Coding & Development

Awesome-World-Models is an open-source repository that serves as a curated list of academic papers focused on world models. This resource is invaluable for researchers and academics interested in general video generation, embodied AI, and autonomous driving. The repository not only lists relevant papers but also includes links to associated code implementations and related websites, making it a comprehensive hub for staying updated on the latest advancements in these fields. Its open-source nature encourages community contributions, ensuring the list remains current and extensive. This tool is particularly useful for those looking to quickly find and organize research materials, facilitating literature reviews and project development in complex AI domains.

Arabic Tokenizer Arena is a specialized platform designed for in-depth analysis of Arabic text tokenization. Users can input their own Arabic text or select from pre-made samples, then choose one or more tokenizers to observe how they split the text. The tool offers comprehensive metrics such as token count, fertility, and Out-Of-Vocabulary (OOV) rate, providing valuable insights into the tokenization process. Additionally, it generates visual representations to help users understand the tokenization results more intuitively. This tool is particularly useful for researchers, developers, and linguists working with Arabic language processing, offering a robust environment for comparing and evaluating different tokenization strategies.

Argilla Space is a free and open-source tool designed for building and iterating on datasets specifically for AI models. It can be easily deployed on the Hugging Face Hub, with Hugging Face OAuth enabled for user authentication. This platform is particularly well-suited for orchestrating community annotation initiatives, allowing multiple contributors to collaborate on data labeling tasks. Its primary purpose is to facilitate the creation and continuous improvement of high-quality datasets, which are crucial for training and refining AI models across various applications.

Chat with Tess provides an interactive platform for engaging with advanced AI assistants, specifically showcasing the capabilities of Tess-R1 models. These models are designed to produce Chain-of-Thought (CoT) reasoning, enabling them to process complex queries and deliver detailed, structured responses. Users can customize various settings, including the AI model itself and the system message, to tailor their conversational experience. The platform highlights models such as migtissera/Tess-R1-Limerick-Llama-3.1-70B and migtissera/Tess-v2.5.2-Qwen2-72B, offering a hands-on opportunity to explore the Tess-R1 series' advanced reasoning abilities. This tool is ideal for those interested in experimenting with and understanding the nuances of sophisticated AI conversational agents.

CoAdapter is an AI tool hosted on Hugging Face Spaces, focusing on model adaptation and transfer learning. It is built using Gradio, making it accessible for users to interact with. The tool operates under the OpenRAIL license, indicating its open-source nature and community-driven development. While the live website currently shows a runtime error during model downloading, suggesting it may be under maintenance or experiencing issues, its core purpose is to facilitate advanced AI model manipulation. Users interested in experimenting with or developing upon existing AI models for specific applications would find CoAdapter relevant.

EcoLogits Calculator is a web-based tool designed to help users understand the environmental impact of their AI model usage. By providing information such as the specific AI model, the number of tokens processed, and other usage details, the application calculates the estimated CO₂ emissions associated with that workload. This tool promotes awareness and encourages more responsible AI practices by making the carbon footprint of generative AI models transparent. It is hosted on Hugging Face Spaces, making it easily accessible for anyone interested in assessing the ecological impact of their AI projects.

EdgeTAM is an on-device executable variant of the SAM (Segment Anything Model) designed for promptable segmentation and tracking within videos. This open-source tool allows users to upload a video, select a specific object, and then generates a masked video that highlights and tracks the chosen object throughout the footage. Optimized for efficiency, EdgeTAM achieves real-time performance on mobile devices like iPhones, making it highly suitable for mobile AI applications where immediate processing is crucial. Its core capability lies in providing precise object tracking and segmentation directly on the device, reducing latency and reliance on cloud infrastructure.

EmbeddingGemma Tuning Lab is a web-based interface built using the Gradio framework, designed for fine-tuning EmbeddingGemma models. This application enables users to customize the EmbeddingGemma model to better understand their personal tastes and specific data. It provides a platform to adapt the model for various applications, such as mood reading or other personalized tasks. The tool is hosted on Hugging Face Spaces, making it accessible through a web browser for multiple users to interact simultaneously. It offers a practical way for developers and data scientists to tailor pre-trained models to their unique requirements.

rnnlib is an open-source recurrent neural network library designed for sequence learning problems, building upon Alex Graves's foundational work. It provides implementations for tasks like online handwriting prediction and synthesis, demonstrating the capabilities of recurrent neural networks, particularly LSTM networks, in learning from sequential input. The library requires a C++11 compiler, Fortran, cmake, libcurl, automake, libtool, and texinfo for building. Auxiliary scripts in the 'utils' directory require Python packages such as SciPy, PyLab, and PIL, while experiments in 'examples' need ScientificPython for NetCDF data manipulation. It offers features like optimized LSTM layers, RMSprop optimizer, and configurable output layers with Gaussian mixtures, making it a robust tool for researchers and developers working with sequence data.

Rogue is an AI Agent Evaluator & Red Team Platform designed to stress-test AI agents for both compliance and security vulnerabilities. It offers automatic evaluation against business policies and expected behaviors, allowing users to define scenarios, verify compliance, and monitor live conversations. Additionally, Rogue provides robust red teaming capabilities, simulating over 75 vulnerabilities across 12 security categories and 20 attack techniques, with CVSS-based risk scoring. It supports 8 compliance frameworks, including OWASP, MITRE, and NIST. The platform operates on a client-server architecture with TUI and CLI interfaces, supporting various protocols and offering reproducible scans for regression testing and security fixes.

The Frontier AI Cybersecurity Observatory is a platform designed to collect and evaluate AI capabilities within the cybersecurity domain. It offers a comprehensive leaderboard that allows users to explore cybersecurity data by filtering through various benchmarks and models. This tool is crucial for understanding emerging impacts and risks associated with AI in cybersecurity. Built with Gradio, it provides an interactive interface for selecting specific aspects of cybersecurity work and inputting model or agent data for evaluation.

Safe-Reinforcement-Learning-Baselines is a GitHub repository dedicated to advancing safe reinforcement learning (RL) research. It serves as a central hub for exploring and comparing different safe RL baselines and benchmarks, encompassing both single-agent and multi-agent reinforcement learning scenarios. The repository is actively maintained and welcomes contributions from the community, encouraging users to add new papers or suggest improvements. It organizes its content into supported environments, safe RL baselines, surveys, theses, books, and tutorials, making it a valuable resource for researchers and practitioners in the field.

GPU Poor LLM Arena is a platform designed for the comparison and evaluation of compact language models, specifically those with up to 14 billion parameters. It offers a battle arena format where users can input a text prompt and receive side-by-side answers from two different language models. This setup facilitates direct comparison, allowing users to vote for the better reply and contribute to a community-driven ranking. The tool is ideal for researchers, developers, and enthusiasts interested in understanding the practical performance of smaller, more resource-efficient AI models without requiring extensive GPU resources. It provides insights into the capabilities of frugal AI options.

cosine_metric_learning offers a repository with code for training a metric feature representation, specifically tailored for person re-identification tasks. This tool is intended to be used in conjunction with the deep_sort tracker, implementing the approach described in the 'Deep Cosine Metric Learning for Person Re-identification' paper. It includes functionalities to train models on datasets like Market1501 and MARS, with options for different loss modes such as cosine-softmax. Users can monitor training progress and evaluation metrics using TensorBoard, export features for testing, and freeze trained models for deployment with Deep SORT. The repository provides detailed instructions for setting up datasets, initiating training, and evaluating model performance.

semisup-learn is a Python framework designed for semi-supervised learning, enabling the use of scikit-learn classifiers with datasets that are only partially labeled. It features implementations of Contrastive Pessimistic Likelihood Estimation (CPLE), a 'safe' framework applicable to classifiers that can yield prediction probabilities, ensuring model performance isn't worse than supervised-only training. The framework also includes Self Learning (self-training) and a wrapper for Semi-Supervised Support Vector Machine (S3VM) for comparison. CPLE is noted for its general applicability, low memory footprint, and reliance only on assumptions made by the chosen classifier, though it has high computational complexity. The project is an early-stage research endeavor.

Facetorch App is a Python library designed for comprehensive facial analysis, available as a Hugging Face Space. It allows users to upload photos or use a webcam to detect faces, generate 3D facial landmarks, and analyze various facial attributes. The app provides detailed reports on detected facial expressions, action units, and emotion scores. It also includes capabilities for extracting facial embeddings and performing face recognition. This tool is particularly useful for developers and researchers in computer vision who require advanced facial analysis functionalities for their projects.

Geocalc MCP is an AI-powered geospatial tool developed during the Agents-MCP-Hackathon, designed to execute various geo-calculations independently, without relying on external third-party APIs. This application offers core functionalities such as converting addresses into precise geographical coordinates, calculating distances between points, and planning optimal routes. Users can also visualize these calculations and routes on maps, and identify nearby points of interest. It provides a self-contained solution for geospatial computations, making it suitable for projects requiring independent geo-processing capabilities.

samples-for-ai is a comprehensive collection of deep learning samples and projects designed to help beginners get started with deep learning. It encompasses a wide range of classic deep learning algorithms and applications, supporting multiple frameworks including TensorFlow, CNTK (BrainScript and Python), PyTorch, Caffe2, Keras, MXNet, Chainer, and Theano. The project offers samples in Visual Studio solution format, making it accessible for users leveraging Microsoft Visual Studio Tools for AI or Open Platform for AI. Users can run samples locally or submit jobs to OpenPAI, providing flexibility in deployment. This open-source initiative encourages contributions and adheres to the Microsoft Open Source Code of Conduct, fostering a collaborative environment for deep learning development.

SPTAG (Space Partition Tree And Graph) is an open-source library developed by Microsoft Research and Microsoft Bing, designed for large-scale vector approximate nearest neighbor search. It represents samples as vectors and compares them using L2 or cosine distances. SPTAG offers two primary methods: kd-tree (SPTAG-KDT) for efficient index building and balanced k-means tree (SPTAG-BKT) for superior search accuracy in high-dimensional data. Key features include fresh updates for online vector deletion and insertion, and distributed serving across multiple machines. The library is inspired by the NGS approach and uses k-nearest neighborhood graphs for enhanced connectivity, with balanced k-means trees replacing kd-trees for improved accuracy with high-dimensional vectors. It provides an iterative search process combining tree and graph searches.

Stable-Baselines3 (SB3) is a robust open-source library offering reliable implementations of reinforcement learning (RL) algorithms built on PyTorch. It serves as the next major version of Stable Baselines, aiming to facilitate the replication, refinement, and identification of new ideas within the RL community and industry. SB3 provides a common interface, supports custom environments and policies, and includes features like Tensorboard integration, custom callbacks, and high code coverage. While designed for ease of use, it assumes some prior knowledge of RL concepts. The library is actively maintained for bug fixes and documentation updates, with newer algorithms and faster variants developed in associated repositories like SB3 Contrib and SBX (SB3 + Jax).

Stagehand is an AI browser automation framework designed to control web browsers using both natural language and code. It addresses the limitations of existing tools by offering a hybrid approach, allowing developers to choose between AI-driven navigation for unfamiliar pages and precise code for known actions. This flexibility makes web automation more maintainable and reliable. Key features include the ability to preview AI actions, cache repeatable actions to save time and tokens, and a self-healing mechanism that remembers previous actions and involves AI when website changes break an automation. Stagehand is open-source and provides an optimized, low-level interface to the browser built for automation.

StarVector is a multimodal vision-language model designed for Scalable Vector Graphics (SVG) generation, capable of performing both image-to-SVG and text-to-SVG conversions. Unlike traditional vectorization methods that often produce artifacts or struggle with diverse SVG primitives, StarVector operates directly in the SVG code space, leveraging visual understanding to create compact and semantically rich outputs. It has been trained on SVG-Stack, a diverse dataset of 2 million samples, and evaluated on SVG-Bench across 10 datasets and 3 tasks. StarVector excels at vectorizing icons, logotypes, technical diagrams, graphs, and charts, offering state-of-the-art performance.

StableAnimator is an open-source, end-to-end ID-preserving video diffusion framework designed for high-quality human image animation. It synthesizes videos directly from a reference image and a sequence of poses, eliminating the need for post-processing tools like face-swapping or restoration. The framework incorporates a global content-aware Face Encoder and a novel distribution-aware ID Adapter to ensure identity consistency. During inference, it utilizes a Hamilton-Jacobi-Bellman (HJB) equation-based optimization to further enhance face quality. StableAnimator supports resolutions like 576x1024 or 512x512 and provides tools for human skeleton and face mask extraction, making it a comprehensive solution for pose-driven human image animation.

sports is an open-source project by SkalskiP dedicated to exploring the intersection of Computer Vision and Sports. It features various experiments, including football player tracking using YOLOv5 and ByteTrack, 3D football player pose estimation with YOLOv7, and assigning players to teams based on uniform color using GPT-4V. The project is designed for researchers and developers interested in applying advanced AI techniques to sports analytics, offering practical examples and code for implementing these vision-based solutions. It serves as a valuable resource for understanding and replicating complex computer vision tasks in a sports context.