Projects

### Overview: This repository focuses on building foundational visual models for large multimodal language models using large-scale datasets such as LAION400M and COYO700M. We employ sample-to-cluster contrastive learning to optimize performance. Our models are primarily used for multimodal visual large language models, such as LLaVA. ### Key features: - Sample-to-cluster contrastive learning - Optimized performance - Primarily used for multimodal visual large language models ### Tech stack: Python ### Why it matters: This repository is essential for developing foundational visual models that can be integrated into large multimodal language models, enhancing their capabilities in understanding and generating visual information alongside language.

### Overview: This repository implements the Voicebox model, a state-of-the-art Text-to-Speech (TTS) network, as described in the paper "Voicebox: A New SOTA Text-to-Speech Model from MetaAI". The model is implemented in Pytorch, making it accessible for developers. The project was funded by the Imminent Grant and sponsored by StabilityAI, and it has been used to train and sample TTS models. ### Key Features: - Pytorch implementation of the Voicebox model. - Uses adaptive normalization for time embedding. - Supports unconditional and conditional training. - Includes training code for SPEAR-TTS conditioned Voicebox. - Provides a simple installation guide using `pip install voicebox-pytorch`. ### Tech Stack: Pytorch ### Why it Matters: The project aims to provide a high-quality open-source implementation of the Voicebox model, which is a significant advancement in the field of text-to-speech technology. The use of Pytorch and the inclusion of training code make it particularly useful for developers looking to incorporate this model into their projects.

Overview: This repository is a curated list of top 100 trending deep learning GitHub repositories sorted by the number of stars gained on a specific day, with a focus on repositories with 40,000 stars or more. Key features: - Trending deep learning repositories sorted by stars - 703 GitHub stars - Python as the primary language - Regular updates Tech stack: Python Why it matters: This repository provides a valuable resource for developers looking to stay updated with the latest trends in deep learning, offering insights into the most active and impactful repositories in the field.

Overview: This repository contains development workflows for Claude Code, designed to keep exploration focused on an approved outcome, ensuring that designs and implementations align with the agreed-upon results. Key features: - **Repository-Specific Quality Checks**: Ensures that changes adhere to the specified quality standards. - **Design and Implementation Verification**: Verifies designs against the repository and commits before finalizing changes. - **Independent Review**: Independently reviews the finished implementation to ensure it meets the agreed outcome, does not include unnecessary changes, and has no serious issues. Tech stack: JavaScript Why it matters: This workflow helps maintain consistency and reliability in the development process, ensuring that changes are aligned with the agreed outcomes and that the implementation is robust and secure. Repository: shinpr/claude-code-workflows GitHub stars: 704

### Overview: Trinity is the operating system for the AI-native company that you own. It compounding intelligence your company owns, governed by receipts: git history, audit logs, human gates. Self-hosted and open source, it runs agents on Claude Code, OpenAI Codex, or Gemini CLI, organized into departments with roles and reporting lines, with our cognitive architecture, Cornelius, for making the work compound. ### Key features: - Persistent agents that keep their memory, share what they learn, and ship their own improvements. - Role-based governance with objectives tied to a number the company actually tracks. - Approvals in your Inbox - anything irreversible waits for a person, and the answer wakes the agent. - Gated skills, approvers by role, skill sets you assign as one unit. - Projects with tasks, a log, and a steward (Enterprise). - Rate the work, reply in-thread, one Skills tab, agents that pull their own repo. ### Tech stack: Python ### Why it matters: Trinity is the operating system for the AI-native company that you own, compounding intelligence your company owns. It runs agents on Claude Code, OpenAI Codex, or Gemini CLI, organized into departments with roles and reporting lines, with our cognitive architecture, Cornelius, for making the work compound.

Overview: VectorFlow is a high-volume vector embedding pipeline that ingests raw data, transforms it into vectors, and writes it to a vector database or returns it quickly and reliably. It's designed for use with Kubernetes in production. Key features: - Ingests raw data - Transforms data into vectors - Writes vectors to a vector database - Returns vectors quickly and reliably - Supports multiple file types (TXT, PDF, HTML, DOCX) Tech stack: Python Why it matters: VectorFlow is an open-source, high-throughput, fault-tolerant vector embedding pipeline that simplifies the process of processing and storing large volumes of raw data into vectors. Its simple API makes it easy to integrate with various vector databases and applications, making it a valuable tool for developers looking to improve their data processing and retrieval capabilities.

Overview: GPBoost is a software library that combines tree-boosting, Gaussian processes, and mixed-effects models (also known as latent Gaussian variable models). It allows for the integration of these techniques and provides a comprehensive set of tools for machine learning and statistical modeling. Key features: - Tree-boosting algorithms - Gaussian processes - Mixed-effects models Tech stack: C++ Why it matters: GPBoost offers a powerful and versatile toolset for machine learning and statistical modeling, combining the strengths of tree-boosting, Gaussian processes, and mixed-effects models to provide a robust framework for various applications.

Overview: Chat 是一个基于自然语言理解与机器学习的语义理解库，适用于从零开始搭建聊天机器人，减少开发工作量，并支持多用户并发及自定义多轮对话场景。 Key features: - 支持多用户并发 - 自定义多轮对话 - 丰富的语义分析工具与知识图构建工具 Tech stack: Python Why it matters: Chat 是一个快速搭建聊天机器人的工具，适用于从零开始的开发需求，减少工程师在实际开发中的重复工作，并提供多轮对话场景支持。

### Overview: MMRec is a modern toolbox designed for MultiModal Recommendation, simplifying research and implementation. It integrates 10+ models, including self-supervised collaborative filtering, layer-refined graph convolutional networks, and more. ### Key Features: - **SelfCF**: A simple framework for self-supervised collaborative filtering. - **LayerGCN**: A model that refines graph convolutional networks for recommendation. - **VBPR**: A model for visual Bayesian personalized ranking from implicit feedback. - **MMGCN**: A multi-modal graph convolutional network for personalized recommendation of micro-video. - **GRCN**: A graph-refined convolutional network for multimedia recommendation with implicit feedback. - **MVGAE**: A multi-modal variational graph auto-encoder for recommendation systems. - **DualGNN**: A dual graph neural network for multimedia recommendation. - **LATTICE**: A model for mining latent structures for multimedia recommendation. - **SLMRec**: A self-supervised learning model for multimedia recommendation. - **BM3**: A model for bootstrap latent representations for multi-modal recommendation. - **FREEDOM**: A model for freezing and denoising graph structures for multimodal recommendation. ### Tech Stack: Python ### Why it Matters: MMRec simplifies the research and implementation of MultiModal Recommendation, making it easier for developers to build and test models. It is a comprehensive toolbox that integrates 10+ models, including self-supervised collaborative filtering, layer-refined graph convolutional networks, and more.

### Overview: This roadmap provides a comprehensive guide for aspiring Full Stack ML engineers, covering essential Python programming, data analysis, visualization, statistics, and machine learning skills necessary for a successful career in machine learning. ### Key Features: - Python Basics - Data Analysis with NumPy and Pandas - Data Visualization with Matplotlib and Seaborn - Statistics for Machine Learning - Machine Learning Algorithms using Scikit-Learn ### Tech Stack: Python ### Why It Matters: This roadmap is crucial for anyone aiming to become proficient in all aspects of machine learning, ensuring a well-rounded understanding of the necessary skills and technologies.

Overview: Blackstone is a spaCy model and library designed for processing long-form, unstructured legal text, built by the Incorporated Council of Law Reporting for England and Wales research lab, ICLR&D. Key features: - Named-Entity Recogniser - Text categoriser - Pipeline - Abbreviation and long-form definition resolution - Compound case reference detections - Legislation linker - Sentence segmenter Tech stack: Python Why it matters: This project aims to democratize legal NLP research by making it available to the broader legal community, focusing on non-transactional legal text, which is often underrepresented in the current research landscape.

### Overview Your Obsidian vault, through the Local REST API plugin. Read, search, and write notes, edit a single heading, block, or frontmatter field in place, and manage tags, with folder-scoped read/write permissions built in. ### Key Features - **Read, Write, Search, and Edit Notes**: Access and modify your Obsidian vault notes, tags, and frontmatter fields. - **STDIO or Streamable HTTP Support**: Use either STDIO or Streamable HTTP for input/output operations. - **Folder-Scope Permissions**: Manage read/write permissions for notes within folders. - **Tool Integration**: Utilize the `obsidian_get_note`, `obsidian_list_notes`, `obsidian_list_tags`, `obsidian_search_notes`, `obsidian_write_note`, `obsidian_append_to_note`, `obsidian_patch_note`, `obsidian_replace_in_note`, `obsidian_manage_frontmatter`, `obsidian_manage_tags`, `obsidian_delete_note`, `obsidian_open_in_ui`, and `obsidian_list_commands` tools for enhanced functionality. ### Tech Stack - TypeScript ### Why It Matters This tool simplifies and enhances the Obsidian vault experience, providing powerful tools for managing and editing your notes efficiently.

### Overview: 这是一个全面整理的人工智能学习资料仓库，涵盖了机器学习、深度学习、计算机视觉、自然语言处理、语音识别、图神经网络、推荐系统以及算法工程师面试题等领域的经典入门资料。资料包括书籍、视频课程、代码示例和实战项目，旨在帮助开发者快速入门和深入学习人工智能技术。 ### Key features: - 机器学习基础：《PRML模式识别与机器学习》、《机器学习》、《机器学习实战》、《机器学习：概率学的观点 Machine Learning A Probabilistic Perspective》、《Maching Learning Yearning》、《机器学习》西瓜书 周志华、《统计学习方法》蓝宝书 李航、《机器学习实战 Hands on Maching Learning》、机器学习算法地图、斯坦福机器学习课程 CS229 吴恩达、《深度学习》花书英文版、《深度学习》花书中文版、《神经网络与深度学习》Michael Nielsen、《动手学深度学习》李沐、《Tensorflow实战Google深度学习框架》郑泽宇、斯坦福深度学习课程 CS230 吴恩达、台湾大学深度学习课程 李宏毅。 - 深度学习基础：《深度学习》花书英文版、《深度学习》花书中文版、《神经网络与深度学习》Michael Nielsen、《动手学深度学习》李沐、《神经网络与深度学习》邱锡鹏、《Tensorflow实战Google深度学习框架

### Overview: This project is a comprehensive implementation of the BiLSTM-CRF model using TensorFlow for sequence labeling tasks. It is designed to be highly scalable, configurable, modular, and easy to understand and customize. ### Key Features: - **Configurable Settings**: All settings can be configured, making it highly adaptable for different use cases. - **Modular Structure**: The project is organized into clear, modular components for easy integration and customization. - **Sequence Labeling Tasks**: Supports various sequence labeling tasks such as POS Tagging, Chunking, Named Entity Recognition (NER), and more. - **BiLSTM+CRF Model**: Implements the BiLSTM+CRF model proposed by Lample et al., making it a stable and classical model for sequential labeling tasks. - **TensorFlow API**: The model is built using TensorFlow, ensuring compatibility and scalability. - **Labeling Schemes**: Supports different label schemes such as BIO, BIESO, and others. - **Training and Testing**: Provides options for training, testing, and single predictions. - **Logging and API Service**: Detailed logging and an API service for easy demonstration and integration. - **Object-Oriented Design**: Uses classes like BILSTM_CRF, DataManager, Configer, and utils to ensure clear structure and ease of DIY. ### Tech Stack: - **Python**: 3.6 - **TensorFlow**: 1.10.0 - **Django**: 1.11.8 - **Jieba**: For Chinese word segmentation ### Why

### Overview: The **Convolutional Reconstruction Model (CRM)** is an efficient model designed to generate 3D textured meshes from single images in just 10 seconds. This project is part of the ECCV 2024 conference and showcases a feed-forward approach that significantly reduces the computational cost compared to traditional methods. ### Key Features: 1. **Single Image to 3D Textured Mesh**: CRM can convert a single image into a 3D textured mesh in a remarkably short time. 2. **Efficient Model**: The model is designed to be highly efficient, making it suitable for real-time applications. 3. **UV Texturing**: Unlike some online demos, CRM uses UV texturing instead of vertex color, providing better texture quality but with longer generation times due to UV texturing. 4. **Gradio Interface**: A visual interface is provided for easy inference, allowing users to interact with the model directly. 5. **Training Scripts**: Comprehensive training scripts are provided for both multiview generation and CCM generation, suitable for overfitting training. ### Tech Stack: - **Python**: The primary programming language used for developing the model and interface. - **Xformers**: Used for GPU acceleration, providing significant performance improvements. - **NVDiffrast**: Used for generating 6-view images and CCMs, offering better quality textures. ### Why It Matters: - **Efficiency**: CRM significantly reduces the time required to generate 3D textures, making it ideal for applications where real-time performance is crucial. - **Versatility**: The

### Overview: OpenClaw Agent Dashboard is a comprehensive, secure real-time monitoring dashboard designed for OpenClaw AI agents. It offers features such as session management, rate limit monitoring, cost analysis, live feed, memory viewer, file manager, system health tracking, service control, log viewer, cron management, tailscale integration, activity heatmap, streak tracking, dark/light theme, keyboard shortcuts, mobile responsiveness, browser notifications, timeline view, Git activity, Claude usage scraper, Gemini usage tracking, provider switching, per-model selector, auto-refresh, lifetime stats, health history, quick actions, macOS compatibility, system security dashboard, config editor, and Docker management. ### Key Features: - **Session Management**: View all agent sessions with real-time activity status. - **Rate Limit Monitoring**: Track Claude and Gemini API usage against rolling windows. - **Cost Analysis**: Detailed spending breakdowns by model, session, and time period. - **Live Feed**: Real-time stream of agent messages across all sessions. - **Memory Viewer**: Browse and read agent memory files (MEMORY.md, HEARTBEAT.md, daily notes). - **Files Manager**: View and edit workspace files, skills, and configs with security hardening. - **System Health**: CPU, RAM, disk, temperature monitoring with sparklines. - **Service Control**: Quick actions to restart OpenClaw, dashboard, or other services. - **Log Viewer**: Real-time system logs with auto-refresh. - **Cron Management**: View, enable/disable, and manually trigger cron jobs. - **Tailscale Integration**: View Tails

### Overview: Image augmentation software designed for object detection, segmentation, and classification. It offers a range of image enhancement techniques to improve model performance. ### Key Features: - **Meta Enhancement**: Enhances images to improve model performance. - **Arithmetic Enhancement**: Applies mathematical transformations to images. - **Artistic Enhancement**: Mimics artistic styles to enhance images. - **Blend Enhancement**: Combines multiple images to create a new, enhanced image. - **Blur Enhancement**: Reduces image details for better focus. - **Collections Enhancement**: Groups multiple images for better context. - **Color Enhancement**: Adjusts image colors to enhance details. - **Contrast Enhancement**: Increases the contrast of images for better visibility. - **Convolutional Enhancement**: Applies convolutional filters to images. - **Edges Enhancement**: Highlights edges to improve image clarity. ### Tech Stack: Python ### Why It Matters: This software is essential for researchers and developers working on object detection, segmentation, and classification tasks. It helps in improving model performance by enhancing images, making them more suitable for training and testing.

Overview: AnyTool is a universal tool-use layer designed to empower AI agents with fast, scalable, and powerful tool orchestration. It addresses three key challenges: overwhelming tool contexts, unreliable community tools, and limited capability coverage. Key features: - Lightning tool retrieval - Self-evolving tool orchestration - Universal tool automation Tech stack: Python Why it matters: AnyTool simplifies the integration of AI agents with tools, enabling them to interact more efficiently and effectively. Repository: HKUDS/AnyTool Primary language: Python GitHub stars: 698 GitHub description: "AnyTool: Universal Tool-Use Layer for AI Agents"

Overview: This repository contains the code for the experiments presented in the EMNLP 2021 paper "The Power of Scale for Parameter-Efficient Prompt Tuning" by Lester et al. Key features: The repository includes the original implementation of Prompt Tuning, using T5X, Flaxformer, Flax, and Jax. It provides detailed implementation details in the `implementation.md` file. Tech stack: The tech stack includes Python, T5X, Flaxformer, Flax, and Jax. Why it matters: This repository is essential for replicating the experiments and understanding the techniques used in the paper for parameter-efficient prompt tuning.

### Overview: Awesome Jev is a system-backed open-source ecosystem radar that provides a comprehensive and easy-to-navigate platform for discovering and managing projects. It offers a live radar feature, categorization, and automatic GitHub sync. The repository has 696 GitHub stars and is curated by the LogicRW team, focusing on projects related to system-1 decision architecture. ### Key Features: - **Live Radar**: A live dashboard for discovering and managing projects. - **Categories**: Organized into 17 categories for easy categorization. - **GitHub Sync**: Automatically syncs projects with GitHub repositories. ### Tech Stack: - JavaScript ### Why It Matters: TypeSafe Jev (System 1) is designed specifically for fast, typed discrete decisions, providing sub-second latency, native typed outputs, and zero vaporware. It includes 864+ projects and resources across 17 categories, making it a valuable resource for developers looking to improve their autonomous agent systems.