Coding & Development

MagicQuill is an intelligent and interactive image editing system, officially implemented for CVPR 2025. This open-source tool offers a user-friendly interface with AI-powered suggestions and precise local editing features. Users can leverage three types of 'magic quills': an add brush to introduce details, a subtract brush to remove or redraw elements, and a color brush for precise color adjustments. The system also includes a 'Draw and Guess' feature that intelligently suggests prompts based on user strokes. With robust canvas tools for uploading, erasing, dragging, rotating, and resizing strokes, MagicQuill streamlines the editing workflow. It supports various base models for different editing styles, including realistic, fantasy, portrait, and anime, and allows for negative prompts to refine generation results. Hardware requirements include a GPU with at least 8GB VRAM, and it offers Docker container setup for isolated environments.

llm-foundry is a comprehensive open-source repository offering code for the entire lifecycle of Large Language Models (LLMs), from training and finetuning to evaluation and deployment. It is specifically designed to integrate with Composer and the MosaicML platform, providing an efficient and flexible environment for rapid experimentation. The codebase supports various LLM workloads, including data preparation, training HuggingFace and MPT models from 125M to 70B parameters, and benchmarking training throughput and MFU. It also facilitates inference by converting models to HuggingFace or ONNX formats, generating responses, and evaluating LLMs on academic or custom in-context-learning tasks. The repository includes support for DBRX and MPT models, with detailed instructions for local use and community contributions.

Meeting Assistant Flow is an open-source project built on the crewAI framework, designed to streamline the entire meeting lifecycle. It automates critical tasks such as loading meeting notes from a text file, generating actionable tasks from meeting transcripts using AI agents, and integrating these tasks with Trello for project management. Additionally, it saves new tasks to a CSV file and sends Slack notifications to keep teams informed. This flow leverages multiple AI agents to handle different aspects of the meeting workflow, offering a modular and efficient solution for automating meeting management processes. Users can customize agents, tasks, and the flow itself to fit specific organizational needs.

MeshCNN is a general-purpose deep neural network specifically designed for 3D triangular meshes, implemented using PyTorch. This framework enables advanced tasks such as 3D shape classification and segmentation by applying convolutional, pooling, and unpooling layers directly on the mesh edges. It offers a robust solution for researchers and developers working with 3D data, providing a novel approach to process geometric information. The repository includes scripts for installation, training, and testing on datasets like SHREC and Humans, making it accessible for practical application and further development in the field of geometric deep learning.

mcp-context-forge is an open-source AI Gateway, registry, and proxy designed to federate Model Context Protocol (MCP) servers, A2A servers, and REST/gRPC APIs into a unified endpoint. It offers centralized governance, discovery, and observability across AI infrastructure, optimizing agent and tool calling. Key capabilities include a Tools Gateway for MCP, REST, and gRPC translation, an Agent Gateway for A2A protocol and OpenAI/Anthropic routing, and an API Gateway with rate limiting, authentication, and retries. The tool supports extensive plugin extensibility with over 40 integrations and provides OpenTelemetry tracing for comprehensive observability. It runs as a fully compliant MCP server, deployable via PyPI or Docker, and scales to multi-cluster Kubernetes environments with Redis-backed federation and caching.

Mocha.jl is a deep learning framework for the Julia programming language, drawing inspiration from the C++ framework Caffe. Although now deprecated, it was designed for efficient training of deep and shallow convolutional neural networks, supporting optional unsupervised pre-training via stacked auto-encoders. The framework boasts a modular architecture with isolated components for layers, activation functions, solvers, and more, allowing for easy extension. Written in Julia, it offers a high-level interface for intuitive deep neural network experimentation. Mocha.jl provides multiple backends, including a portable pure Julia backend, a faster native extension backend, and a highly efficient GPU backend utilizing NVidia® cuDNN and CUDA kernels. It also supports HDF5 for data and model storage, ensuring compatibility with other computational tools, and can import Caffe model snapshots.

MockingBird is an open-source voice cloning tool designed for real-time speech generation. It allows users to clone a voice in approximately 5 seconds and generate arbitrary speech. The tool supports Chinese Mandarin and has been tested with multiple datasets, including aidatatang_200zh, magicdata, and aishell3. It is compatible with Windows, Linux, and even M1 macOS, offering flexibility for various environments. MockingBird leverages PyTorch and provides options for training custom models for encoders, synthesizers, and vocoders, or utilizing community-shared pretrained models. It offers a web server, a toolbox, and a command-line interface for generating voices.

StackBob is an agentic Identity Governance (IGA) solution designed to complete enterprise Identity and Access Management (IAM) stacks by integrating applications that typically lack SCIM, APIs, or connectors. This tool significantly reduces manual provisioning efforts and mitigates compliance risks by extending governance over previously unintegrated applications. Key features include automated identity lifecycle management for on/offboarding and RBAC, secure password sharing via encrypted vaults, and streamlined security compliance with audit-ready evidence for SOC 2 and ISO 27001. StackBob also helps finance teams cut software license waste by detecting and removing unused licenses and orphaned accounts, claiming to save up to 25% on software costs. It offers a no-code setup, goes live in hours, and connects any app in less than 48 hours without requiring API access.

Mojo AI offers an AI-powered safety management platform designed for the construction and oil & gas sectors. Its flagship product, Safety Mojo, streamlines safety processes by unifying risk, compliance, and frontline data across projects, trades, and crews. Key features include AI-scored Pre-Task Plans (PTP/JSA) for quality and risk coverage, conversational AI tools like Ask Mojo, and multilingual support. The platform automatically generates OSHA 300/301/300A logs, TRIR, DART, and lost-time reports from submitted forms and field data. It also provides real-time dashboards for visibility across sites, allowing users to prioritize audits and identify high-risk work. Mojo AI helps meet OCIP data requirements and integrates with other software via API.

Ollamac is a free and open-source native Mac application designed to seamlessly integrate with Ollama, enabling users to run and interact with various Ollama models directly on their macOS 14.0 Sonoma or later devices. The application is exclusively available from its official GitHub repository, ensuring authenticity and direct access to updates. Key features include compatibility with all Ollama models, customizable host settings, and syntax highlighting for an enhanced user experience. Ollamac prioritizes simplicity and ease of use, providing a native interface for local AI model interaction without requiring internet access once models are pulled. This makes it an ideal tool for developers, data scientists, and students looking to experiment with large language models offline.

dgl-ke is an open-source package designed for learning large-scale knowledge graph embeddings, built on top of the Deep Graph Library (DGL). It offers high performance, ease of use, and scalability, making it suitable for various machine learning tasks involving knowledge graphs. The package supports training knowledge graph embeddings using popular models like TransE, TransR, RESCAL, DistMult, ComplEx, and RotatE. Users can perform training on single machines (CPU/GPU) or distributed environments, evaluate pre-trained embeddings with link prediction tasks, and conduct inference for entity/relation linkage prediction or embedding similarity. DGL-KE is optimized for scale, capable of processing knowledge graphs with millions of nodes and billions of edges efficiently.

Project News Alan AI is an open-source code repository that showcases how to build a conversational voice-controlled React News Application using Alan AI. Alan AI is a powerful speech recognition software designed to integrate voice capabilities into various applications, enabling users to control app functionalities entirely through voice commands. This project serves as a practical tutorial, guiding developers through the process of integrating Alan AI into a React application to create interactive, voice-enabled experiences. It highlights the ease of integration and the potential for developing custom voice-controlled applications, making it a valuable resource for those looking to add advanced speech recognition features to their projects.

PyTorch Tabular offers a unified and accessible framework for applying deep learning models to tabular data. Designed with principles of low resistance usability, easy customization, and scalability, it simplifies the development and deployment of advanced models. The library integrates with PyTorch and PyTorch Lightning, enabling efficient training on both GPUs and CPUs, alongside automatic logging for experiment tracking. It supports a variety of state-of-the-art models including FeedForward Networks, NODE, TabNet, Mixture Density Networks, AutoInt, TabTransformer, GATE, GANDALF, and DANETs, as well as semi-supervised Denoising AutoEncoders. Users can also implement custom models, making it suitable for both real-world applications and research.

PyTorch-BYOL offers a robust PyTorch implementation of the Bootstrap Your Own Latent (BYOL) self-supervised learning approach. This tool is designed for researchers and developers to experiment with and apply BYOL algorithms for representation learning. It includes configurable parameters for network architecture (ResNet-18 or ResNet-50), projection and prediction heads, data transformations, and trainer settings such as batch size, momentum update, and epochs. The repository provides clear installation instructions and configuration options, making it accessible for those looking to delve into self-supervised learning without starting from scratch. It also details feature evaluation methods, including linear separability using logistic regression and KNN on datasets like STL10.

pytorch-cnn-finetune is an open-source Python library designed to simplify the process of fine-tuning pre-trained Convolutional Neural Networks (CNNs) within the PyTorch framework. It offers a streamlined approach for adapting powerful, pre-trained models, such as those from ImageNet, to new, custom image recognition challenges. The tool automatically handles the replacement of the network's top-level classifier, allowing users to focus on training the model for their specific datasets. This makes it particularly useful for researchers and developers looking to leverage state-of-the-art CNN architectures without extensive manual configuration, accelerating the development of specialized image classification solutions.

rnn is a specialized library designed for building Recurrent Neural Networks within the Torch7's nn framework. It offers functionalities to construct different types of RNN architectures, including LSTMs (Long Short-Term Memory), GRUs (Gated Recurrent Units), and BRNNs (Bidirectional Recurrent Neural Networks). This tool is particularly useful for developers and researchers working on deep learning projects that require sequential data processing and advanced neural network models. While the original repository is deprecated, its principles and functionalities laid a foundation for subsequent RNN implementations in Torch.

Sematic is an open-source platform designed for ML engineers and data scientists to develop and manage machine learning pipelines. It enables users to write complex end-to-end pipelines using simple Python code, which can then be executed locally on a laptop, in a cloud VM, or on a Kubernetes cluster to leverage cloud resources. The platform emphasizes easy onboarding with no deployment or infrastructure needed to get started, offering local-to-cloud parity. Key features include end-to-end traceability of pipeline artifacts, reproducibility of results, dynamic graphs, lineage tracking, and runtime type-checking. Sematic also provides a modern web dashboard for monitoring, tracking, and visualizing pipelines and artifacts, along with integrations for Apache Spark, Ray, Snowflake, Plotly, Matplotlib, and Pandas.

SuperGluePretrainedNetwork is a research project from Magic Leap, presented at CVPR 2020, focusing on learning feature matching using Graph Neural Networks. The core of the project is the SuperGlue network, which integrates a Graph Neural Network with an Optimal Matching layer. This architecture is specifically designed to perform matching tasks on two distinct sets of sparse image features. The repository offers both the PyTorch code implementation and pretrained weights, making it accessible for researchers and developers interested in computer vision and feature matching applications. It serves as a valuable resource for those looking to implement or build upon advanced feature matching techniques.

StellarGraph is a comprehensive Python library designed for machine learning on various types of graphs and networks. It provides a rich collection of state-of-the-art algorithms, including GraphSAGE, GCN, GAT, Node2Vec, and Metapath2Vec, enabling users to perform tasks such as representation learning for nodes and edges, classification of nodes or entire graphs, and link prediction. The library supports diverse graph structures, from homogeneous to heterogeneous and knowledge graphs, and integrates seamlessly with TensorFlow 2, Keras, Pandas, and NumPy. This makes it user-friendly, modular, and extensible, allowing for smooth interoperability with existing machine learning workflows and easy augmentation of its core algorithms.

sumo-rl is an open-source tool designed to simplify the creation and management of Reinforcement Learning (RL) environments for Traffic Signal Control using SUMO. It offers a straightforward interface, ensuring compatibility with widely used RL libraries and frameworks such as Gymnasium, PettingZoo, stable-baselines3, and RLlib. The tool supports both single-agent and multi-agent RL scenarios, allowing for flexible experimentation. Users can easily customize observation spaces and reward functions to suit their specific research or application needs. sumo-rl is particularly useful for developers and researchers focused on advancing AI agents for traffic management and optimization, providing a robust platform for simulating and evaluating different control strategies.

T-MAC is an open-source AI Frameworks & Infra tool specifically designed for efficient low-bit Large Language Model (LLM) inference on CPU/NPU architectures. It utilizes a lookup table approach to accelerate the execution of LLMs, making it suitable for deployment on resource-constrained devices. The tool supports models like BitNet and offers a significant advantage over traditional dequantization-based methods by providing faster inference speeds. T-MAC aims to optimize the performance of AI models in environments where computational resources are limited, making advanced AI capabilities more accessible and practical for a wider range of applications.

SpeedTorch is a Python library designed to optimize data transfer between CPU and GPU in PyTorch, particularly for deep learning applications. It achieves faster transfer speeds for pinned CPU to GPU tensors and GPU to CPU tensors, in some cases up to 410x faster for GPU to CPU transfers. The library is especially beneficial for training large numbers of embeddings by allowing them to be hosted on CPU RAM when idle, thereby sparing GPU RAM. It also enables the use of non-sparse optimizers like Adamax for sparse training, which is typically not supported. SpeedTorch leverages Cupy tensors and custom memory allocators to achieve its performance gains, making it a valuable tool for developers working with memory-intensive PyTorch models.

text_renderer is an open-source tool designed to generate synthetic text line images, primarily for training deep learning Optical Character Recognition (OCR) models like CRNN. It features a modular design, allowing users to easily add different components such as Corpus, Effect, and Layout. A key capability is its integration with Albumentations, providing a wide range of image augmentation effects to enhance dataset diversity. The tool supports rendering multiple corpora on a single image with varying effects, generating vertical text, and creating LMDB datasets compatible with PaddleOCR. It also includes a web-based font viewer and corpus sampler for character balance.

torchscale is a PyTorch library specifically engineered to facilitate the scaling of Transformer models, which are fundamental to modern large language models. It emphasizes key aspects such as modeling generality and capability, ensuring that the models can be applied across a wide range of tasks and perform robustly. The library also prioritizes training stability and efficiency, crucial for developing and managing large-scale foundation models. By providing tools and frameworks within the PyTorch ecosystem, torchscale aims to empower researchers and developers to build, train, and deploy increasingly complex and powerful AI models more effectively.