Projects

### Overview: This project introduces a Chinese Named Entity Recognition (NER) model using BERT, including models like BERT+Softmax, BERT+CRF, and BERT+Span, and their performance on the CLUENER dataset. ### Key Features: - **BERT+Softmax**: Uses a softmax layer for classification. - **BERT+CRF**: Applies Conditional Random Fields for improved accuracy. - **BERT+Span**: Utilizes span-based models for more precise entity tagging. ### Tech Stack: PyTorch ### Why it Matters: This project provides a comprehensive solution for Chinese NER using BERT, demonstrating the effectiveness of different BERT-based models on the CLUENER dataset.

### Overview: Kernel Memory is a research project aimed at providing best practices and a reference implementation for memory management in specific AI and LLMs application scenarios. The code serves as a demonstration and is not an officially supported Microsoft offering. It includes features such as efficient indexing of datasets, support for Retrieval Augmented Generation (RAG), synthetic memory, prompt engineering, and custom semantic memory processing. ### Key features: - Efficient indexing of datasets through custom continuous data hybrid pipelines. - Support for Retrieval Augmented Generation (RAG). - Synthetic memory. - Prompt engineering. - Custom semantic memory processing. ### Tech stack: C# ### Why it matters: Kernel Memory is designed to enable seamless integration with Microsoft's Semantic Kernel, Microsoft Copilot, and ChatGPT, making it a valuable tool for developers looking to enhance their AI applications. The system is optimized for natural language querying and supports citations and links to original sources, making it a powerful tool for researchers and developers alike.

### Overview: This MATLAB and Simulink Challenge Project Hub provides a collection of research and design project ideas that aim to foster practical experience and insight into technology trends and industry directions. ### Key features: - Explore a diverse range of projects across various technologies and applications. - Gain access to MATLAB and Simulink resources for students. - Participate in regular contests hosted by the MATLAB Central community to test skills and win prizes. - Receive official recognition for problem-solving skills from technology leaders at MathWorks. - Learn about technology trends and industry directions through these projects. ### Tech stack: MATLAB and Simulink. ### Why it matters: These projects offer a unique opportunity to apply theoretical knowledge to real-world challenges, enhancing both technical skills and problem-solving abilities. By participating, you can contribute to science and engineering advancements and gain valuable recognition from industry leaders.

Overview: PyGAD is an open-source Python library for building genetic algorithms and training machine learning models, supporting Keras and PyTorch. Key features: - Supports single-objective and multi-objective optimization - Integrates with Keras and PyTorch - Provides easy-to-use functions for genetic algorithm implementation - Includes various types of crossover, mutation, and parent selection methods Tech stack: Python Why it matters: PyGAD simplifies the process of building and optimizing genetic algorithms, making it accessible for developers and researchers alike. Repository: ahmedfgad/GeneticAlgorithmPython Primary language: Python GitHub stars: 2232 GitHub description: Source code of PyGAD, a Python 3 library for building the genetic algorithm and training machine learning algorithms (Keras & PyTorch).

Overview: Tracking Any Point (TAP) is a benchmark and dataset designed for evaluating point tracking algorithms, including TAPIR, BootsTAPIR, RoboTAP, and TAPNext. The repository contains various demos, benchmarks, and metrics to help developers improve their tracking models. Key features: - TAPIR: A two-stage algorithm that locates and refines point matches. - BootsTAPIR: Uses large unlabeled video data to improve tracking accuracy. - TAPNext: A simple yet effective tracker that tracks points by propagating information through a network. - Trajan: A trajectory autoencoder that reconstructs trajectories from support points. - TAPVid Benchmark: A benchmark for point tracking algorithms. - RoboTAP Benchmark: A dataset for robotics manipulation tasks using TAPIR point tracks. - TAPVid-3D Benchmark: Metrics and evaluation code for 3D point tracking. Tech stack: Jupyter Notebook Why it matters: TAP is crucial for developers aiming to improve their point tracking algorithms, providing benchmarks, datasets, and metrics to evaluate and enhance their models' performance.

Overview: This curated list provides resources for machine learning in Ruby, including tutorials, libraries, applications, and more. Key features: - Tutorials on various machine learning topics using Ruby - Machine learning libraries and frameworks - Applications of machine learning in Ruby - Data structures and visualization techniques Tech stack: Ruby programming language Why it matters: This list is useful for developers looking to integrate machine learning into their Ruby projects. Repository: arbox/machine-learning-with-ruby GitHub stars: 2230 GitHub description: Curated list: Resources for machine learning in Ruby README excerpt: <img title="Awesome Machine Learning with Ruby" alt="Awesome Machine Learning with Ruby" src="header.png" align="center">

Overview: Kalosm is an ecosystem of crates designed for developers to easily integrate local or remote AI models into their Rust applications. Key features: - Kalosm: A simple interface for pre-trained models in Rust. - Fusor: A compiler-backed runtime for quantized ML inference on CPUs and WebGPU accelerators. Tech stack: Rust Why it matters: Kalosm simplifies the integration of AI models into Rust applications, making it easier for developers to leverage pre-trained models for various tasks such as text, audio, and image processing. The ecosystem includes utilities for extracting context, transcribing audio, and structured generation, making it a comprehensive solution for AI model usage in Rust projects.

**Overview:** any-llm provides a unified interface for communicating with various LLM providers, making it easy to switch between different providers without changing the code. **Key features:** - Supports multiple LLM providers including OpenAI, Anthropic, Azure / Microsoft Foundry, Mistral, and more. - Allows seamless switching between providers without modifying the code. - Simplifies the integration process by handling API keys and environment variables automatically. **Tech stack:** Python **Why it matters:** any-llm simplifies the process of integrating with different LLM providers by providing a single, unified interface, making it easier to switch between providers without changing the code. This makes it particularly useful for developers working on projects that require interaction with multiple LLM providers.

Overview: Proma is a comprehensive Agent experience designed for professionals, offering a seamless workflow integration and proactive capabilities. Key features: - Chat mode - Agent mode - Task management - Project partitioning - Memory management - Embedded browser - Embedded terminal - Timed tasks - Agent parallel exploration - Sub- and main sessions - Skills / MCP / CLI - Remote link support: WeChat,飞书,钉钉, Slack - File preview and editing capabilities Tech stack: TypeScript Why it matters: Proma provides a professional-level Agent experience, making it easier for developers to integrate and utilize AI tools seamlessly. Repository: proma-ai/Proma Primary language: TypeScript GitHub stars: 2225 GitHub description: Proma brings a seamless general-purpose Agent experience to your workflow. Built for 100× professionals and the proactive Agent era.

### Overview: UnrealCV is a project that enables computer vision researchers to create virtual worlds using Unreal Engine, extending the engine with a plugin to interact with the virtual world and communicate with external programs. ### Key Features: - **UnrealCV Commands**: Provides a set of commands to interact with the virtual world. - **Communication with External Programs**: Facilitates communication between Unreal Engine and external programs such as Pytorch/Tensorflow. - **Two Usage Modes**: - Embedded Game Binary: Simple to use, no knowledge of Unreal Engine required. - Unreal Engine Plugin: Allows building a new virtual world using the Unreal Engine editor. - **Support for Unreal Engine 5.2 and Later**: Ensures compatibility with the latest version of Unreal Engine. - **Optical Flow Image Capture**: `vget /camera/[id]/optical_flow [format]`. - **Python Commands**: `vbp [obj_name] [func_name] [arg1] [arg2] ...` to call Blueprint functions. - **Improved RPC Communication**: Higher FPS and more reliable communication between Server and Client. - **New Camera and Object Commands**: Extensive set of commands for camera control and object manipulation. - **UnrealZoo Features**: Continuously developed features tested in UnrealZoo environments, available in supported UnrealZoo environments. - **UnrealCV Runtime MCP**: Public clients, examples, and agent skills available at unrealcv-runtime-mcp. - **UnrealCV Dev For UnrealZoo**: Features available in UnrealZoo environments, not part of the open-source

Overview: Deep Research Web UI is an AI-driven tool that converts research questions into cited reports, providing real-time feedback, visualization, and evidence verification. It supports multiple providers and is self-hosted with Docker. Key features: - Safe & Secure: Config and API requests stay in the browser locally. - Realtime Feedback: AI responses and UI updates are reflected in real-time. - Search Visualization: Shows the research process in a tree structure. - Export as PDF: Final research report can be exported as Markdown or PDF. - Supports more models: Uses plain prompts instead of newer, less widely supported features. - Docker support: Deploy in your environment with a single command. - Server Mode: Deploy with environment variables, no need for API key configuration. - Evidence and follow-ups: Inspect source excerpts and research further, updating related passages. - Focused search: Separate queries and filters, check relevance, and retry with revised queries. - On-demand full-page reading: Fetches full source pages for insufficient search snippets. Why it matters: This tool simplifies the research process, ensuring accurate and up-to-date information, and supports a wide range of providers, making it versatile and reliable. Tech stack: TypeScript GitHub stars: 2223 Repository: AnotiaWang/deep-research-web-ui

### Overview: This repository is a collection of diffusion model papers categorized by their subareas, focusing on visual illusions, color generation, and various applications in image restoration, storytelling, and more. ### Key Features: - **Visual Illusion**: Papers exploring color visual illusions and their generation using diffusion models. - **Color**: Techniques for generating color images using diffusion models. - **Count**: Guidance for count-based diffusion models. - **Poster**: Categorized papers related to diffusion models. - **Accelerate**: Papers on train-free diffusion models and acceleration techniques. - **Image Restoration**: Methods for image restoration, including colorization, face restoration, image compression, and super-resolution. - **Storytelling**: Papers on diffusion models for creating visual storytelling. - **Virtual Try-On**: Techniques for virtual try-on using diffusion models. - **Drag Edit**: Tools for drag-and-editing using diffusion models. - **Text-Guided Editing**: Papers on text-guided image editing, including diffusion inversion and multiple concept learning. - **Continual Learning**: Papers on continual learning with multi-concept and decomposition-based approaches. - **Remove Concept**: Techniques for removing concepts using diffusion models. - **In Context Learning**: Papers on in-context learning with multi-concept, decomposition, and general personalization. - **Image Translation**: Methods for image translation using diffusion models. - **Seg & Detect & Track**: Papers on segmentation, detection, and tracking using diffusion models. - **Adding Conditions**: Techniques for adding conditions to diffusion models. - **Few-Shot**: Papers on few-shot learning using diffusion models

### Overview: APIAuto 是一款敏捷开发中不可或缺的HTTP接口工具，集成了机器学习零代码测试、生成代码与静态检查、生成文档与光标悬浮注释等功能。它广泛应用于腾讯、华为、SHEIN、传音、工行等知名企业，助力开发者高效开发和测试API接口。 ### Key Features: - **机器学习零代码测试**：自动识别并修复常见错误，提高测试效率。 - **生成代码与静态检查**：自动生成代码并进行静态分析，确保代码质量。 - **生成文档与光标悬浮注释**：实时生成文档并提供光标悬浮注释，提高开发效率。 - **一站式智能开发管理**：涵盖文档、测试、Mock、调试、管理等全面功能。 ### Tech Stack: JavaScript ### Why it Matters: APIAuto 是敏捷开发中不可或缺的工具，能够显著提高开发效率和代码质量。它通过机器学习和零代码测试，帮助开发者快速构建和测试API接口，同时提供全面的文档和注释功能，确保开发过程的顺利进行。在众多企业中广泛应用，包括腾讯、华为、SHEIN、传音等，证明了其在实际项目中的价值。 ### Repository: - **Repository**: TommyLemon/APIAuto - **Stars**: 2222 - **Description**: ☔ 敏捷开发最强大易用的接口工具，机器学习零代码测试与 AI 问答、生成代码与静态检查、生成文档与光标悬浮注释

### Overview: Helios is a real-time long video generation model that achieves minute-scale, high-quality video synthesis with a 19.5 FPS on a single H100 GPU, making it 10 times faster than 1.3B models without relying on conventional strategies or techniques. ### Key Features: - **Real-Time Video Generation**: Helios can generate videos in real-time, enabling immediate feedback and interactive experiences. - **High-Quality Synthesis**: The model maintains strong coherence and quality, ensuring that the generated videos are visually appealing and consistent. - **Efficient Inference**: Helios achieves 19.5 FPS on a single H100 GPU, significantly reducing inference time compared to conventional methods. ### Tech Stack: - Python ### Why It Matters: Helios represents a significant advancement in video generation, offering faster and higher-quality video synthesis without compromising on the quality of the output. This model is particularly useful in applications requiring real-time video generation, such as virtual reality, augmented reality, and interactive storytelling. ### Repository: The official implementation of Helios is available on GitHub at [PKU-YuanGroup/Helios](https://github.com/PKU-YuanGroup/Helios). This repository includes the model, scripts, and documentation necessary for both training and inference. ### GitHub Stars: The repository has been starred by 22,150 users, indicating high interest and engagement in the project. ### Video Demos: For a visual demonstration of Helios, please refer to the following links: - [Demo Video of

Overview: This project is an unofficial PyTorch implementation of VALL-E, a zero-shot text-to-speech synthesizer model. VALL-E is designed to synthesize speech that maintains speaker identity, potentially carrying risks of misuse. Key features: - Zero-shot text-to-speech synthesis - Reproduces a demo from the original paper - Supports training on one GPU - Includes a reproducible demo Tech stack: PyTorch Why it matters: VALL-E's ability to generate speech that mimics a specific speaker's voice could be used for malicious purposes, such as impersonating a known speaker or spoofing voice identification. Therefore, the project's primary focus is to ensure that the trained model and its services are well-trained and not provided to avoid such misuse. Repository: lifeiteng/vall-e Primary language: Python GitHub stars: 2221 GitHub description: PyTorch implementation of VALL-E(Zero-Shot Text-To-Speech), Reproduced Demo https://lifeiteng.github.io/valle/index.html

Overview: PyTextRank is a Python implementation of TextRank algorithms, specifically designed for phrase extraction and related natural language processing tasks. It serves as a library for graph-based natural language work, including TextRank, PositionRank, Biased TextRank, and TopicRank. Key features: - Phrase extraction - Low-cost extractive summarization - Concept inference from unstructured text Tech stack: Python Why it matters: PyTextRank provides a powerful toolset for natural language processing tasks, making it invaluable for researchers and developers looking to enhance their text analysis capabilities. Repository: DerwenAI/pytextrank Primary language: Python GitHub stars: 2221 GitHub description: Python implementation of TextRank algorithms ("textgraphs") for phrase extraction README excerpt: PyTextRank is a Python implementation of *TextRank* as a spaCy pipeline extension, for graph-based natural language work. It includes TextRank, PositionRank, Biased TextRank, and TopicRank. Popular use cases include phrase extraction, extractive summarization, and concept inference. The library is used in various applications, including solving systems of linear constraints and finding minimal generating sets of solutions. It can be installed via pip or conda, and examples are provided in the tutorial notebooks.

Overview: Agent Mail is an asynchronous coordination layer for AI coding agents, providing identities, inboxes, searchable threads, and advisory file leases over FastMCP, Git, and SQLite. Key features: - Asynchronous email-like coordination for coding agents - Memorable identities for agents - Searchable message history - Voluntary file reservation leases to avoid stepping on each other Tech stack: Python Why it matters: Agent Mail simplifies asynchronous communication and collaboration among multiple coding agents, reducing conflicts and improving productivity. Repository: Dicklesworthstone/mcp_agent_mail Primary language: Python GitHub stars: 2217

Overview: PyTorch-NLP is a basic utilities library for PyTorch Natural Language Processing (NLP) that extends PyTorch with essential text processing functions. Key features: - Load datasets - Tokenize and encode text - Batch data for training Tech stack: Python Why it matters: This library provides a simple yet powerful set of tools for NLP tasks, making it easier to work with text data in PyTorch projects.

### Overview: Ainovel-cli is a Go-based project that automates the creation of long-form AI novels through a multi-agent approach. It ensures that the engine runs the entire book deterministically, with models being used precisely at each decision point. The system includes features like semantic arbitration, step-level checkpoint recovery, and intelligent chapter recommendations, making it a powerful tool for generating detailed and coherent AI-generated novels. ### Key features: - **Deterministic Engine + Multi-Agent Collaboration**: The Engine routes decisions based on a fact-driven route table, while the Arbiter handles semantic arbitration through single calls. This ensures zero LLM cost and allows for exhaustive behavior testing. - **Semantics Can Be Audited**: The Arbiter makes decisions through structured calls, with each verdict traceable and replayable. It simplifies complex arrangements by avoiding unnecessary complexity. - **Step-Level Breakpoint Recovery**: Each tool's execution is checkpointed, allowing for precise recovery from any point in the process. - **Rolling Arcs with Layered Planning**: Long chapters are not planned in one go; initial planning focuses on the first two arcs, with subsequent arcs and arcs being expanded based on the writing progress. - **Chapter-Specific Recommendations**: The system suggests related historical chapters based on the chapter's context, ensuring continuity in 500+ chapter novels. - **Self-Adaptive Context Strategies**: The system automatically switches between full, window, and layered summaries based on the total number of chapters, supporting 500+ chapters. - **Seven-Dimensional Quality Review**: The Editor evaluates the novel

Overview: ZhuSuan is a Python library designed for Bayesian deep learning, focusing on probabilistic programming. It combines the strengths of Bayesian methods and deep learning, providing tools for building and applying probabilistic models. Key features: - Variational Inference (VI) - Importance Sampling (IS) - Hamiltonian Monte Carlo (HMC) - Stochastic Gradient Markov Chain Monte Carlo (SGMCMC) Tech stack: Python Why it matters: ZhuSuan offers a unique approach to Bayesian deep learning by providing deep learning-style primitives and algorithms for building probabilistic models, making it a valuable tool for researchers and practitioners in the field of Bayesian deep learning and generative models. Repository: thu-ml/zhusuan Primary language: Python GitHub stars: 2218 GitHub description: A probabilistic programming library for Bayesian deep learning, based on TensorFlow. Unlike existing deep learning libraries, it focuses on probabilistic models and Bayesian inference. Installation: Users can install ZhuSuan by running `pip install .` in the main directory. It requires TensorFlow 1.13.0 or later. Documentation: ZhuSuan provides tutorials and API documentation, as well as a white paper available on the project's documentation site. Examples: The library includes examples for traditional hierarchical Bayesian models, deep generative models, and advanced techniques like Variational Autoencoders, Sigmoid Belief Networks, and more. Citing ZhuSuan: If you find ZhuSuan useful, please cite it in your publications using the BibTeX entry provided in the README.