Projects

Overview: OpenEdit is an open-source command-line tool designed for agent-driven video creation and editing, powered by VEED. It provides a streamlined workflow for subtitle creation, motion graphics, slide editing, and video rendering. Key features: - Transcription of videos using various providers (WhisperX, VEED, custom). - Whisper integration for transcribing media files. - Generate, sample, and background removal functionalities. - Agent-driven pipeline for efficient video creation. Tech stack: TypeScript Why it matters: OpenEdit streamlines the video creation process, making it accessible and efficient for developers and content creators alike. Its agent-driven approach ensures consistent quality and reduces the need for manual intervention, significantly speeding up the workflow. Repository: veedstudio/open-edit Primary language: TypeScript GitHub stars: 1704 GitHub description: Open-source, agent-driven editing pipeline: create subtitles, motion graphics, slides, edit and render videos.

### Overview: This repository is a curated list of research papers, datasets, and tools related to the application of machine learning techniques in compilers and program optimisation. It aims to provide a comprehensive overview of the latest advancements in this field. ### Key features: - **Machine Learning in Compiler Optimisation**: Zheng Wang and Michael O'Boyle's paper from 2018. - **Compiler Autotuning with Machine Learning**: Papers on how machine learning can be used to improve compiler options and tuning. - **Big Code and Naturalness**: Survey on machine learning for big code and natural language. - **ML for Systems Problems**: A taxonomy of machine learning applications in systems problems. - **The Deep Learning Compiler**: A comprehensive survey on the synergy of large language models and compilers. - **Compiler Stack with LLMs**: A survey on the new compiler stack combining LLMs and compilers. ### Tech stack: Unknown ### Why it matters: This repository serves as a valuable resource for researchers, developers, and practitioners interested in leveraging machine learning to enhance compiler optimization and program optimisation. It provides a platform to stay updated with the latest advancements and to contribute to the field through feedback and contributions. ### Repository: zwang4/awesome-machine-learning-in-compilers ### GitHub stars: 1704 ### GitHub description: Must read research papers and links to tools and datasets that are related to using machine learning for compilers and systems optimisation.

### Overview: GPT2 for Multiple Languages is a project focused on fine-tuning the GPT-2 model for various languages. This repository includes a 1.5B parameter Chinese pre-trained model, which is trained on a large corpus of 30GB. The model is designed to be easily integrated into various applications, making it accessible for developers. ### Key Features: - **Pretrained Model**: A 1.5B parameter Chinese pre-trained model (10,000 steps) and a 1.5B parameter Chinese pre-trained model (22,000 steps). - **Colab Demo**: A fully functional Colab notebook for quick access to the pre-trained model. - **Training**: The model is trained using Cloud TPUs, providing efficient training capabilities. - **Corpus**: The model is trained on a corpus of 30GB, sourced from THUCNews and nlp_chinese_corpus. ### Tech Stack: Python ### Why It Matters: This project is significant for developers looking to integrate language models into their applications. The pre-trained model can be easily integrated into various projects, making it a valuable tool for natural language processing tasks. ### Repository: The repository is hosted on GitHub and has been starred by 1,703 developers, indicating its popularity and utility in the community. ### Primary Language: Python ### GitHub Stars: 1,703 ### GitHub Description: GPT2 for Multiple Languages, including pretrained models. GPT2 多语言支持, 15亿参数中文预训练

Overview: This project is a research-based deep learning model designed for collaborative filtering, specifically for recommendation systems. The model uses deep AutoEncoders to enhance collaborative filtering techniques. Key features: - Deep AutoEncoders for collaborative filtering - Training with mixed precision using Tensor Cores for improved performance - Comprehensive data preprocessing and statistical analysis - Detailed tutorial and example scripts for easy implementation Tech stack: Python, Pytorch, CUDA Why it matters: This model offers a novel approach to collaborative filtering by integrating deep learning techniques, potentially improving recommendation accuracy and efficiency in recommendation systems.

### Overview: This repository serves as a hub for innovative experiments and applications using the OpenAI Vision API. It provides a space for both beginners and experts to explore the capabilities of the Vision API, share findings, and collaborate on pushing the boundaries of visual AI. ### Key features: - WebcamGPT: A chatbot using a video stream. - HotDogGPT: A simple image classification application. - Zero-shot image classifier with GPT-4V. - Zero-shot object detection with GroundingDINO + GPT-4V. - GPT-4V vs. CLIP. - GPT-4V with Set-of-Mark (SoM). - GPT-4V on Web. - Automated voiceover of NBA game. - Screenshot-to-code. ### Tech stack: Python ### Why it matters: This repository offers a wealth of resources for anyone interested in experimenting with and building on the OpenAI Vision API, showcasing its capabilities in various applications.

### Overview: Alias is a teachable "parasite" designed to enhance control over smart home assistants and improve privacy. Users can train Alias to react to custom wake-words and commands, then control their smart devices directly from the app. ### Key features: - Multiple wake-up words - Custom commands - Change gender and language of Alias - Adjust sensitivity and delay settings - Enhanced noise detection using Pocketsphinx ### Tech stack: Python ### Why it matters: Alias offers users more control over their smart assistants and privacy by allowing them to customize wake-words and commands, and adjust sensitivity and delay settings. This tool can disrupt the built-in microphone of smart devices, introducing noise to confuse the assistant's algorithms, leading to interesting interactions and unexpected answers. ### Repository: bjoernkarmann/project_alias ### GitHub stars: 1700 ### Primary language: Python ### Build Guide: 1. Download and flash the alias2.0.img file onto a micro SD card. 2. Use Etcher to flash the SD card with the image. 3. Insert the SD card into a Raspberry Pi 3A+. 4. Power on the Pi and wait for the yellow light to glow. ### Easy Setup: The software can be flashed onto a micro SD card for a quick and easy setup. Follow the instructions provided in the README for detailed steps. ### Start using Alias: 1. Connect to the local Wi-Fi of the Alias device. 2. Open a browser and enter the URL: alias.local

### Overview: Alphafold 3 is an advanced model for predicting protein structures using Pytorch. This implementation is based on the research paper published by Google DeepMind, providing a comprehensive toolset for researchers and developers. ### Key features: - **Alphafold 3 Implementation**: The repository contains the complete implementation of Alphafold 3, a state-of-the-art method for protein structure prediction. - **Pytorch Framework**: Utilizes Pytorch for efficient and scalable model training and inference. - **Lightning and Hydra Support**: A fork includes full support for Lightning and Hydra, enhancing the model's performance and ease of deployment. - **Visualization Tools**: Provides interactive visualizations of molecular structures, allowing researchers to explore and manipulate the data. - **Optimized Version**: Offers an optimized version leveraging Triton kernels, improving computational efficiency. - **Visualization of Molecules**: A visualization tool allows users to interact with the molecules used in the repository. - **PDB Dataset**: Includes a dataset for training the model, with scripts for preparing and clustering the PDB files. - **Contributions**: Acknowledges contributions from various researchers, including Joseph, Felipe, Alex, Heng, Wei Lu, Jimin, Andrei, Milot, and others, who have contributed to the development and optimization of the model. ### Tech stack: The repository uses Pytorch for the model implementation and Lightning and Hydra for enhanced support and deployment. ### Why it matters: This implementation is crucial for researchers and developers looking to advance their research in protein structure prediction, making it

Overview: This repository contains a curated set of specialized AI subagents designed to enhance development workflows with domain-specific expertise and intelligent automation. Key features: - Intelligent Auto-Delegation - Domain Expertise - Multi-Agent Orchestration - Quality Assurance - Performance Optimized Tech stack: Unknown Why it matters: These subagents provide a valuable extension to Claude Code, enhancing development efficiency and quality across various domains.

### Overview: Efficient, scalable, and enterprise-grade CPU/GPU inference server for 🤗 Hugging Face transformer models. Deploy models on NVIDIA Triton inference servers for up to 10X faster inference compared to vanilla Pytorch. ### Key features: - Heavily optimize transformer models for inference (CPU and GPU) with speedup between 5X and 10X. - Deploy models on NVIDIA Triton inference servers for 6X faster inference compared to FastAPI. - Add quantization support for both CPU and GPU. - Simple to use: optimization done in a single command line. - Supported model: any model that can be exported to ONNX (most of them). - Supported tasks: document classification, token classification (NER), feature extraction (dense embeddings), text generation. ### Tech stack: Python ### Why it matters: At Lefebvre Dalloz, we run semantic search engines in the legal domain, where latency is key for providing good user experience. We have tested many solutions, and the combination of Nvidia TensorRT and Triton inference server provides up to 10X faster inference compared to Pytorch, making it the best choice for enterprise-grade deployments.

### Overview: AutoCrawler is a high-quality, fast, and customizable image web crawler designed for Google and Naver. It supports multi-threaded downloads, supports Google and Naver sites, and offers customizable features such as full resolution images and data imbalance detection. ### Key Features: - **Multi-threaded Downloads**: Efficiently downloads images from multiple sites simultaneously. - **Customizable**: Supports Google and Naver sites, customizable features like full resolution images, and data imbalance detection. - **High Quality & Speed**: Ensures high-quality images and fast performance. - **Customizable Collectors**: Allows developers to create their own collectors for advanced customization. ### Tech Stack: - **Python**: The primary programming language used for developing AutoCrawler. - **Selenium**: Used for automating browser interactions and scraping images from websites. ### Why It Matters: - **Efficiency**: AutoCrawler is designed to handle large volumes of images quickly and efficiently. - **Customization**: The ability to customize the crawler to specific needs, making it highly versatile. - **Quality**: Ensures high-quality images, which is crucial for SEO and other image-based applications. ### GitHub Stars: - **GitHub Stars**: 1696, indicating a significant amount of interest and adoption of the project. ### Repository: - **Repository**: YoongiKim/AutoCrawler - **Primary Language**: Python - **GitHub Stars**: 1696 ### How to Use: 1. **Install Chrome**: Ensure Chrome is installed. 2. **Install Dependencies**: Use `pip install

Overview: This repository contains a collection of TensorFlow 2.0 examples, designed to help beginners easily dive into the framework. It includes both Jupyter Notebooks and source codes with detailed explanations, covering topics such as basic operations, activation functions, gradient tape, and more. Key features: - **Hello World** (print "hello world" using TensorFlow) - **Variable** (using variable in TensorFlow) - **Basical operations** (simple examples of TensorFlow operations) - **Activation functions** (introduction to activation functions) - **Gradient tape** (key technique for automatic differentiation) Tech stack: Jupyter Notebook Why it matters: This tutorial is particularly useful for beginners looking to get started with TensorFlow 2.0, providing a clear and practical introduction to the framework through hands-on examples. Repository: GitHub repository: YunYang1994/TensorFlow2.0-Examples GitHub stars: 1696 GitHub description: Difficult algorithm, Simple code.

### Overview: Speech Note is a Linux desktop and Sailfish OS app designed for note taking, reading, and translating with offline speech-to-text, text-to-speech, and machine translation capabilities. ### Key Features: - Offline speech-to-text and text-to-speech processing. - Supports multiple languages and models. - Compatible with various speech processing engines. ### Tech Stack: - C++ ### Why It Matters: - Ensures privacy and security by processing text and speech locally. - Supports multiple languages and models for diverse needs. ### Repository: mkiol/dsnote - GitHub stars: 1694 - Description: Speech Note Linux app for note taking, reading and translating with offline Speech to Text, Text to Speech and Machine Translation. - Languages and Models: - Afrikaans - Amharic - Ar - ... (more languages and models) ### How to Install: - Download from Flathub: <https://flathub.org/apps/net.mkiol.SpeechNote> ### Flatpak Packages: - Available on Flathub ### Beta Version: - Not specified ### Extra Features: - Not specified ### Building from Sources: - Not specified ### How to Enable a Custom Model: - Not specified ### Contributing to Speech Note: - Not specified ### How to Support: - Not specified ### Reviews and Demos: - Not specified ### License: - Not specified

### Overview: PyTorch Tabular is a unified framework designed for deep learning models on tabular data. It simplifies the process of training deep learning architectures by providing a high-level API and leveraging the power of PyTorch Lightning for efficient training on GPU or CPU, with automatic logging. ### Key Features: - **Unified Interface**: Provides a single API for various deep learning models. - **High-Level API**: Simplifies the process of building and training models. - **Automatic Logging**: Automatically logs training metrics and progress. - **Scalability**: Supports training on both CPU and GPU, making it accessible for a wide range of environments. ### Tech Stack: - **PyTorch**: The core framework used for deep learning. - **PyTorch Lightning**: Used for automatic training, logging, and model deployment. ### Why It Matters: PyTorch Tabular aims to democratize deep learning by making it accessible to researchers and practitioners alike. It simplifies the process of building and deploying models on tabular data, making it easier to handle complex tabular datasets in real-world applications and research projects. ### Repository: - **Repository**: pytorch-tabular/pytorch_tabular - **GitHub Stars**: 1696 - **GitHub Description**: A unified framework for Deep Learning Models on tabular data. ### Documentation: - **Documentation**: [https://pytorch-tabular.readthedocs.io/en/latest/](https://pytorch-tabular.readthedocs.io/en/latest/) - **Tutorials**: [https://pytorch-tabular.readthedocs.io/en

### Overview: Swark is a powerful tool that automatically generates architecture diagrams from code using large language models (LLMs). It seamlessly integrates with GitHub Copilot, making it free and open source with no authentication or API key required. ### Key Features: - **Automatic Architecture Diagrams**: Swark creates diagrams from code automatically. - **Free and Open Source**: Works with GitHub Copilot, no additional setup needed. - **LLM-Based Logic**: All logic is encapsulated within the LLM, supporting all languages. - **Seamless Integration**: Directly integrated with GitHub Copilot, no extra setup required. - **Privacy First**: Source code is shared only with GitHub Copilot, no external APIs involved. - **Mermaid.js Support**: Generates diagrams in Mermaid.js, making it easy to edit and refine. - **Use Cases**: Ideal for onboarding, reviewing AI-generated code, improving documentation, understanding legacy code, and spotting design flaws. ### Why Swark? - **Free and Open Source**: Free and open source with no additional setup needed. - **LLM-Based Logic**: Supports all languages with no additional language support required. - **Seamless Integration**: Directly integrated with GitHub Copilot, no extra setup required. - **Privacy First**: Source code is shared only with GitHub Copilot, no external APIs involved. - **Mermaid.js Support**: Generates diagrams in Mermaid.js, making it easy to edit and refine. - **Use Cases**: Ideal for onboarding, reviewing AI-generated code, improving documentation, understanding legacy code, and spotting design flaws

Overview: TensorFlow Addons (TFA) is a collection of additional functionality for TensorFlow 2.x, maintained by the SIG-addons community. It provides useful features and patterns that are not available in core TensorFlow, focusing on new developments in the field of machine learning. Key features: - Implement new functionality not available in core TensorFlow - Conforms to well-established API patterns - Offers a wide range of contributions, including activations, callbacks, image processing, layers, losses, metrics, optimizers, RNNs, sequence-to-sequence models, text processing, and more Tech stack: Python Why it matters: TensorFlow Addons provides valuable tools and patterns for developers working with TensorFlow, especially for new developments in the field of machine learning that may not be integrated into core TensorFlow yet. Repository: tensorflow/addons GitHub stars: 1696 GitHub description: Useful extra functionality for TensorFlow 2.x maintained by SIG-addons README excerpt: <h2 align="center"> :warning: :warning: :warning: </h2> <h4 align="center"> TensorFlow Addons (TFA) has ended development and introduction of new features. TFA has entered a minimal maintenance and release mode until a planned end of life in May 2024. Please modify downstream libraries to take dependencies from other repositories in our TensorFlow community (e.g. Keras, Keras-CV, and Keras-NLP) For more information see: [https://github.com/tensorflow/addons/issues/2807](https://github.com/tensorflow

Overview: This project demonstrates how to detect objects in Augmented Reality (AR) using Core ML and displays 3D labels above them. It serves as a basic template for an ARKit project to use Core ML. Key features: - Detecting objects in AR - Displaying 3D labels on detected objects - Using Core ML for the detection model Tech stack: Swift 4.0 Why it matters: This project provides a simple yet effective way to integrate Core ML with ARKit, showcasing the potential of combining machine learning with AR technology. Repository: hanleyweng/CoreML-in-ARKit GitHub stars: 1695 GitHub description: Simple project to detect objects and display 3D labels above them in AR. This serves as a basic Template for an ARKit project to use CoreML.

Overview: Spec Kitty is an open-source CLI designed for turning product intent into a Git-native AI coding workflow. It streamlines the process from specification to acceptance, ensuring clear work-package boundaries, parallel execution, and traceable progress. Key features: - Specification -> Plan -> Tasks -> Next -> Review -> Accept -> Merge - Repository-native mission artifacts under `kitty-specs/` - Isolated git worktrees under `.worktrees/` - Clear work-package boundaries and parallel implementation - Visible progress and review gates - Retrospective learning and governance Tech stack: Python Why it matters: Spec Kitty simplifies the transition to a governed software factory, ensuring clear communication and repeatable processes across AI coding agents and teams. Repository: spec-kitty/spec-kitty GitHub stars: 1695

Overview: Axon is a high-performance neural network library built on the Nx compiler, providing a functional API for deep learning in Elixir. Key features: - Functional API for low-level numerical definitions - Model Creation API for high-level model creation - Training API for quick model training Tech stack: Elixir, Nx Why it matters: Axon simplifies the process of creating and training neural networks in Elixir, leveraging the power of Nx for optimization and integration with other libraries.

### Overview: This repository is a curated collection of state-of-the-art research materials related to trajectory prediction, including datasets, surveys, physics systems, intelligent vehicles, pedestrians, mobile robots, and benchmark evaluation metrics. ### Key Features: - Datasets: Comprehensive selection of datasets for trajectory prediction, including Waymo Open Dataset, Argoverse, nuScenes, highD, inD, roundD, and BDD100k. - Literature: Surveys and papers on trajectory prediction, focusing on intelligent vehicles, pedestrians, mobile robots, and benchmark evaluation metrics. - Codes: Public codes and implementations for trajectory prediction models, including EvolveGraph and Conditional Generative Neural System. ### Tech Stack: State-of-the-Art Research Materials ### Why it Matters: This repository aims to provide a valuable resource for both academia and industry, offering insights into the latest research in trajectory prediction. ### Repository: jiachenli94/Awesome-Interaction-Aware-Trajectory-Prediction ### GitHub Stars: 1694 ### GitHub Description: A selection of state-of-the-art research materials on trajectory prediction ### Maintainers: Jiachen Li (Stanford University), Hengbo Ma (University of California, Berkeley), Jinning Li (University of California, Berkeley) ### Contact: jiachen_li@stanford.edu, {hengbo_ma, jinning_li}@berkeley.edu ### Citation: Please consider citing our work if you found this repo useful: ``` @inproceedings{li2020evolvegraph, title

### Overview: EdgeML is a library providing code for machine learning algorithms designed for edge devices, developed at Microsoft Research India. It focuses on models with minimal footprint to fit in a few kilobytes, essential for resource-scarce IoT devices and sensors. ### Key Features: - **Bonsai**: Strong and shallow non-linear tree-based classifier. - **ProtoNN**: Proto-based k-nearest neighbors classifier. - **EMI-RNN**: Training routine to recover critical signatures for faster and accurate RNN predictions. - **Shallow RNN**: Meta-architecture for training RNNs on streaming data. - **FastRNN & FastGRNN - FastCells**: Fast, Accurate, Stable, Tiny RNN cells. - **DROCC**: Deep Robust One-Class Classification for robust anomaly detectors. - **RNNPool**: Efficient non-linear pooling operator for RAM-constrained inference. ### Tech Stack: C++ ### Why it Matters: These algorithms can train models for classical supervised learning problems with memory requirements orders of magnitude lower than other modern ML algorithms. The trained models can be loaded onto edge devices such as IoT devices/sensors, and used to make fast and accurate predictions offline.