Projects

### Overview: Libra is an ergonomic machine learning library designed for non-technical users, providing a simple and efficient way to perform machine learning tasks without requiring advanced technical knowledge. ### Key features: - **Easy Data Loading**: Supports various data formats and can be easily integrated with existing datasets. - **Model Building**: Facilitates the creation of machine learning models using intuitive query syntax. - **Model Evaluation**: Provides detailed metrics and plots to evaluate model performance. - **Interpretability**: Offers a user-friendly interface for model interpretation, making complex models accessible. ### Tech stack: Python ### Why it matters: Libra simplifies the process of machine learning, making it accessible to a wide range of users, from beginners to advanced users, without the need for extensive technical expertise.

Overview: This tutorial guides you through the process of building your own AI agent, starting from a simple chat loop and gradually transitioning to a more advanced version of OpenClaw. Key features: - A step-by-step guide - Key components and design decisions explained in READMEs - A runnable codebase Tech stack: Python Why it matters: This tutorial provides a comprehensive approach to building an AI agent, from a basic chat loop to a more advanced version of OpenClaw, making it a valuable resource for developers looking to create their own AI agents.

### Overview: DemoGPT is a comprehensive platform designed to streamline the creation of LLM (Large Language Model) Agents. It provides a suite of tools, prompts, frameworks, and models all in one place, making it easier to generate automatic LangChain pipelines. ### Key Features: - 🚀 Access to a wide range of tools, prompts, frameworks, and models. - 📝 Comprehensive documentation and tutorials. - 🚀 Streamlined agent development process. - 📊 Knowledge hub for LLM models. ### Tech Stack: Python ### Why It Matters: DemoGPT simplifies the process of creating LLM Agents by providing a one-stop solution for all necessary components. This makes it easier for developers to create, test, and deploy their agents quickly and efficiently.

### Overview: ocrs is a Rust library and CLI tool designed for extracting text from images, also known as OCR (Optical Character Recognition). The goal is to create a modern OCR engine that works well on a wide variety of images with zero or minimal preprocessing effort compared to earlier engines like Tesseract. The library uses machine learning extensively and is easy to compile and run across a variety of platforms, including WebAssembly. It is trained on open and liberally licensed datasets and has a codebase that is easy to understand and modify. ### Key features: - OCR (Optical Character Recognition) functionality - Modern, efficient OCR engine - Works well on a wide variety of images - Zero preprocessing effort compared to Tesseract - Easy to compile and run on various platforms - Trained on open and liberally licensed datasets - Codebase easy to understand and modify ### Tech stack: - Rust - PyTorch - ONNX - RTen ### Why it matters: ocrs is a modern, efficient OCR engine that works well on a wide variety of images with minimal preprocessing effort compared to earlier engines like Tesseract. It is easy to compile and run on various platforms, including WebAssembly, and is trained on open and liberally licensed datasets. The codebase is easy to understand and modify, making it a valuable tool for developers looking to extract text from images.

### Overview: BruteForceAI is an advanced penetration testing tool that leverages Large Language Models (LLM) for intelligent form analysis to automate and enhance brute-force attacks. This tool intelligently identifies login form selectors using AI, executes sophisticated multi-threaded attacks with human-like behavior patterns, and provides advanced evasion techniques for better stealth. ### Key Features: - **LLM-Powered Form Analysis**: Identifies login form elements and selectors using AI, reducing the need for manual intervention. - **Multi-threaded Execution**: Executes attacks with synchronized delays and randomization for better performance. - **Bruteforce & Password Spray**: Executes brute-force and password spray attacks with intelligent timing patterns. - **User-Agent Rotation**: Rotates User-Agent for better evasion. - **Webhook Notifications**: Sends notifications on success to Discord, Slack, Teams, and Telegram. - **Comprehensive Logging**: Logs attacks with SQLite database and verbose output. - **Automatic Update Checking**: Checks for updates automatically from mordavid.com. ### Tech Stack: - Python 3.8 or higher - Playwright for browser automation - Requests for HTTP requests - PyYAML for YAML parsing ### Why It Matters: BruteForceAI combines AI intelligence with automated login attacks to revolutionize traditional brute-force methods. By using LLMs, it reduces the need for manual form analysis, making the process more efficient and effective. The advanced features, including multi-threaded execution, human-like timing, and evasion techniques, ensure that the tool remains stealthy and effective in real-world penetration testing scenarios. ### Repository

Overview: This repository contains the code developed by TensorFlow for the paper "3D Convolutional Neural Networks for Cross Audio-Visual Matching Recognition," authored by Amirsina Torfi, Seyed Mehdi Iranmanesh, Nasser Nasrabadi, and Jeremy Dawson. The code aims to provide an implementation for Coupled 3D Convolutional Neural Networks for audio-visual matching, with a specific focus on lip-reading applications. Key features: - Coupled 3D Convolutional Neural Networks for Cross Audio-Visual Matching Recognition - Lip-reading implementation Tech stack: Python Why it matters: This project is significant for researchers and developers interested in audio-visual matching and speech recognition, as it provides a practical solution for improving the accuracy of speech recognition in noisy audio environments and verifying speaker identities in multi-speaker scenarios.

### Overview: This repository is a curated list of resources related to Knowledge Graphs, including databases, tools, and learning materials. It aims to provide a comprehensive overview of the latest advancements and best practices in the field of knowledge graphs. ### Key features: - Graph Databases: A variety of graph databases, such as AgensGraph, ArangoDB, ArcadeDB, and Memgraph. - Triple Stores: Triple stores for storing and querying knowledge graphs. - Graph Computing Frameworks: Tools like JanusGraph and Kuzu for graph computing. - Graph Visualization: Tools and libraries for visualizing graph data. - Graph Construction: Resources for creating and managing graph databases. - Languages: Support for various programming languages, including SQL and Cypher. - Managed Hosting Services: Services like Blazegraph and Sparksee for hosting and managing graph databases. - Knowledge Engineering: Resources for knowledge fusion and knowledge engineering. - Knowledge Graph Dataset: Datasets for semantic networks, academic knowledge graphs, and general knowledge graphs. ### Tech stack: The repository uses Rust for the Atomic-Server, Go for Cayley, and PostgreSQL for AgensGraph. It also includes various databases like ArangoDB, Dgraph, and Memgraph, as well as tools like WeaveHQ and Kuzu. ### Why it matters: Knowledge graphs are crucial for understanding complex relationships and knowledge across different domains. This repository provides a valuable resource for developers and researchers looking to enhance their knowledge graph capabilities and applications.

Overview: AutoViz is a powerful library that automatically visualizes any dataset, regardless of size, with just a single line of code. It offers quick assessments of dataset quality and fixes data quality issues through the FixDQ() function, making it an essential tool for both beginners and experts in data analysis. Key features: - Automatically visualizes any dataset with a single line of code - Quick assessment of dataset quality - Fixes data quality issues - Easy to use and understand Tech stack: Python Why it matters: AutoViz simplifies data visualization, making it accessible to everyone, while also providing a quick way to assess and improve data quality, enhancing the overall data analysis process.

### Overview: This repository tracks the latest AI models for audio generation, starting from 2023. ### Key Features: - Latest AI models for waveform-based audio generation. - Includes releases, samples, papers, and code for each model. - Tracks the evolution of audio AI models since 2023. ### Tech Stack: - GitHub ### Why it Matters: This repository provides a comprehensive overview of the latest advancements in audio AI models, helping developers and researchers stay updated on the cutting-edge technologies in the field.

Overview: vLLM Metal is a hardware plugin designed to run vLLM on Apple Silicon Macs, utilizing MLX as the primary compute backend. This plugin unifies MLX and PyTorch under a single lowering path, enabling efficient model execution. Key features: - Unified MLX and PyTorch - Supports a growing set of models on Apple Silicon - Runs on macOS 15 (Sequoia) or later, on Apple Silicon Tech stack: Python, MLX, PyTorch Why it matters: vLLM Metal simplifies the deployment of vLLM on Apple Silicon, providing a unified approach to model execution that optimizes performance and memory usage.

Overview: 这是一个为独立开发者精心挑选的技术栈和工具集合，旨在提升开发效率、降低开发成本，并确保使用的工具在市场上具有较高的热度和可靠性。 Key features: - 提升开发效率 - 节省开发成本 - 热门工具验证 - 市场认可度高 Tech stack: TypeScript + tRPC + Next.js + AI Why it matters: 这些工具是独立开发者提升效率、降低成本、确保工具市场认可度的首选方案，帮助他们在开发过程中更加专注和高效。

Overview: InferenceX™ is an open-source platform designed to continuously benchmark and analyze the performance of the world's leading open-source inference frameworks used by major token factories and models. It offers a live dashboard for real-time performance tracking and improvement. Key features: - End-to-end Inference Serving Benchmarks - Networking & Collective Communication Benchmarks - Operator & Kernel Level Benchmarks - OSS System Level Power Modelling - Shared components - Remaining experiments Tech stack: Python Why it matters: InferenceX™ provides a live indicator of inference performance progress, tracking improvements in the world's most popular open-source inference frameworks used by major token factories and models.

### Overview: NCRF++ is an open-source toolkit for neural sequence labeling tasks, including Named Entity Recognition (NER), part-of-speech (POS) tagging, and word segmentation. It supports character LSTM/CNN, word LSTM/CNN, and softmax/CRF components, making it highly configurable and efficient. ### Key Features: - **Character LSTM/CNN**: Captures character-level features. - **Word LSTM/CNN**: Processes word-level features. - **Softmax/CRF**: Outputs the final classification. - **Configuration File**: Configures neural model structures and parameters. - **Training and Decoding**: Supports both training and decoding modes. - **N-best Decoding**: Outputs the top N most probable sequences. ### Tech Stack: Python and PyTorch ### Why it Matters: NCRF++ provides a flexible and efficient solution for sequence labeling tasks, offering comparable or better performance compared to state-of-the-art models. Its configuration flexibility and fast running speed make it suitable for both training and inference tasks. ### Repository: jiesutd/NCRFpp ### Primary Language: Python ### GitHub Stars: 1901 ### Data Format: NCRF++ supports various input formats, including character sequence and word sequence representations. ### Performance: NCRF++ models can replicate state-of-the-art results with minimal coding, making it highly configurable and efficient. ### Add Handcrafted Features: Users can easily add handcrafted features to enhance the model's performance. ### Speed: NCRF++ utilizes fully batched operations, making it highly efficient

### Overview: LLMCompiler is an efficient and effective framework for orchestrating parallel function calls with LLMs, including both open-source and close-source models. It enables users to specify tools and optional in-context examples, automatically computing an optimized orchestration for function calls, resulting in reduced latency, cost savings, and improved accuracy. ### Key Features: - Automatically identifies tasks that can be performed in parallel. - Orchestrates multi-function calling with LLMs. - Supports both open-source and close-source models. - Enables efficient execution of complex problems. - Reduces latency, cost, and improves accuracy. ### Tech Stack: Python ### Why it Matters: LLMCompiler addresses the limitations of LLMs by enabling parallel function calling, leading to improved performance and accuracy across various tasks.

Overview: This repository contains a curated list of resources related to Large Language Models for Autonomous Driving (LLM4AD), focusing on Vision-Language Models for Autonomous Driving (VLM4AD) and Vision-Language-Action models for Autonomous Driving (VLA4AD). It aims to track the latest developments in the field and provide a comprehensive overview of research papers, datasets, and tools used in the development of autonomous driving models. Key features: - Papers: A collection of research papers on LLM4AD, including VLM4AD and VLA4AD. - Datasets: Resources for datasets used in LLM4AD research, such as Bench2Drive. - Projects: Links to projects and tools related to LLM4AD. - Code: Access to code repositories where relevant projects are hosted. - Simulators: Resources for simulators used in LLM4AD research. - Publications: Links to the survey paper on LLM4AD, which includes more detailed discussions and is continuously updated. Tech stack: The repository uses Markdown for documentation and links to external resources. Why it matters: This repository serves as a valuable resource for researchers, developers, and enthusiasts interested in the latest advancements in autonomous driving technology, specifically focusing on the integration of large language models with vision and language understanding for driving tasks. It provides a platform to explore and contribute to the ongoing research in this field, ensuring that the latest developments in LLM4AD are accessible and up-to-date.

### Overview: This repository provides a practical implementation of "Bootstrap Your Own Latent" (BYOL) self-supervised learning, developed by DeepMind, in Pytorch. BYOL achieves a new state of the art performance without the need for contrastive learning and negative pairs, making it an efficient method for utilizing unlabelled image data. ### Key features: - **BYOL Implementation**: A module that can be easily integrated into any image-based neural network to start benefiting from unlabelled image data. - **Practical Application**: The repository offers a module that can be used with any image-based neural network, such as ResNet, Discriminator, or Policy Network, to improve its performance on downstream tasks. - **State-of-the-Art Performance**: BYOL surpasses SimCLR by a significant margin, demonstrating its effectiveness in self-supervised learning. - **Evolving Evidence**: The README mentions three updates that provide new evidence supporting the importance of batch normalization in BYOL's success, as well as a new paper that successfully replaced batch normalization with group norm + weight standardization. ### Tech stack: The repository uses PyTorch, a popular deep learning framework, for its implementation. ### Why it matters: BYOL is an efficient and practical method for self-supervised learning that can significantly improve the performance of image-based neural networks without the need for expensive label acquisition. By leveraging unlabelled data, it can save organizations from the high costs associated with obtaining and labeling images. ### GitHub stars: The repository has received 1,900 stars,

### Overview: This repository is a curated list of papers and resources focused on fraud, anomaly, and outlier detection using Graph/Transformer-based methods. It includes a dashboard for interactive exploration and a local RAG-based LLM chatbot for deep research. ### Key Features: - **Interactive Dashboard**: View, filter, and search papers listed in the repository. - **LLM and Transformer Papers**: A curated list of papers that utilize LLMs and Transformers for fraud detection. - **Non-Deep-Learning Graph Papers**: Papers that focus on fraud detection using non-deep learning methods. - **Toolbox**: A collection of tools and resources for fraud detection. - **Dataset**: A dataset for fraud detection. - **Survey Paper**: A survey paper on fraud detection. - **Other Resource**: Additional resources and links for further reading. ### Tech Stack: - **GitHub**: Repository hosting. ### Why It Matters: This repository provides a comprehensive overview of the latest advancements in fraud detection using Graph/Transformer methods, making it a valuable resource for developers and researchers in the field.

Overview: Jido is an autonomous agent framework for Elixir, designed for workflows and multi-agent systems. It helps you build agent systems as ordinary Elixir and OTP software, keeping agent decisions and state transitions explicit. Key features: - Agents hold state and implement `cmd/2` - Actions do work and transform that state - Signals route events into the system - Directives describe effects for the runtime to execute Tech stack: Elixir Why it matters: Jido provides a robust framework for building autonomous and distributed agent systems in Elixir, making it ideal for applications requiring complex workflows and multi-agent coordination. Repository: agentjido/jido GitHub stars: 1899

Overview: Penzai is a JAX library designed to make it easy to work with models after they have been trained, focusing on making it simple to modify, visualize, and analyze neural networks. Key features: - **Treescope**: A pretty-printer for JAX pytrees that helps understand complex models. - **JAX Tree and Array Manipulation Utilities**: Tools for modifying and analyzing pytrees. - **Declarative Combinator-Based Neural Network Library**: Models represented as data structures that can be easily modified. - **Transformer Architecture Implementation**: Supports research into interpretability, model surgery, and training dynamics. Tech stack: Python Why it matters: Penzai simplifies the process of working with neural networks, making it easier to modify, visualize, and analyze models after training, making it a valuable tool for research and development in the field of neural networks.

### Overview In this project, we present **Make-It-3D**, a method that enables high-fidelity 3D creation from a single image. This approach leverages a diffusion prior to enhance the realism of the generated 3D models, achieving significant improvements over existing methods. ### Key Features - **High-Fidelity 3D Creation**: Generate detailed 3D models from a single image. - **Diffusion Prior Supervision**: Utilize a 2D diffusion model as 3D-aware supervision. - **Two-Stage Optimization Pipeline**: First stage optimizes a neural radiance field, followed by texture refinement. - **Realistic Textures**: Incorporate textures from the reference image to improve realism. ### Tech Stack - **Python**: The primary programming language used for implementation. - **Jittor**: A deep learning framework for training the neural radiance field. - **PyTorch**: Used for training and inference of the model. ### Why It Matters - **General Object 3D Creation**: Our method can create high-quality 3D models for various objects, including complex scenes and detailed textures. - **Versatile Applications**: Enables applications such as text-to-3D creation and texture editing, making it useful in various fields like gaming, architecture, and art. ### Repository - **Repository**: `junshutang/Make-It-3D` - **GitHub Stars**: 1898 - **Description**: **"ICCV 2023" - "Make-It-3D: