Projects

Overview: The SLING project aims to develop a natural language frame semantics parser for knowledge base completion, specifically for adding facts from Wikipedia to Wikidata. Key features: - Jointly trained model using bi-directional LSTMs for input encoding and TBRU for output decoding. - Efficient and scalable frame store implementation. - Neural network JIT compiler for fast training and parsing. Tech stack: C++ Why it matters: The SLING project is a crucial step towards enabling the automatic extraction of facts from text, which can significantly enhance the accuracy and completeness of knowledge bases like Wikidata.

### Overview: 花叔全部开源 Agent Skills 总目录：16 旗舰 + 14 人物视角 + 22 内置共 52 个 skill，分层分类 + AI Agent 安装协议 + 机器可读 skills.json + 更新检查机制。 ### Key features: - 旗舰Skills（21个独立仓库） - 人物视角Skills（14个独立仓库） - 内置Skills（22个，本仓库） - AI Agent 安装协议 - 机器可读 skills.json - 更新检查机制 ### Tech stack: Python ### Why it matters: 花叔在GitHub上全部开源Skill的总目录，包含16个旗舰系统、14个人物视角系统和22个内置技能，涵盖了从设计到数据分析、从原型制作到AI视频生成的广泛领域，为开发者提供了丰富的技能资源。

### Overview: kvcached is a virtualized elastic key-value cache library designed for efficient GPU sharing and utilization in dynamic workloads. By leveraging OS-style virtual memory abstraction, it enables flexible and demand-driven KV cache allocation, optimizing GPU memory usage under various conditions. ### Key Features: - **Elastic KV Cache**: Dynamically allocate and reclaim KV memory to match live load. - **GPU Virtual Memory**: Decouples logical KV from physical GPU memory via runtime mapping. - **Memory Control**: Enforces memory limits using the kvcached CLI or embedding revisioned limits in a production controller. - **Frontend Router and Sleep Mode**: Routes requests to the target models and puts models to sleep when idle. - **Support Mainstream Serving Engines**: Integrates with SGLang and vLLM. - **Prefix Caching**: Supports automatic prefix caching with a configurable memory bound. See the example doc for details. ### Tech Stack: Python ### Why It Matters: By enabling flexible and efficient GPU memory management, kvcached helps reduce GPU costs and improve overall system performance. Its ability to dynamically allocate and reclaim KV memory based on workload demands ensures optimal GPU utilization, making it a valuable tool for dynamic and mixed workload environments.

Overview: PySINDy is a package designed for system identification, specifically focusing on the Sparse Identification of Nonlinear Dynamical systems (SINDy) method introduced in Brunton et al. (2016a). It also includes other methods from related literature. Key features: - Fits sparse dynamical systems from data - Provides interpretable and generalizable models - Supports various optimization techniques (e.g., CVXPy, MIOSR, SBR) Tech stack: Python Why it matters: PySINDy enables researchers and practitioners to quickly identify and understand complex dynamical systems from observational data, making it invaluable for fields such as physics, engineering, and data science. Repository: dynamicslab/pysindy Primary language: Python GitHub stars: 1924 GitHub description: A package for the sparse identification of nonlinear dynamical systems from data

### Overview: BERTScore is an automatic evaluation metric designed to assess the quality of text generation models. It is based on the research paper "BERTScore: Evaluating Text Generation with BERT" (ICLR 2020), which aims to provide a more accurate and reliable evaluation metric compared to traditional methods. ### Key features: - **Supports Multiple Models**: BERTScore supports over 130 models, including DeBERTa, Roberta, BERTweet, and more. - **Correlation with Human Evaluation**: It correlates highly with human evaluations, providing a more accurate assessment of text generation models. - **Versatile Tokenizers**: Compatible with various tokenizers, including fast tokenizers for improved performance. - **Updated Versions**: Regular updates ensure the metric stays relevant and accurate, with features like support for new models and improved baseline files. ### Tech stack: Python ### Why it matters: BERTScore is essential for researchers and developers looking to evaluate the quality of text generation models. Its high correlation with human evaluations makes it a valuable tool for improving model performance and ensuring fair comparisons across different models. ### Repository: Tiiiger/bert_score ### Primary language: Jupyter Notebook ### GitHub stars: 1922 ### GitHub description: BERTScore is an automatic evaluation metric designed to assess the quality of text generation models. It is based on the research paper "BERTScore: Evaluating Text Generation with BERT" (ICLR 2020), which aims to provide a more accurate and reliable evaluation metric

### Overview: Vibe Coding Toolkit is a curated, battle-tested AI-coding toolkit designed to streamline the development process. It includes documentation, prompts, and templates for configuring machines and individual projects. The agent inspects the environment, organizes decisions, and configures components to ensure quality and context throughout the work. ### Key Features: - **Documentation and Prompts**: Comprehensive guides and prompts for setting up and configuring projects. - **Machine Setup**: Prepares machines for development and upgrades. - **Project Configuration**: Configures individual projects, improving and maintaining quality. - **Agents**: Integrates AI agents for efficient development tasks. ### Tech Stack: JavaScript ### Why It Matters: This toolkit provides a structured approach to machine setup and project configuration, ensuring that developers can focus on their work without worrying about the underlying infrastructure. The ready-to-copy prompts and quality gates help maintain high standards and reduce the risk of errors, making the toolkit valuable for both new projects and existing ones. ### Repository: - **Repository**: soumatheusgomes/vibe-coding-toolkit - **GitHub Stars**: 1922 - **GitHub Description**: A curated, battle-tested AI-coding toolkit: Claude Code plugins, subagent orchestration, quality gates, and ready-to-copy prompts, extracted from real production use. ### Community: - **Star History**: 1922 - **License**: MIT ### How to Use This Repository: - **Foundation**: Set up a foundation that fits your project. - **Agents**: Use AI agents for efficient development tasks. - **

### Overview: MiniOneRec is an open-source framework designed for scalable generative recommendation. It provides an end-to-end workflow that includes SID construction, supervised fine-tuning (SFT), and recommendation-oriented reinforcement learning (RL). ### Key Features: - **SID Construction:** Converts every product into a compact, semantically meaningful token by concatenating its title and description, then quantizing the embedding with a three-level RQ-VAE. - **SFT (Supervised Fine-Tuning):** Trains the model to generate the next product in a user's sequence of choices, using the user's chronological history as input. - **RL (Reinforcement Learning):** Encourages the model to make more effective recommendations by rewarding it for generating relevant and valuable items. ### Tech Stack: - Python ### Why It Matters: MiniOneRec offers a comprehensive solution for generative recommendation, making it an essential tool for developers looking to enhance their recommendation systems with a focus on scalability and effectiveness.

### Overview: MedicalZooPytorch is a PyTorch-based deep learning framework designed for multi-modal 2D/3D medical image segmentation. The project aims to provide an open-source library for state-of-the-art 3D deep neural networks in PyTorch, with a focus on 3D multi-modal brain MRI segmentation. ### Key Features: - **U-Net3D**: A dense volumetric segmentation model. - **V-Net**: A fully convolutional neural network for volumetric medical image segmentation. - **ResNet3D-VAE**: A 3D MRI brain tumor segmentation using auto-encoder regularization. - **U-Net**: A convolutional network for biomedical imaging. ### Tech Stack: - Python ### Why it Matters: MedicalZooPytorch aims to democratize medical image segmentation research by providing an open-source, reproducible framework. It supports a wide range of medical image datasets and datasets, making it a valuable tool for researchers and practitioners in the field of medical imaging. ### Repository: - **GitHub Repository**: black0017/MedicalZooPytorch - **Stars**: 1921 ### Quick Start: - **Blog Post**: A general overview of deep learning in medical imaging. - **Manual**: Detailed instructions for setting up the project. - **Colab Demo**: A quick start guide using Google Colab. ### Implemented Architectures: - **U-Net3D**: Dense volumetric segmentation. - **V-Net**: Fully convolutional network. - **ResNet3D

Overview: GPflow is a Python package designed for building Gaussian process models using TensorFlow 2.4+ and TensorFlow Probability. It provides modern Gaussian process inference capabilities. Key features: - Supports modern Gaussian process inference with composable kernels and likelihoods. - Built on TensorFlow 2.4+ and TensorFlow Probability for efficient GPU execution. - Available via PyPI and GitHub `develop` branch. Tech stack: Python, TensorFlow 2.4+, TensorFlow Probability. Why it matters: GPflow simplifies Gaussian process modeling, making it accessible for a wide range of applications in machine learning and scientific computing.

### Overview: BuildingAI is an enterprise-grade open-source platform designed for AI developers, entrepreneurs, and organizations. It provides a visual configuration interface for building native enterprise AI applications without coding. Key features include intelligent agents, knowledge bases, large-model aggregation, and context engineering, along with user registration, membership subscriptions, and compute billing. ### Key Features: - **AI Conversations**: Conversational AI and text generation powered by large language models, with support for multimodal models. - **AI Agents**: Create agents with memory, goals, and tool usage for autonomous task execution. - **Knowledge Base**: Build knowledge bases from documents with vector search and RAG-enhanced generation. - **MCP Integration**: Call MCP tools via SSE and Streamable HTTP protocols. - **Model Management**: Integrate mainstream large models under a unified API specification. - **Extension Mechanism**: Expand system capabilities and AI skills by installing extensions. - **Billing & Payments**: Built-in membership management, billing, and payment features ready to use. ### Tech Stack: TypeScript ### Why It Matters: BuildingAI offers a simplified and efficient way to build AI applications, making it accessible to everyone. It provides a comprehensive set of tools and features that developers can use to create intelligent agents, knowledge bases, and large-model integration, all without the need for complex coding.

Overview: pi_agent_rust is a high-performance AI coding agent CLI written in Rust, designed to replace slow, resource-intensive, and unreliable alternatives. It offers a native Rust implementation with zero unsafe code, making it a reliable and efficient solution for developers. Key features: - **Streamlined Startup**: No need for managed runtimes or heavy runtimes, ensuring a quick start. - **Resource Efficiency**: Rust's lightweight nature makes it more resource-efficient compared to Electron apps or heavy runtimes. - **Reliable Operation**: No sessions can break, and tools work reliably without silent failures. - **Extensible**: Built on Rust-native architecture, allowing for easy extension and customization. Tech stack: Rust 2024 edition, with a license of MIT + Rider, and no unsafe code. Why it matters: pi_agent_rust provides a native, Rust-based solution that addresses the shortcomings of existing AI coding tools, offering faster startup times, lower resource usage, and more reliable performance.

### Overview: SeekStorm is a high-performance in-process library and multi-tenancy server for vector and lexical search. It is built in Rust and has been in production since 2020, with over 1,900 GitHub stars. ### Key features: - **In-process library**: Optimized for sub-millisecond performance. - **Multi-tenancy server**: Supports multiple tenants with isolated data. - **Vector & lexical search**: Fast and efficient search capabilities. - **Rust implementation**: Built with Rust, ensuring reliability and performance. ### Tech stack: - **Rust**: The primary language used for development. ### Why it matters: SeekStorm is a crucial tool for developers looking to implement advanced search functionalities in their applications, particularly in environments requiring high performance and scalability. Its native vector and lexical search capabilities make it an ideal choice for applications that need to handle large datasets efficiently.

### Overview: This repository contains a collection of Jupyter Notebooks on Bayesian Machine Learning, focusing on various techniques such as Bayesian regression, Gaussian processes, sparse Gaussian processes, Bayesian optimization, Bayesian neural networks, and latent variable models. The notebooks are designed to provide a comprehensive understanding of Bayesian methods for machine learning. ### Key features: - Bayesian regression with linear basis function models - Gaussian processes for regression and classification - Sparse Gaussian processes using variational approaches - Bayesian optimization for hyper-parameter tuning - Bayesian neural networks with variational inference - Reliable uncertainty estimates for neural network predictions using noise contrastive priors ### Tech stack: Jupyter Notebook ### Why it matters: These notebooks are essential resources for developers and researchers interested in applying Bayesian methods to machine learning problems. They provide practical implementations and explanations of complex concepts, making it easier to understand and apply Bayesian techniques in real-world scenarios.

### Overview: Dot is a standalone, open-source application designed for seamless interaction with documents and files using local LLMs and Retrieval Augmented Generation (RAG). It is inspired by solutions like Nvidia's Chat with RTX, providing a user-friendly interface for those without a programming background. ### Key features: - Supports multiple document types: PDF, DOCX, PPTX, XLSX, and more. - Engages with Big Dot for inquiries not directly related to documents. - Built with Electron JS, using FAISS for local vector stores, Langchain for conversation chains, and additional tools for document management and interaction. - Uses the Phi-3.5 LLM by default. ### Tech stack: - JavaScript ### Why it matters: Dot simplifies the interaction with documents and files, making it accessible and easy to use for those without programming knowledge. It leverages local LLMs for a more intuitive and user-friendly experience. ### GitHub stars: 1,917 ### Repository: - **Repository**: alexpinel/Dot - **Primary language**: JavaScript - **GitHub stars**: 1,917 ### Visit the Dot Website: - **Visit the Dot Website**: [https://dotapp.uk/](https://dotapp.uk/) ### Installation: - **To use Dot**: - Visit the [Dot website](https://dotapp.uk/) to download the application for Apple Silicon or Windows. - Clone the repository: `$ https://github.com/alexpinel/Dot.git` - Install Node.js

Overview: fastdup is a fast and scalable tool designed to generate insights from image and video datasets, enhancing quality and reducing costs. Key features: - Rapid data analysis - Improved image and label quality - Cost-effective data processing - Unmatched scalability Tech stack: Python Why it matters: fastdup is a powerful, free tool that helps organizations enhance the quality of their image and video datasets while significantly reducing data operation costs, all with unmatched scalability.

### Overview: This repository contains pre-trained models for the spaCy Natural Language Processing (NLP) library, designed to enhance the capabilities of spaCy for various NLP tasks. ### Key Features: - **spaCy v3.x models directory**: Contains pre-trained models for spaCy v3.x. - **spaCy v3.x model comparison**: Provides a detailed comparison of spaCy v3.x models. - **spaCy v2.x models directory**: Contains pre-trained models for spaCy v2.x. - **spaCy v2.x model comparison**: Offers a comparison of spaCy v2.x models. - **Individual release notes**: Provides notes for each individual release. ### Tech Stack: Python ### Why It Matters: These models are essential for enhancing the performance and capabilities of spaCy, making it more versatile for a wide range of NLP tasks.

Overview: This repository is a curated list of awesome resources for speaker diarization, including papers, libraries, datasets, and other tools. Key features: - Papers - Libraries - Datasets - Evaluation - Clustering - Speaker embedding - Speaker change detection - Audio feature extraction - Audio data augmentation - Other software Tech stack: Python, TensorFlow, PyTorch, and other machine learning frameworks. Why it matters: This repository helps researchers, developers, and practitioners stay updated on the latest advancements in speaker diarization, enabling better understanding and implementation of state-of-the-art techniques.

### Overview: Diffgram is an AI Datastore designed for integrating Compliant PII, BLOBs, and Predictions into your applications. It offers built-in Human Supervision, Data Workflow, and a User Interface Catalog to maximize value from your AI data. ### Key Features: - **Human Supervision (Data Labeling)**: Label all media types and scale annotation. - **AI Data Application Workflow**: Move data between AI Apps and control AI through a friendly UI/UX. - **UI Catalog**: Visually explore and manage your AI Datastore. ### Tech Stack: Python ### Why it Matters: Diffgram simplifies the integration of AI data into applications, ensuring compliance and quality control. Its built-in human supervision and AI data workflow capabilities make it a valuable tool for businesses looking to leverage AI without compromising data integrity. ### Repository: - **GitHub**: [diffgram/diffgram](https://github.com/diffgram/diffgram) - **Stars**: 1,911 - **Description**: The AI Datastore for Schemas, BLOBs, and Predictions. Use with your apps or integrate built-in Human Supervision, Data Workflow, and UI Catalog to get the most value out of your AI Data. ### Tech Stack: - Python ### Why it Matters: Diffgram simplifies the integration of AI data into applications, ensuring compliance and quality control. Its built-in human supervision and AI data workflow capabilities make it a valuable tool for businesses looking to leverage AI without compromising data integrity.

Overview: Ollamac is a macOS app designed specifically for the Ollama model, providing a native and customizable experience. Key features: - Works with all Ollama models. - Customizable host. - Syntax highlighting. - Simple and easy to use. - Free and open source. Tech stack: Swift Why it matters: Ollamac offers a seamless and efficient way to interact with Ollama models directly from the macOS app, making it a valuable tool for developers and users alike. Repository: kevinhermawan/Ollamac Primary language: Swift GitHub stars: 1911 GitHub description: Mac app for Ollama

Overview: This project provides a generic U-Net implementation for image segmentation using TensorFlow. Key features: - TensorFlow-based U-Net implementation - Supports image segmentation on arbitrary imaging data - Includes usage documentation and example notebooks - Compatible with TensorFlow 1.x and 2.x Tech stack: Python Why it matters: This project is essential for researchers and developers working on image segmentation tasks, providing a versatile tool for various applications such as radio frequency interference mitigation and galaxy/star detection in wide-field imaging data.