlow basics - Implementations of regression, classification, clustering, hidden markov models, autoencoders, reinforcement learning, convolutional neural networks, recurrent neural networks, and sequence-to-sequence models Tech stack: Jupyter Notebook Why it matters: This repository serves as a practical guide for developers interested in applying TensorFlow to machine learning tasks, offering detailed explanations and code examples from a book that is widely recognized in the field of machine learning.

### Overview: VoTT is a powerful open-source annotation and labeling tool designed for image and video assets. It provides a comprehensive end-to-end machine learning pipeline, enabling users to build and deploy object detection models from images and videos. ### Key Features: - **Labeling Images and Video Frames**: Easily tag images or video frames for machine learning. - **Extensible Model for Data Import and Export**: Supports importing and exporting data from local or cloud storage providers. - **Machine Learning Workflow**: Facilitates a complete machine learning pipeline, from data collection to model training and deployment. ### Tech Stack: - **TypeScript**: Written in TypeScript, the primary language for VoTT. ### Why It Matters: VoTT is essential for anyone working on image or video-based machine learning projects. It streamlines the annotation process, making it easier to create and deploy accurate models. The open-source nature of VoTT ensures that the community can contribute to and improve the tool, making it a valuable resource for developers and researchers alike. ### Repository: - **Repository**: [microsoft/VoTT](https://github.com/Microsoft/VoTT) - **GitHub Stars**: 4427 - **GitHub Description**: Visual Object Tagging Tool: An electron app for building end to end Object Detection Models from Images and Videos. ### Table of Contents: - **Getting Started** - Download and install a release package for your platform (recommended) - Build and run from source - Run as Web Application - **V1 & V2** - Where is V1

### Overview: A curated list of awesome data labeling tools, covering image, video, text, and audio annotation for various applications. ### Key features: - **LabelImg**: A graphical image annotation tool for labeling objects in images. - **CVAT**: A powerful and efficient Computer Vision Annotation Tool. - **Labelme**: Image Polygonal Annotation with Python. - **VoTT**: An open source annotation and labeling tool for image and video assets. - **imglab**: A web-based tool for labeling images for object detection. - **Yolo_mark**: GUI for marking bounded boxes of objects in images for training Yolo v3 and v2. - **PixelAnnotationTool**: Software for manually and quickly annotating images. - **OpenLabeling**: Label images and videos for Computer Vision applications. - **imagetagger**: An open source online platform for collaborative image labeling. - **Alturos.ImageAnnotation**: A collaborative tool for labeling image data. - **deeplabel**: A cross-platform image annotation tool for machine learning. - **MedTagger**: A collaborative framework for annotating medical datasets using crowdsourcing. - **Labelbox**: The fastest way to annotate data for building and shipping computer vision applications. - **turktool**: A modern React app for scalable bounding box annotation of images. - **Pixie**: A GUI annotation tool for bounding box, polygon, free drawing, and semantic segmentation. - **OpenLabeler**: An open source desktop application for annotating objects for AI applications. - **Anno-Mage**: A semi-automatic image annotation tool for

### Overview: GLM-4.5 is a foundational model designed to enhance agentic, reasoning, and coding capabilities, offering significant improvements in multilingual agentic coding, UI quality, tool usage, and complex reasoning tasks. ### Key Features: - **Core Coding**: GLM-4.5 improves multilingual agentic coding and terminal-based tasks, achieving notable gains in benchmarks such as SWE-bench, SWE-bench Multilingual, and Terminal Bench 2.0. - **Vibe Coding**: Enhances UI quality with cleaner and more modern webpages and better-looking slides. - **Tool Using**: Significantly improves performance in tool usage, particularly on benchmarks like τ^2-Bench and web browsing via BrowseComp. - **Complex Reasoning**: Boosts mathematical and reasoning capabilities, achieving substantial improvements on the HLE benchmark compared to GLM-4.6. ### Tech Stack: Python ### Why it Matters: GLM-4.5 is essential for developers looking to enhance their coding, reasoning, and agentic skills, making complex tasks more efficient and stable. Repository: zai-org/GLM-4.5 GitHub stars: 4424 GitHub description: GLM-4.5: Agentic, Reasoning, and Coding (ARC) Foundation Models

Overview: This repository contains a collection of 21 tensor puzzles designed to help developers practice and improve their understanding of PyTorch and NumPy. Key features: - 21 tensor puzzles - Practice reimplementing functions from the NumPy standard library without using magic - Simplified environment for learning tensor programming - Colab notebook for running the puzzles - YouTube walkthrough available for interested users Tech stack: Jupyter Notebook Why it matters: These puzzles are designed to help developers practice and improve their understanding of tensor programming languages like PyTorch and NumPy, focusing on the expressive power of these languages and the importance of understanding broadcasting rules. Repository: srush/Tensor-Puzzles GitHub stars: 4420 GitHub description: Solve puzzles. Improve your pytorch.

### Overview: This repository contains plain Python implementations of basic machine learning algorithms, focusing on providing a basic understanding of the algorithms and their underlying structure without relying on external libraries. The primary language used is Jupyter Notebook, with implementations for various algorithms such as Bayesian Linear Regression, Decision Trees, k-Nearest Neighbors, k-Means Clustering, Linear Regression, Logistic Regression, Multinomial Logistic Regression, Perceptron, Principal Component Analysis, Simple Neural Network, Softmax Regression, Support Vector Machines, and more. ### Key features: - Bayesian Linear Regression - Decision tree for classification - Decision tree for regression - k-nearest-neighbor - k-Means clustering - Linear Regression - Logistic Regression - Multinomial Logistic Regression - Perceptron - Principal Component Analysis - Simple neural network with one hidden layer - Softmax regression - Support vector machines ### Tech stack: Jupyter Notebook ### Why it matters: These notebooks aim to provide a basic understanding of the algorithms and their underlying structure without focusing on efficiency, making them valuable resources for developers looking to gain a fundamental knowledge of machine learning concepts.

### Overview: 从零开始构建大语言模型，深入了解LLM原理和架构。 ### Key features: - **从头开始构建LLM**：从基础理论到实际编码的系统化学习路径。 - **实践导向**：通过实际操作掌握LLM的开发和训练。 - **重点关注LLM架构**：微调、部署相关教程较为丰富的背景下，着重关注大模型的架构实现。 ### Tech stack: Python, Jupyter Notebook ### Why it matters: 通过本项目，开发者和研究者可以掌握创建大语言模型和大语言模型架构的核心技术，从而深入理解大型语言模型的工作原理。

Overview: franc is a powerful natural language detection library that supports over 400 languages, making it highly versatile and useful for applications requiring language identification. Key features: - Supports multiple languages, including 82, 186, and 419 languages. - Provides a CLI for easy integration. - Detects the language of text efficiently. Tech stack: JavaScript Why it matters: franc is highly versatile, supporting a wide range of languages, making it suitable for applications that need to identify the language of text. Its support for multiple languages and easy CLI integration make it a valuable tool for developers working with international content.

Overview: This repository contains Python code for selected tables, figures, and LAB sections from the book "An Introduction to Statistical Learning with Applications in R" by James, Witten, Hastie, and Tibshirani (2013), focusing on the first edition. The code is based on the book's public datasets and includes applications in R. Key features: - Python code for statistical learning models and methods - Implementation of various machine learning algorithms using libraries such as scikit-learn, python-glmnet, and statsmodels - Reproducible and accessible code for educational purposes Tech stack: Python, pandas, numpy, scipy, scikit-learn, python-glmnet, statsmodels, patsy, matplotlib, seaborn Why it matters: This repository provides a practical and accessible way to learn and understand the concepts presented in the book "An Introduction to Statistical Learning with Applications in R" through Python code. It serves as a valuable resource for students and practitioners looking to deepen their understanding of statistical learning techniques and their implementation in Python.

Overview: ONNX Simplifier simplifies complex ONNX models, making them easier to understand and use. Key features: - Constant folding: Evaluates constant parts of the graph and replaces redundant operators with their computed outputs. - Graph optimization: Runs onnx-optimizer's fusions and eliminations to simplify the model. - Shape inference: Propagates tensor shapes through the graph to unlock more folding. - Correctness checking: Validates the simplified model against the original. - Fixed and dynamic input shapes: Pin shapes for simplification or check correctness. - Custom operators: Keeps custom ops unchanged and automatically registers schemas. - Opset conversion: Upgrades or downgrades the model's opset while simplifying. - Function inlining: Flattens local functions before simplifying the model. Tech stack: Python Why it matters: Simplifying complex ONNX models makes them easier to understand and use, reducing the risk of errors and improving performance.

### Overview: FLAML is a fast library for Automated Machine Learning and Hyperparameter Tuning, designed to automate and optimize machine learning workflows efficiently. It supports various tasks such as classification, regression, and more, leveraging large language models and machine learning models. FLAML is lightweight and easy to customize, making it suitable for both beginners and advanced users. ### Key Features: - **Efficient Workflow Automation**: FLAML automates the process of model selection and hyperparameter optimization under resource constraints. - **Model Selection and Tuning**: It finds high-quality models quickly using low computational resources, making it ideal for tasks like classification and regression. - **Customizability**: Users can easily customize or extend FLAML to meet their specific needs. - **Integration with MLflow**: FLAML integrates seamlessly with MLflow, a popular platform for machine learning experiments, making it easier to track and manage experiments. ### Tech Stack: - Python 3.10+ and < 3.14 ### Why It Matters: FLAML is a powerful tool for automating machine learning and AI operations, making it easier and faster to build and optimize models. Its lightweight nature and ease of customization make it a valuable addition to any developer's toolkit. ### Repository: - **GitHub Repository**: [microsoft/FLAML](https://github.com/microsoft/FLAML) - **Stars**: 4,413 - **Description**: A fast library for AutoML and tuning. Join our Discord: https://discord.gg/Cppx2vSPVP. ### Installation: - **Python Requirements

Overview: Cognita is a modular, open-source framework for building production-ready applications using Langchain/LlamaIndex, designed to be easily scalable and extendable. Key features: - Supports Langchain/LlamaIndex abstractions for quick experimentation - Provides a modular, API-driven organization for RAG components - Offers a production-ready environment with no-code UI support - Supports incremental indexing - Introduces model gateway for managing models and configurations - Supports metadatastore powered by Prisma and Postgres - Adds support for AudioParser, VideoParser, and Reranking - Offers one-click local deployment - Supports multi-modal vision parsing with GPT-4 Tech stack: Python Why it matters: Cognita simplifies the process of building production-ready applications by providing a modular, scalable, and easy-to-use framework that integrates Langchain/LlamaIndex and supports various indexing and querying functionalities. Repository: truefoundry/cognita GitHub stars: 4412 GitHub description: RAG (Retrieval Augmented Generation) Framework for building modular, open source applications for production by TrueFoundry

Overview: Yellowbrick is a suite of visual diagnostic tools designed to help facilitate the machine learning model selection process. It integrates scikit-learn with matplotlib, providing visualizations that guide the model selection workflow. Key features: - Feature Visualization: Displays pairwise comparisons of features with a specific metric. - Model Visualization: Integrates scikit-learn models with Yellowbrick visualizers. Tech stack: Python, scikit-learn, matplotlib. Why it matters: Yellowbrick simplifies the model selection process by providing intuitive visualizations that help guide the user through the selection of machine learning models, making the process more accessible and less intimidating.

Overview: This repository contains papers on computational advertising, including optimization methods, topic models, and other relevant resources. It aims to provide insights and practical knowledge for students and professionals in the field of computational advertising. Key features: - Papers on optimization methods, such as online optimization, parallel stochastic gradient descent, and regularized convex optimization. - Papers on topic models, including probabilistic latent semantic analysis (PLSA), latent Dirichlet allocation (LDA), and distributed representations of words. - Papers on deep interest evolution network for click-through rate prediction. Tech stack: Python Why it matters: This repository serves as a valuable resource for anyone interested in computational advertising, offering insights into optimization techniques and topic modeling methods that are essential for understanding and improving advertising algorithms. Repository: wzhe06/Ad-papers Primary language: Python GitHub stars: 4407 GitHub description: Papers on Computational Advertising

### Overview: This repository contains detailed notes, programming assignments, and quizzes from all courses within the Coursera Deep Learning specialization offered by deeplearning.ai. The specialization covers topics such as neural networks, deep learning, hyperparameter tuning, regularization, optimization, and convolutional neural networks. ### Key Features: - Detailed programming assignments and quizzes for each course. - Notes and diagrams taken directly from the Coursera Deep Learning Specialization. - Updated for the 2021 version, including developments in deep learning and programming frameworks. - Includes pre-trained VGG-19 dataset and model extraction for all assignments. ### Tech Stack: Jupyter Notebook ### Why It Matters: This repository serves as a comprehensive resource for anyone looking to enhance their knowledge of deep learning and machine learning, providing practical insights and hands-on experience through detailed programming assignments and quizzes.

### Overview: This repository provides a comprehensive Data Science roadmap, covering essential topics from beginner to advanced levels. It includes resources for understanding the field, learning programming languages, and mastering data analysis techniques. ### Key Features: - **Data Science Roadmap Overview**: A detailed guide to the Data Science field. - **Programming Languages**: Python, R, and SQL for data analysis. - **Data Science Tools**: Jupyter Notebook, Google Colab, and more. - **Projects and Applications**: Real-world projects and case studies. - **Soft Skills**: Importance of communication, critical thinking, and problem-solving. ### Tech Stack: - Python - R - SQL - Jupyter Notebook - Google Colab ### Why It Matters: This roadmap is essential for anyone interested in breaking into the field of Data Science, providing a structured approach to learning and applying data analysis techniques. It covers the basics of data science, including programming languages, tools, and real-world applications, making it a valuable resource for beginners and advanced learners alike.

Overview: MCP Server Chart is a TypeScript-based Model Context Protocol server designed for chart generation and data analysis. It supports 26+ charts, including area, bar, boxplot, column, district map, dual axes, fishbone diagram, flowchart, funnel chart, histogram, and line charts. Key features: - 📊 Generate various types of charts using MCP tools - 📊 Supports 26+ charts, including area, bar, boxplot, column, district map, dual axes, fishbone diagram, flowchart, funnel chart, histogram, and line charts - 📈 Data analysis capabilities Tech stack: TypeScript Why it matters: MCP Server Chart simplifies the process of creating and analyzing charts, making it an essential tool for data visualization and analysis.

Overview: Dagu is a self-hostable workflow orchestrator designed for teams where the primary focus is not orchestration. It uses a declarative YAML format to define workflows, allowing scripts, containers, and remote commands to be integrated seamlessly without modifications. Key features: - Declarative YAML for workflow definition - Runs on Linux, macOS, and Windows - Supports shell commands, Docker containers, Kubernetes Jobs, and SSH commands - Native support for cron syntax, timezones, and retry policies - Runs where data and credentials are located, on-prem, air-gapped, edge, or cloud - Composes reusable Sub-DAGs and runs work in parallel with concurrency controls - Schedules workflows with cron syntax, timezones, overlap policies, and catch-up windows - Logs, run history, retries, notifications, and webhook triggers in one place - Built-in MCP server for inspecting workflows and runs, maintaining Wiki pages, applying changes, and controlling runs - Optional monitoring in Dagu Console: every server's runs and trends in one place, and an email when a workflow fails or a server stops reporting - Runs without an external database or message broker Tech stack: Go Why it matters: Dagu simplifies the process of scheduling and managing workflows, allowing teams to focus on their core operations without the overhead of orchestration systems. It eliminates the need for a separate orchestrator, making it more efficient and easier to maintain. Repository: dagucloud/dagu GitHub stars: 4398 GitHub description: Self-hostable workflow