Projects

### Overview: This repository, "Awesome-KV-Cache-Compression," serves as a curated collection of papers and resources on key topics in KV cache compression for large language models. It aims to provide developers with the latest insights and methodologies to optimize the performance and efficiency of these models. ### Key Features: - **Pruning / Evicting / Sparse**: - **Scissorhands**: Exploiting the Persistence of Importance Hypothesis for LLM KV Cache Compression at Test Time. - **SnapKV**: LLM Knows What You are Looking for Before Generation. - **H2O**: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models. - **Tech Stack**: - Python, Rust, and various transformers libraries. - **Why it Matters**: This repository offers valuable resources and methodologies for developers looking to improve the performance and efficiency of large language models by optimizing KV cache compression techniques. ### Tech Stack: Python, Rust, transformers libraries. ### Why it Matters: This repository provides insights and methodologies for optimizing KV cache compression in large language models, making it a valuable resource for developers aiming to enhance the performance and efficiency of their models.

Overview: Paella is a novel text-to-image model that requires only 10 steps to generate high-fidelity images, sampling in less than 500 ms with 573M parameters. It operates on a compressed latent space and uses improved sampling functions over previous works, making it accessible to a broader audience. Key features: - Paella is a text-to-image model that requires fewer inference steps than most state-of-the-art models. - It operates on a compressed and quantized latent space, making it faster and more efficient. - Paella can generate images with CLIP embeddings and improved sampling functions. - It supports latent space interpolation and image manipulations such as inpainting, outpainting, and structural editing. Tech stack: Paella is built using Jupyter Notebook, making it easy to understand and modify. Why it matters: Paella is designed to be more accessible to non-experts in the field of generative AI, making it easier for people outside the field to understand and use text-to-image models. The minimalistic training and sampling code, available in just 12 lines of code, allows for quick extensions, tests, and idea testing, making further developments and improvements more straightforward.

Overview: This package allows developers to integrate Stanford NLP models directly into the spaCy pipeline, leveraging the top-performing models from the CoNLL 2017 and 2018 shared tasks. These models are known for their high accuracy across 68 languages. Key features: - Tokenization - Lemmatization - Part-of-speech tagging - Morphological analysis - Dependency parsing - Named entity recognition Tech stack: Python Why it matters: This package simplifies the integration of Stanford NLP models into spaCy, making it easier for developers to leverage advanced language processing capabilities without needing to install and manage multiple libraries. Repository: explosion/spacy-stanza Primary language: Python GitHub stars: 748 GitHub description: 💥 Use the latest Stanza (StanfordNLP) research models directly in spaCy GitHub actions badge: [](https://github.com/explosion/spacy-stanza/actions/workflows/tests.yml) GitHub PyPi badge: [](https://pypi.python.org/pypi/spacy-stanza) GitHub GitHub badge: [](https://github.com/explosion/spacy-stanza) GitHub Code style: black badge: [![Code style

Overview: Clawe is a multi-agent coordination system powered by OpenClaw, designed to streamline the work of AI agents through a Kanban-style task management interface. Key features: - Multiple AI agents with distinct roles and personalities - Agents wake up on cron schedules to check for work - Kanban-style task management with assignments and subtasks - Instant delivery of @mentions and task updates - Agents collaborate through shared files and Convex backend - Monitor squad status, tasks, and chat with agents from a web dashboard Tech stack: TypeScript Why it matters: Clawe simplifies the coordination and communication among AI agents, making complex tasks manageable and ensuring timely updates and notifications. The Kanban-style interface and real-time collaboration features make it an ideal solution for teams working with multiple agents, enhancing productivity and efficiency.

### Overview: This repository is a curated list of Text-to-Video (T2V) products, models, and tools. It covers various aspects of the field, including commercial platforms, open-source models, research papers, and datasets. ### Key Features: - **Native 4K and Synchronized Audio**: Products now offer native 4K output and synchronized audio, making the experience more immersive. - **Open-Source Models**: Open-source models like Wan 2.2, HunyuanVideo 1.5, and LTX-2.3 are now competitive with closed commercial systems. - **Multi-Shot Storytelling and Character Consistency**: The landscape has shifted towards multi-shot storytelling and character consistency, becoming the new battlegrounds. - **Inference Acceleration**: Techniques like sliding tile attention, token carving, and caching have made local generation feasible, making it faster and more efficient. ### Tech Stack: Python ### Why it Matters: This repository is crucial for developers and creators looking to integrate Text-to-Video technology into their projects. It provides a comprehensive overview of the current landscape, helping users make informed decisions about the best tools and models for their needs. ### Repository: jianzhnie/awesome-text-to-video ### GitHub Stars: 746 ### GitHub Description: A Survey on Text-to-Video Generation/Synthesis.

Overview: 这个项目是地球上最全面的华语现代诗歌语料库，收录了3489位诗人和80000多首诗歌，总字数超过1543万字。 Key features: - 丰富的诗人资源 - 大量的诗歌作品 - 超过1543万字的诗歌文本 Tech stack: Python Why it matters: 该项目为现代诗歌研究和爱好者提供了丰富的资源，有助于深入研究华语现代诗歌的多样性和发展。 Repository: sheepzh/poetry GitHub stars: 748 GitHub description: 地球上最全的华语现代诗歌语料库，3k+诗人，80K+诗歌，15M+字 README excerpt:

### Overview: UltimateALPR is a fast and accurate Automatic Number/License Plate Recognition (ANPR/ALPR) implementation that runs on CPUs, GPUs, VPUs, and NPUs using Transformers (ONNX, Tensorflow lite, TensorRT, Hailo, OpenVINO, RKNN2). It supports multiple character sets (Latin, Korean, Chinese) and operating systems (Jetson, Android, Raspberry Pi, Linux, Windows), as well as various architectures (ARM, x86). ### Key Features: - **Fast Processing**: Runs at up to 64 FPS on a $99 ARM device, making it suitable for low-cost devices. - **High Accuracy**: #33 times faster than OpenALPR on Android. - **SIMD ARM NEON Optimizations**: Utilizes fixed-point math operations and multithreading for efficient performance. - **Versatile Hardware Support**: Accelerated on CPUs, GPUs, VPUs, and FPGAs via CUDA, TensorRT, and OpenVINO. - **Multi-Charset Support**: Latin, Korean, Chinese character sets. - **Multi-OS Support**: Android, Raspberry Pi, Linux, Windows, and others. - **Multi-Arch Support**: ARM, x86 architectures. - **Support for Multiple Languages**: C++, C#, Java, and Python APIs. ### Tech Stack: - **Hardware Acceleration**: CUDA, NVIDIA TensorRT, Intel OpenVINO. - **Deep Learning Techniques**: State-of-the-art deep learning for ANPR/ALPR. ### Why It Matters:

### Overview: AnythingMCP is a self-hosted MCP gateway that converts any REST, OpenAPI, SOAP, GraphQL, OData, or SQL system into MCP tools for Claude, ChatGPT, and Copilot. It includes 547 ready-made connectors, including SAP, Etsy, weclapp, and Amazon Seller, with 26 of them requiring no API keys. ### Key Features: - **547 Ready Connectors:** 547 pre-configured connectors for various systems like SAP, Etsy, weclapp, and Amazon Seller. - **No API Keys Needed:** 26 connectors require no API keys. - **Self-Hosted Gateway:** Runs on any server, including macOS, without the need to clone the repository. - **Open Source, AGPL-3.0 License:** Available on GitHub and Docker Hub. ### Tech Stack: - **TypeScript:** The main programming language used. - **Docker:** Used for containerization and deployment. ### Why It Matters: - **Versatility:** Converts any system into a MCP-compatible system, making it accessible to a wide range of developers. - **Security:** No need to write an MCP server, reducing the risk of security vulnerabilities. - **Ease of Use:** The quickstart script ensures a smooth setup process, making it easy to get started quickly. ### GitHub Stars: - **GitHub Stars:** 740, indicating a significant community interest and adoption of the project. ### Tech Stack: - **TypeScript:** The primary programming language used. - **Docker:**

### Overview: OpenClaw is a curated list of resources for the OpenClaw ecosystem, including official projects, skills, dashboards, deployment tools, memory systems, and guides. This repository aims to provide a comprehensive overview of the various components and tools used in the OpenClaw ecosystem. ### Key Features: - **Official Projects**: The core project, official website, documentation, and official workflow shell. - **Skills Registries**: Skills listed separately. - **Dashboards & Control Centers**: Official workflows and resumable automations. - **Deployment & Operations**: Official Windows companion suite and macOS screenshot CLI. - **Models & Providers**: Official TypeScript API and CLI. - **Memory & Context Systems**: Official headless CLI client and sandbox helper. - **Security & Safety**: Official WhatsApp CLI. - **Plugins & Channel Integrations**: Official Apple Messages.app CLI. - **Alternative Clients & Runtimes**: Official headless CLI client and sandbox helper. - **Learning Resources**: Official documentation and skills. ### Tech Stack: OpenClaw. ### Why it Matters: This repository serves as a go-to resource for developers and enthusiasts looking to explore and utilize the OpenClaw ecosystem. It provides a wealth of information on projects, skills, dashboards, deployment tools, memory systems, and more, making it an invaluable resource for anyone interested in the OpenClaw ecosystem.

### Overview: Best of AI is a curated directory of the best AI tools across 143 categories, selected for quality, practicality, and real-world impact. ### Key features: - **Quality over quantity**: Every tool is reviewed and ranked, not just scraped from a list. - **Practical focus**: Tools are selected for real-world usefulness, not hype. - **Broad coverage**: 143 categories from Code Assistants to Video to Healthcare. - **Always updated**: New tools added regularly as the AI landscape evolves. - **Browse by profession**: Curated tool stacks for designers, developers, marketers, and more. ### Tech stack: HTML ### Why it matters: Best of AI is a curated, community-driven directory of the best AI tools across 143 categories, selected for quality, practicality, and real-world impact.

### Overview: This repository is an awesome list of summer schools on Artificial Intelligence, Machine Learning, and Healthcare. It serves as a resource for developers interested in staying updated with the latest advancements in these fields. ### Key Features: - **Summer Schools in 2027:** - [8th International Artificial Intelligence Summer School – IAISS 2027](https://2027.iaiss.cc/) - [10th Advanced Course on Data Science & Machine Learning – ACDL 2027](https://acdl2027.icas.events/) - [2nd International School on Human-AI Collaboration in the Era of Agentic AI (AICES 2027)](https://aices.irdta.eu/2027/) - [Chemistry and Materials Machine Learning School 2027 (CAMML 2027)](https://spring2027.camml.ac.uk/about/) - [International Artificial Intelligence Winter School – IAIWS 2027](https://2027.iaiws.cc/) - [DeMythif.AI Winter School 2027 – DMWS 2027](https://www.dataia.eu/en/formations/demythifai-winter-school-2026) - [Göttingen Emergent Minds (GoEMMI) 2027 Winter School](https://events.gwdg.de/event/1440/)

Overview: Swarm is a multi-agent swarm coordination system designed to handle complex tasks in OpenCode, integrating agent issue tracking and learning capabilities. Key features: - Decomposes tasks into parallelizable subtasks. - Creates cells in the Hive (git-backed task tracker) for parallel execution. - Spawns agents with file reservations to avoid conflicts. - Coordinates tasks via Swarm Mail for event-based communication. - Reviews task completions before approval. - Learns from past task executions for improved future performance. Tech stack: TypeScript. Why it matters: Swarm simplifies the coordination of complex tasks by breaking them down into manageable subtasks, creating parallel execution cells, and leveraging learning capabilities to improve future performance. Repository: joelhooks/swarm-tools GitHub stars: 748 GitHub description: 🐝 Multi-agent swarm coordination for OpenCode with learning capabilities, agent issue tracking, and management.

### Overview: Octopii is an AI-powered Personal Identifiable Information (PII) scanner designed to detect and extract sensitive information such as Government IDs, addresses, emails, and phone numbers from public images, PDFs, and documents. This tool is crucial in identifying and mitigating the risk of PII leaks in cybersecurity practices. ### Key Features: - **Scanning Methods**: Utilizes Optical Character Recognition (OCR) and Natural Language Processing (NLP) to detect PII from images and documents. - **Supports**: S3 URLs, Apache open directory listings, and local filesystems. - **Output**: Generates a JSON file containing the detected PII information, including file paths, country of origin, faces, identifiers, emails, phone numbers, and addresses. ### Tech Stack: - **OCR**: Tesseract OCR - **NLP**: NLTK - **Dependencies**: `pip install -r requirements.txt`, `sudo apt install tesseract-ocr -y` on Ubuntu, `sudo pacman -Syu tesseract` on Arch Linux - **Python**: The primary language used for development. ### Why It Matters: The detection of PII leaks is often overlooked in cybersecurity practices, making Octopii a valuable tool for organizations and services to identify and mitigate these risks. By automating the process of discovering and extracting PII, Octopii helps in ensuring the security and privacy of sensitive information. ### Repository: - **GitHub Repository**: redhuntlabs/Octopii - **GitHub Stars**: 748 - **Primary Language**: Python

### Overview: Affitor's affiliate-skills repository is a comprehensive collection of 52 AI-powered affiliate marketing skills, designed to help marketers optimize their affiliate campaigns with any AI agent. These skills cover various stages of the affiliate marketing funnel, from research and content creation to landing page optimization and automation. ### Key Features: - **Research Trending Content**: Identify the latest affiliate marketing trends and content formats. - **Write Data-Backed Posts**: Generate compelling and data-driven blog posts and articles. - **Generate Infographics**: Create visually appealing infographics to enhance website traffic and conversions. - **Build Landing Pages**: Automate the creation of effective landing pages for affiliate programs. - **Deploy Affiliate Programs**: Scale and automate the deployment of affiliate programs. - **Track and Optimize**: Monitor and optimize the performance of affiliate programs using analytics. - **Scale and Automate**: Implement automation tools to streamline the entire affiliate marketing process. ### Tech Stack: The repository utilizes various AI agents such as Claude Code, Pi, ChatGPT, Gemini CLI, Cursor, Windsurf, and any AI that reads text. ### Why It Matters: With the affiliate-skills repository, marketers can leverage the power of AI to enhance their affiliate marketing strategies. By providing a closed-loop flywheel of skills, the repository ensures that each step of the marketing funnel is supported by data-driven insights, leading to more effective and efficient campaigns. This tool simplifies the process of researching, creating, and automating affiliate marketing efforts, making it easier for marketers to scale their operations and achieve better results.

### Overview: Memora is a powerful AI agent memory system designed for developers. It provides persistent, collective memory storage, semantic search, and a graph-based interface. Key features include deduplication, supersession lineage, and seamless cross-session context management. ### Key Features: - **Core Storage**: Persistent SQLite storage with optional cloud sync (S3, R2, D1), multi-database routing, hierarchical organization, and export/import capabilities. - **Absorb & Lineage**: Feed facts into the system, update supersession lineage, and retrieve relevant memories using semantic search. - **Search & Intelligence**: Advanced queries, cross-references, LLM deduplication, and memory linking for enhanced search and intelligence. ### Tech Stack: Python ### Why It Matters: Memora is essential for developers looking to build robust AI systems that remember their agents' work, ensuring that knowledge is preserved and easily accessible across multiple sessions and databases.

### Overview: Scene Text Recognition with Permuted Autoregressive Sequence Models (PARSEQ) is a state-of-the-art method for scene text recognition. This project introduces a novel approach that leverages Permuted Autoregressive Sequence Models to achieve high accuracy on various datasets, including COCO Text, IC19 Art, ICDAR 2013, and more. ### Key Features: - **High Accuracy**: PARSEQ achieves state-of-the-art performance on multiple datasets, including COCO Text, IC19 Art, ICDAR 2013, and more. - **Easy-to-Use**: The project includes a Gradio demo for quick and easy model deployment. - **Community Contributions**: PARSEQ has been integrated into popular OCR libraries such as PaddleOCR and docTR, providing robust scene text recognition capabilities. - **Research Contributions**: The project was accepted at ECCV 2022 and has been featured in NVIDIA Developer Blog and other academic publications. ### Tech Stack: - **Python**: The primary programming language used for development. - **PaddleOCR**: A popular multilingual OCR toolkit that has integrated PARSEQ. - **docTR**: A deep learning-based library for OCR that has added PARSEQ. ### Why It Matters: PARSEQ offers a significant improvement in scene text recognition accuracy, making it a valuable tool for developers and researchers working in the field of computer vision and natural language processing. Its integration into popular OCR libraries ensures broad accessibility and utility for various applications.

### Overview: Karukan 是一个支持Linux和macOS的日本语输入方法系统，包含神经韩字韩字转换引擎。它提供实时转换功能，支持多种输入方式，包括IME、CLI和Web界面。 ### Key features: - **神经韩字韩字转换**: 使用GPT-2/Qwen3模型进行推理，提供高度准确的韩字韩字转换。 - **实时转换**: 用户输入时实时显示转换结果，无需按下Space键。 - **考虑上下文**: 考虑到周围文本的日本语转换。 - **学习记忆**: 记忆用户选择的转换结果，优先显示预测转换（前向匹配）。 - **系统辞书**: 从SudachiDict构建系统辞书。 - **生成候选**: 自动生成半角カタカナ、英字、数字等的候选词，附有Mozc由来注释。 - **单字韩字转换**: 读音全部转换为汉字1个字，模型和辞书无法转换的字可以参考后方的候选列表。 - **Emoji输入**: 支持かな读音和Slack风格的`:trigger`查询。 - **转换人物**: 通过config.toml配置的用户文章和相关关键词，可以提高转换的便利性。 ### Tech stack: Rust ### Why it matters: Karukan 是一个功能强大且实用的日本语输入方法系统，适用于需要实时转换和准确韩字韩字转换的用户。它结合了

Overview: TensorFlow Android Demo 旨在为开发者提供在Android设备上运行TensorFlow模型的示例代码，特别是针对车辆检测、车道线检测、人脸检测、动作识别和骨架识别等功能。 Key features: - 车道线检测 - 车辆检测 - 人脸检测 - 动作识别 - 骨架识别 Tech stack: Java Why it matters: 本项目为开发者提供了一个方便的工具，使他们能够将TensorFlow模型集成到自己的Android应用中，无需了解复杂的模型实现细节。通过提供详细的代码示例和注释，本项目帮助用户快速上手，同时鼓励用户探索和改进。 GitHub stars: 746 Repository: yuxitong/TensorFlowAndroidDemo

Overview: Aether is a powerful AI agent designed for Android, iOS, and macOS, built on the Pi framework and featuring a minimalist, extensible, and highly customizable user interface. Key features: - Pi framework integration for seamless tool calling and extension support. - Pi Web Access extension for web-based search and content extraction. - Pi MCP Adapter for model context protocol gateway with lazy loading and interactive authentication. - Pi Subagents for autonomous sub-agents with parallel execution and interactive prompt handling. Tech stack: Kotlin. Why it matters: Aether offers a modern, local AI experience through its extensible architecture and rich extension ecosystem, making it suitable for developers looking to integrate AI capabilities into their applications.

### Overview: SparkVSR is an interactive video super-resolution (VSR) framework that leverages sparse keyframes to enhance video quality. This project aims to provide users with a more controllable and robust video super-resolution experience, ensuring that the restored video maintains high quality even when reference keyframes are absent or imperfect. ### Key Features: 1. **Interactive Keyframe Propagation:** Users can super-resolve or select keyframes using any off-the-shelf image super-resolution (ISR) model, and SparkVSR then propagates these keyframe priors across the entire video sequence. 2. **Flexible Keyframe Selection:** Users can choose keyframes manually, extract I-frames from codecs, or sample randomly, providing flexibility in the restoration process. 3. **Reference-Free Guidance:** SparkVSR's reference-free guidance mechanism ensures robust performance even without reference keyframes, making it suitable for tasks where reference keyframes are not available or are imperfect. ### Tech Stack: The project is implemented using Python, leveraging libraries such as ComfyUI for UI integration and Hugging Face for model training and inference. ### Why It Matters: SparkVSR addresses the limitations of traditional VSR approaches by providing a more controllable and robust video super-resolution experience. This framework enables users to correct unexpected artifacts and maintain high-quality video restoration, making it particularly useful for applications where reference keyframes are not available or are imperfect.