Projects

### Overview: 本项目旨在为无人驾驶车辆领域提供一个全面的资源列表，涵盖人工智能、机器人学、计算机视觉、课程、研究实验室、数据集、期刊、SOTA代码、软件、硬件、小游戏、公司和媒体等内容。资源精选自多个知名机构，包括斯坦福大学、卡内基梅隆大学等，旨在帮助开发者和研究人员在无人驾驶车辆领域快速找到所需信息。 ### Key features: - 人工智能/机器学习框架、库和软件的精选列表 - 深度学习论文阅读路线图 - 深度学习课程 - 计算机视觉资源精选清单 - 机器人技术书籍和课程列表 - 自动驾驶视觉感知研究生课程 - 自动驾驶汽车编程算法工具 - 自动驾驶仪指导和协调概念介绍 - 自动驾驶汽车应用主题介绍 - 自动驾驶车辆和高级自动化课程 - 自动驾驶出租车系统设计项目 - 自动驾驶汽车项目主题介绍 - 自动驾驶车辆伦理和法律问题讨论 - 自动驾驶车辆社会影响研究 - 自动驾驶车辆从政策到伦理到法律的社会影响研究 ### Tech stack: Python, JavaScript, HTML/CSS, Markdown, Git, Docker, TensorFlow, PyTorch, OpenCV, ROS, Unity, Unreal Engine, C++, C#, Arduino, Python, R, MATLAB, SQL, LaTeX, LaTeX, LaTeX, LaTeX, LaTeX, LaTeX, LaTeX, LaTeX, LaTeX, LaTeX, LaTeX, LaTeX, LaTeX, LaTeX,

Overview: 这本《DeepSeek Harness 白皮书》是专为开发者设计的，从零开始介绍如何安装、使用、开发插件以及进行性能调优。它不仅包含中文内容，还提供了英文版本，帮助读者快速上手并深入理解dsh框架。 Key features: - 安装指南 - 实时开发插件 - 性能调优技巧 - 详细实测案例 - 同模型多Agent对比分析 Tech stack: HTML Why it matters: 本书为任何开发者提供了一个从零到一的路径，从入门到实战，再到深度理解dsh框架，帮助读者快速上手并持续学习。 Repository: Electricitysheep/dsh-handbook Primary language: HTML GitHub stars: 904 GitHub description: DeepSeek Harness (dsh) 从 0 到 1 深度手册：安装/插件开发/性能调优/实测案例/同模型多 Agent 实测对比（中文 + 英文 PDF） GitHub description excerpt: "DeepSeek Harness 中文手册 × 生态观察中心"——从 0 到 1 玩转 dsh，跟着 780 帖讨论区看懂生态。包含 15 章手册、195 帖社区响应、280+ Stars。目标是：任何一个开发者，跟着这本书都能从 0 到 1 用起来、写起来。包含从入门到实战

Overview: The Neo Series is a series of native vision-language models designed to unify pixel-word encoding, alignment, and reasoning within an encoder-free dense model architecture. This approach aims to overcome the constraints of modular VLMs and achieve superior efficiency by utilizing under 10% of image-text data. Key features: - Native Vision-Language Models (Neo) - Evaluation Guide (NEO-ov) - Training Guide - 2B and 9B models across different training stages Tech stack: Python Why it matters: The Neo Series represents a significant advancement in the field of vision-language models, offering a more efficient and powerful alternative to modular VLMs, particularly in terms of scalability and performance. Repository: EvolvingLMMs-Lab/NEO Primary language: Python GitHub stars: 907 GitHub description: NEO Series: Native Vision-Language Models from First Principles

### Overview: CodeGraph is a 100% Rust implementation of code graphRAG, offering blazing fast AST+FastML parsing, a surrealDB backend, and advanced agentic code analysis tools through MCP for efficient code agent context management. ### Key features: - **Graph + Embeddings = True Understanding**: CodeGraph builds a real knowledge graph from your codebase, enabling AI agents to reason about relationships and context within your architecture. - **Graph Indexing**: Supports three indexing tiers: fast, balanced, and full, each offering different levels of richness and storage efficiency. - **Agent Memory**: Enables persistent memory for architectural decisions, findings, failed approaches, and reusable procedures, allowing retrieval of knowledge alongside current code evidence across sessions. - **CLI and CLI Hooks**: Offers a CLI for shell commands and project-local Claude Code/Codex hooks, providing agent instructions before indexing. ### Tech stack: Rust ### Why it matters: CodeGraph transforms your entire codebase into a semantically searchable knowledge graph that AI agents can actually *reason* about—not just grep through. ### Repository: Jakedismo/codegraph-rust ### GitHub stars: 906 ### Primary language: Rust ### GitHub description: 100% Rust implementation of code graphRAG with blazing fast AST+FastML parsing, surrealDB backend and advanced agentic code analysis tools through MCP for efficient code agent context management.

Overview: This repository contains skills for coding agents that follow the Synalinks framework, providing a structured way to learn and implement the framework correctly. Key features: - `synalinks` skill: A single skill that covers the entire Synalinks framework, including core concepts, programs, modules, control flow, decision models, agents, knowledge, training, rewards, optimizers, providers, and datasets. Tech stack: Python Why it matters: This skill is crucial for developers working with Synalinks, ensuring that agents use the framework correctly and efficiently, leading to better and more reliable code.

Overview: OpenCode Manager is a self-hosted web application designed for managing OpenCode sessions, Git, terminal access, and schedules on your desktop and mobile devices. Key features: - Sessions management - Git integration - Terminal access - Schedules setup Tech stack: TypeScript Why it matters: OpenCode Manager simplifies the management of OpenCode sessions, Git operations, terminal access, and schedule management, providing a unified interface for users on both desktop and mobile devices. This tool is essential for developers who need to manage their OpenCode sessions efficiently and securely. Repository: chriswritescode-dev/opencode-manager GitHub stars: 908 GitHub description: A self-hosted command center for OpenCode. Sessions, Git, terminal, and schedules in one web app, on your desktop and your phone.

### Overview: Graftcode is a powerful development framework that enables you to call methods across languages and runtimes as if they were local dependencies. It eliminates the need for integration layers, SDKs, and client code, making your codebase leaner and more efficient. ### Key Features: - **Public Methods**: Write public methods in any language of your choice. - **Strong Typing**: Automatically generate strongly-typed clients using package managers. - **Integration-Free**: No need for integration layers, SDKs, or client code. - **Type Safety**: Ensures type safety and strong typing. - **Efficiency**: Reduces your codebase by up to 50%. - **AI-Enabled Development**: Improves AI-assisted development by lowering token usage, higher efficiency, and simpler code reviews. - **Polyglot Systems**: Enables modular monoliths, zero integration overhead, and seamless communication channels. ### Tech Stack: PowerShell ### Why It Matters: - **Elimination of Integration Layers**: Removes the need to write integration code, SDKs, and client layers. - **Codebase Efficiency**: Reduces your codebase by up to 50%, improving AI-assisted development. - **Type Safety**: Ensures type safety and strong typing, making your code more maintainable and efficient. - **AI-Enabled Development**: Improves AI-assisted development by lowering token usage, higher efficiency, and simpler code reviews. - **Polyglot Systems**: Enables modular monoliths, zero integration overhead, and seamless communication channels. ### GitHub Stars: 9

Overview: This project details the Context Encoders, a method for unsupervised feature learning by image inpainting, presented in the CVPR 2016 paper. The code includes quick demos, training/testing scripts for center and random region inpainting, and is based on a modified DCGAN implementation. Key features: - Quick demo for semantic image inpainting - Training scripts for center and random region inpainting - Adapted from Soumith's DCGAN implementation Tech stack: Lua Why it matters: This project provides a practical implementation of Context Encoders, a technique for unsupervised feature learning, and can be used in various applications requiring unsupervised image inpainting.

### Overview: ZhiLight is a highly optimized LLM inference engine developed by Zhihu and ModelBest Inc. It accelerates the inference of models like Llama and its variants, especially on PCIe-based GPUs. Compared to mainstream open-source inference engines, ZhiLight demonstrates significant performance advantages. ### Key Features: - Asynchronous OpenAI compatible interface adapted from vllm - Custom defined tensor and unified global memory management - Encode and all-reduce overlap, named "dual streams" * Support Int8-quantized all-reduce to further reduce all-reduce cost * Host all-reduce based on SIMD instructions * Optimized fused kernels, qkv, residual & layernorm etc. * Fused batch attention for decoding based on tensor core instructions * Support TP and PP on one node, TP is recommended * Support dynamic batch * Support flash attention prefill * Support chunked prefill * Support prefix cache * Support Native INT8/SmoothQuant/FP8/AWQ/GPTQ quantization * Support Marlin kernel for GPTQ * Support MoE, DeepseekV2 MoE and DeepseekV2 MLA * Support Llama/Llama2, Mixtral, Qwen2 series and similar models ### Tech Stack: C++ ### Why it Matters: ZhiLight significantly accelerates the inference of models like Llama and its variants, especially on PCIe-based GPUs. It offers performance advantages over mainstream open-source inference engines

Overview: The DSH Vision Toolkit is a powerful tool designed to enhance the capabilities of text-only DeepSeek Harness agents. It allows agents to understand images and complete visual tasks directly by pasting an image, without the need for manual intervention. Key features: - Direct image recognition: Automatically switches to the `(Vision Toolkit)` variant when pasting an image. - One command installation: Configures the vision provider in the settings and starts using the tools. - Context-specific content: The model extracts evidence relevant to the current task, such as "Where is the error?" or "Where is the button?" - Native integration: Built with reference to `agent-vision-toolkit`, it seamlessly integrates into Web and Headless Profiles. Tech stack: TypeScript Why it matters: This project provides a comprehensive solution for visual tasks within the DeepSeek Harness ecosystem, enhancing the capabilities of text-only agents and making image-based tasks more intuitive and efficient. Repository: Anionex/dsh-vision-toolkit GitHub stars: 903 GitHub description: A more powerful vision toolkit—give text-only models in DeepSeek Harness eyes: image Q&A, long-screenshot OCR, UI restoration, and GUI visual tasks in one toolkit and Skill.

### Overview: The dataset viewer is a backend that powers the dataset viewer on Hugging Face's dataset pages through a public API. This repository provides pre-computed data for all datasets on the Hub, ensuring a smooth user experience without the need for frequent API calls. ### Key features: - Table displaying dataset contents in pages of 100 rows. - Navigation between pages using buttons. - Filtering and searching capabilities. - Basic statistics view. - Public API for backend data. ### Tech stack: Python ### Why it matters: This backend simplifies the dataset viewer functionality, reducing the load on the Hugging Face servers and improving the performance of the viewer component.

### Overview: Claude Code is a developer-focused plugin that allows multiple AI coding agents to answer a question simultaneously. It helps developers make informed decisions by comparing answers from different models and providing a synthesized recommendation. ### Key Features: - **Ask Providers Together**: Combine answers from multiple AI agents for a comprehensive perspective. - **Choose Providers**: Select and pin providers for a specific task. - **Specialize Roles**: Assign roles to different models for specialized tasks. - **Challenging Each Other**: Allow models to challenge each other's answers. - **Agent Mode**: Enhance analysis by using Claude agents. - **Local Council**: Use Claude subagents without external providers. - **Auto-Context Injection**: Include relevant files or choose your own context. - **Image Input**: Attach a screenshot to your question. - **Ask About Conversation**: Get a note from Claude while it's working. - **Background Jobs**: Run in the background and fetch results. - **Export Answers**: Save answers to a file. - **Quiet Mode**: Show only the synthesized answer. - **Watch Answers**: Monitor answers as Claude works. - **Stop-Gate Review**: Have a second model review the diff before Claude stops. - **Direct Script Usage**: Call scripts without Claude Code. ### Tech Stack: The plugin uses Shell commands for interaction with Claude Code. ### Why It Matters: Claude Code provides a reliable and informed approach to decision-making by leveraging multiple AI models, ensuring developers get the best possible advice.

### Overview: Similarities: A Toolkit for Similarity Calculation and Semantic Search Similarities is a toolkit designed for calculating similarities and performing semantic search on text and images. It supports text and image data, enabling efficient similarity calculations and semantic search across billions of documents. The toolkit is built with Python 3 and is available for use with the command line, making it easy to integrate into projects. ### Key Features: - **Text Similarity and Search**: Utilizes the CoSENT model for text similarity calculation and semantic search. Supports various SentenceBERT models and multiple similarity calculation methods including Cos Similarity, Dot Product, Hamming Distance, and Euclidean Distance. The toolkit also includes various text search algorithms like Faiss, Annoy, Hnsw, and Faiss-based services for efficient text search. - **Image Similarity and Search**: Implements the CLIP model for image similarity and search. Supports various CLIP models including OpenAI's CLIP-ViT and Chinese-CLIP models. The toolkit provides FastAPI backend services and Gradio frontend for image search, enabling efficient image similarity and search across billions of images. It also supports image embedding extraction and image feature extraction using various algorithms like pHash, dHash, wHash, aHash, and SIFT. ### Tech Stack: Python 3 The toolkit is built with Python 3, ensuring compatibility and ease of integration with existing projects. ### Why It Matters: Similarities is a powerful tool for text and image similarity calculations and semantic search. It supports efficient similarity and search across billions of documents and images, making

### Overview: This repository contains all the slides, accompanying code, and exercises from DataCamp's Python Data Science track. It is a comprehensive collection of courses that cover various aspects of data science using Python. ### Key Features: - **Courses:** - Introduction to Python for Data Science - Intermediate Python for Data Science - Python Data Science Toolbox (Part 1) - Python Data Science Toolbox (Part 2) - Importing Data in Python (Part 1) - Importing Data in Python (Part 2) - Cleaning Data in Python - Pandas Foundations - Manipulating DataFrames with pandas - Merging DataFrames with pandas - Introduction to Databases in Python - Introduction to Data Visualization with Python - Interactive Data Visualization with Bokeh - Statistical Thinking in Python (Part 1) - Statistical Thinking in Python (Part 2) - Supervised Learning with scikit-learn - Machine Learning with the Experts: School Budgets - Unsupervised Learning in Python - Deep Learning in Python - Network Analysis in Python (Part 1) - Natural Language Processing Fundamentals in Python - Correlation and Regression ### Tech Stack: Python ### Why it Matters: This repository serves as a valuable resource for anyone looking to enhance their skills in Python-based data science. It provides a structured learning path through DataCamp's courses, ensuring comprehensive understanding and practical application of Python in data science.

### Overview: TextGAN is a PyTorch framework designed for Generative Adversarial Networks (GANs) focused on text generation. It serves as a benchmark platform for research on GAN-based text generation models. ### Key features: 1. **General Text Generation** - **SeqGAN** - Sequence Generative Adversarial Nets with Policy Gradient - **LeakGAN** - Long Text Generation via Adversarial Training with Leaked Information - **MaliGAN** - Maximum-Likelihood Augmented Discrete Generative Adversarial Networks - **JSDGAN** - Adversarial Discrete Sequence Generation without Explicit Neural Networks as Discriminators - **RelGAN** - Relational Generative Adversarial Networks for Text Generation - **DPGAN** - Diversity-Promoting Generative Adversarial Network for Generating Informative and Diversified Text - **DGSAN** - Discrete Generative Self-Adversarial Network - **CoT** - Cooperative Training for Generative Modeling of Discrete Data 2. **Category Text Generation** - **SentiGAN** - Generating Sentimental Texts via Mixture Adversarial Networks - **CatGAN** - Category-aware Generative Adversarial Networks with Hierarchical Evolutionary Learning for Category Text Generation ### Tech stack: PyTorch ### Why it matters: TextGAN provides a comprehensive framework for researchers and developers to experiment with GAN-based text generation models, making it easier to transition from TensorFlow to

Overview: This production-ready template is designed to streamline the development of Generative AI projects, providing a structured and scalable foundation for long-term maintainability. Key features: - Modular code structure - Clear module boundaries - YAML configuration for models and prompts - Separate configs using YAML files - Cache responses to reduce latency and cost - Handle errors with custom exceptions - Use notebooks for rapid testing and iteration - Monitor API usage and set rate limits - Keep code and docs in sync Tech stack: Python Why it matters: This template ensures that Generative AI projects are built with proven structure and practices, making them more maintainable and scalable over time. Repository: HeyNina101/generative_ai_project GitHub stars: 903

Overview: The repository, TheWhisper, aims to provide optimized speech-to-text and text-to-speech inference solutions, focusing on self-hosting, cloud hosting, and on-device inference across multiple devices. It includes open-source transcription models with streaming inference support, high-performance TheStage AI inference engines, CoreML engines for macOS and Apple Silicon, and a local REST API with frontend examples. Key features: - Open weights fine-tuned versions of Whisper models - Fine-tuned models support inference with 10s, 15s, 20s, and 30s - CoreML engines for macOS and Apple Silicon, with low power consumption and low RAM usage - Optimized engines for NVIDIA GPUs through TheStage AI ElasticModels - Streaming implementation - Benchmarks for latency, memory, power, and ASR accuracy - Simple Python API, examples, and a tutorial for deployment on macOS with Electron and ReactJS Tech stack: Python, Hugging Face, NVIDIA GPUs, TheStage AI ElasticModels Why it matters: The repository is optimized for low latency, low power usage, and scalability, making it ideal for real-time captioning, live meetings, voice interfaces, and edge deployments. GitHub stars: 903 GitHub description: Optimized Whisper models for streaming and on-device use GitHub repository: TheStageAI/TheWhisper

### Overview: IRIS is a groundbreaking data-efficient agent that leverages the power of Transformers to learn from millions of imagined trajectories in a world model. This approach enables the agent to perform well in complex environments with minimal real-world experience, making it highly valuable for applications requiring efficient learning and adaptation. ### Key Features: - **Data-Efficient Learning**: IRIS is trained using millions of imagined trajectories, significantly reducing the need for extensive real-world experience. - **Discrete Autoencoder**: The world model is composed of a discrete autoencoder, allowing for efficient encoding and decoding of images. - **Autoregressive Transformer**: The Transformer architecture composes the language of image tokens over time, enabling dynamic learning and efficient inference. ### Tech Stack: - **Python**: The primary programming language used for developing the IRIS agent. - **PyTorch**: The deep learning library powering the implementation of the Transformer and autoencoder components. ### Why It Matters: IRIS represents a significant advancement in the field of reinforcement learning, particularly in the realm of sample-efficient learning. By enabling agents to perform well with minimal real-world experience, IRIS opens up new possibilities for applications in robotics, autonomous systems, and more, where efficient learning is crucial. ### Repository: - **Repository**: eloialonso/iris - **GitHub Stars**: 903 - **GitHub Description**: Transformers are Sample-Efficient World Models. ICLR 2023, notable top 5%. ### Setup: - **Dependencies**: Install PyTorch and other required dependencies via `pip install -r requirements

### Overview: YOLOS is a project that studies the transferability of the vanilla ViT pre-trained on mid-sized ImageNet-1k to the challenging COCO object detection benchmark. It demonstrates that a vanilla ViT can be successfully transferred to perform object detection with minimal modifications, showcasing the versatility and transferability of Transformers. ### Key Features: - Directly inherited from ViT and DeiT - Uses mid-sized ImageNet-1k as pre-training dataset - Only looks at one sequence for transferable object detection - Demonstrates 2D object detection in a pure sequence-to-sequence manner - Utilizes minimal 2D inductive biases - Feasible for object detection in any dimensional space unaware of the input size ### Tech Stack: Jupyter Notebook ### Why it Matters: This project highlights the versatility and transferability of Transformers in vision tasks, particularly in the realm of object detection. It shows that a vanilla ViT pre-trained on ImageNet-1k can be successfully transferred to perform challenging object detection tasks with minimal modifications, making it a valuable contribution to the field of computer vision.

Overview: fauxnix is a powerful tool that allows you to run Linux-style commands on Windows, translating them into PowerShell for deterministic execution without the need for a VM or WSL. Key features: - Run Linux commands on Windows - No VM or WSL required - Deterministic translation of commands - Supports AI agents like Claude Code, Codex, OpenCode, and Qwen Code - GNU-style output, bash-style errors, and UTF-8/GBK handling Tech stack: TypeScript Why it matters: fauxnix simplifies the process of running Linux commands on Windows, making it easier for developers to use Linux-like tools without the complexity of virtual machines or WSL. This is particularly useful for AI agents that need to run scripts in a Linux environment. Repository: 20000419/fauxnix Primary language: TypeScript GitHub stars: 903 GitHub description: Run Linux-style commands on Windows via deterministic bash→PowerShell translation. No VM, no WSL. MCP server + CLI built for AI agents (Claude Code, Codex, OpenCode...). GNU-style output, bash-style errors, UTF-8/GBK handled.