Projects

### Key features: - Unconditional base model - Fine-tuned models for specific properties like `chemical_system`, `space_group`, `dft_mag_density`, `dft_band_gap`, `ml_bulk_modulus`, `dft_mag_density_hhi_score`, and `chemical_system_energy_above_hull` ### Tech stack: Python ### Why it matters: MatterGen is an official implementation of a generative model for inorganic materials design across the periodic table, providing a tool for fine-tuning to steer the generation towards specific property constraints. The model is based on PyTorch Geometric and supports fine-tuning with pre-trained checkpoints, making it accessible for researchers and developers alike.

### Overview: This Kotlin client for OpenAI provides a comprehensive and efficient way to interact with OpenAI's API, leveraging multiplatform capabilities and coroutines for seamless integration across different environments. ### Key Features: - **Multiplatform Support**: Integrates with Kotlin Multiplatform, allowing seamless API access across JVM, JS, and Native platforms. - **Coroutines**: Utilizes Kotlin Coroutines for asynchronous operations, improving performance and responsiveness. - **API Responses**: Supports responses from OpenAI, including text, images, embeddings, and more. - **Models**: Handles various models for different use cases, such as text generation, image creation, and more. - **Chat**: Facilitates conversational interactions through chat functionalities. - **Images**: Generates high-quality images using the OpenAI API. - **Embeddings**: Provides vector representations for text data. - **Files**: Manages file uploads and downloads. - **Fine-tuning**: Enables fine-tuning of models for specific applications. - **Moderations**: Offers moderation tools for filtering inappropriate content. - **Audio**: Supports audio processing and generation. - **Batch**: Facilitates batch processing for multiple requests efficiently. - **Beta Features**: Includes new features like assistants, threads, messages, and runs, making it a versatile tool. - **Deprecated Features**: Some older features like completions, fine-tunes, and edits are now deprecated, but still available for backward compatibility. ### Tech Stack: The project is built using Kotlin, leveraging multiplatform capabilities and coroutines for efficient and cross-platform integration. ### Why It Matters:

### Overview: This repository contains the official implementation of the paper "Score-Based Generative Modeling through Stochastic Differential Equations" by Yang Song, Jascha Sohl-Dickstein, Diederik P. Kingma, Abhishek Kumar, Stefano Ermon, and Ben Poole. The paper presents a unified framework that generalizes and improves previous work on score-based generative models through the lens of stochastic differential equations (SDEs). The code enables better understanding of existing approaches, new sampling algorithms, exact likelihood computation, uniquely identifiable encoding, latent code manipulation, and introduces new conditional generation abilities such as class-conditional generation, inpainting, and colorization. ### Key features: - Unified framework for score-based generative models - Stochastic differential equations (SDEs) for data transformation - Score matching for estimating marginal distributions - Reverse SDE for sample generation - Improved likelihood computation and exact likelihood estimation - Unique encoding and latent code manipulation - Conditional generation capabilities (class-conditional, inpainting, colorization) ### Tech stack: Python (Jupyter Notebook) ### Why it matters: This code provides a comprehensive and modular framework for score-based generative models, enabling researchers and practitioners to easily evaluate and extend existing models. The code supports training new models, evaluating sample quality, and likelihood computation, making it a valuable resource for the field of generative modeling.

Overview: GPTDiscord is a comprehensive GPT interface for Discord, offering chat, image generation, moderation, and more, powered by OpenAI's models. Key features: - ChatGPT-style conversations - Image generation using GPT Image API - Moderation using current OpenAI models - Custom indexes and knowledgebase - YouTube summarizer - Optional integrations with Qdrant/Pinecone, Replicate, and LangChain Tech stack: Python 3.11–3.13 Why it matters: GPTDiscord simplifies the integration of advanced AI capabilities into Discord, making it easier for developers to create custom chatbots and enhance their server experiences. Repository: Kav-K/GPTDiscord GitHub stars: 1850

### Overview: AI File Sorter is a cross-platform desktop application designed to organize and rename files using AI, making them more consistent and easier to find. It supports local and remote language models for improved accuracy and preview-based workflows. ### Key Features: - **AI-Driven File Organization**: AI File Sorter analyzes files and suggests meaningful, human-readable names. - **Remote Model Support**: Uses remote language models for improved accuracy and better results. - **Local Model Support**: Analyzes files locally with built-in visual language models. - **Category Suggestion**: Automatically suggests file categories based on file types and context. - **Rename Suggestions**: Provides rename suggestions based on file content and metadata. - **Consistent Sorting**: Automatically sorts files into predefined categories based on their names, types, and context. - **Optional Whitelists**: Allows you to whitelist specific files or folders to avoid changes. - **User-Controlled Changes**: You can review and adjust changes before applying any changes. ### Tech Stack: - **Programming Language**: C++ - **Platform Support**: Cross-platform support for Windows, macOS, Linux, and Vulkan, CUDA, Metal, and Windows. ### Why It Matters: AI File Sorter helps reduce clutter, improve file consistency, and make files easier to find, whether for review, archiving, or long-term storage. It combines AI suggestions with optional whitelists and user-controlled changes, ensuring a consistent and safe sorting process. ### Repository: - **Repository**: hyperfield/ai-file-sorter - **GitHub Stars**: 1848 ### Primary Language

### Overview: HunyuanVideo-I2V is a customizable image-to-video model based on HunyuanVideo, designed to accelerate open-source community exploration. This project provides official PyTorch model definitions, pre-trained weights, and inference/sampling code. It also includes LoRA training code for customizable special effects, enhancing video creation. ### Key Features: - Official PyTorch model definitions - Pre-trained weights - Inference/sampling code - LoRA training code for customizable effects ### Tech Stack: - Python ### Why it Matters: This project is a significant contribution to the open-source community, providing a powerful tool for image-to-video generation. The customizable effects and LoRA training capabilities make it highly versatile and useful for various applications.

### Overview: Responsible AI Toolbox is a suite of tools designed to enhance the responsible development and deployment of AI systems. It includes visualization widgets for model assessment, error analysis, interpretability, and fairness, empowering developers and stakeholders to make informed decisions and take data-driven actions. ### Key Features: - **Responsible AI Dashboard**: A single pane of glass for comprehensive model assessment and debugging, identifying and mitigating model errors. - **Error Analysis Dashboard**: Identifies model errors and discovers data cohorts with underperforming models. - **Interpretability Dashboard**: Understands model predictions through InterpretML-powered dashboard. - **Fairness Dashboard**: Analyzes model fairness issues using Fairlearn metrics across sensitive features and cohorts. ### Tech Stack: TypeScript ### Why It Matters: Responsible AI Toolbox helps developers and stakeholders ensure that AI systems are developed and deployed responsibly, enabling better decision-making and data-driven actions.

Overview: This repository contains implementations of various research papers on deep learning, natural language processing, computer vision, and machine learning using Python, Keras, TensorFlow, and Scikit-learn. It includes code for projects such as correlation neural networks, reasoning with neural tensor networks, common representation learning, attention-based convolutional neural networks, and more. Key features: - Implementation of correlation neural networks for computer vision tasks. - Use of neural tensor networks for knowledge base completion. - Development of common representation learning using step-based correlation multi-modal CNN. - Attention-based convolutional neural network for sentence pair modeling. - Learning to rank short text pairs with convolutional deep neural networks. - Combining neural, statistical, and external features for fake news stance identification. - A challenge dataset for open-domain question answering. - Siamese recurrent architectures for learning sentence similarity. - Convolutional neural tensor network architecture for community question answering. - Map-reduce for machine learning on multicore systems. - Teaching machines to read and comprehend. - Improved representation learning for question answer matching. - External features for community question answering. - Language identification and disambiguation in Indian mixed-script. - Construction of a semi-automated model for FAQ retrieval via short message service. Tech stack: Python, Keras, TensorFlow, Scikit-learn. Why it matters: This repository provides practical implementations of cutting-edge research in deep learning and natural language processing, making it valuable for developers looking to apply these techniques in their projects.

Overview: This repository contains the official Python implementation for our full paper at IROS 2020 and short paper at ECCVW 2020, focusing on 3D multi-object tracking (MOT) for autonomous driving and assistive robotics applications. Our project aims to provide a baseline and new evaluation metrics for 3D MOT. Key features: - 3D Multi-Object Tracking - Baseline System - New Evaluation Metrics Tech stack: Python Why it matters: Our project provides a straightforward yet accurate baseline for 3D MOT, improving performance and running at a high frame rate compared to state-of-the-art 2D MOT systems.

### Overview: Awesome Interview is a collection of curated resources and guides designed to help developers prepare for technical interviews. It includes a variety of topics such as algorithms, coding challenges, and interview preparation tips. ### Key Features: - **CS-Notes**: A repository with notes on computer science, including topics like algorithms, data structures, and performance optimization. - **HIT-Alibaba/interview**: A collection of interview preparation materials, including both written and behavioral content. - **tech-interview-handbook**: A comprehensive guide on algorithms, behavioral content, and tips for coding interviews. - **awesome-interview-questions**: A curated list of interview questions, organized by categories such as algorithms, data structures, and programming languages. - **CS-Interview-Knowledge-Map**: A map of interview knowledge, covering various topics including JavaScript, networking, and security. - **notes**: Full-stack web development notes. - **arkingc/note**: A collection of learning notes. - **Skill-Tree**: A guide for preparing for technical interviews, focusing on skills and knowledge. - **OverNote**: A note-taking application for organizing and managing notes. ### Tech Stack: The project uses Markdown, GitHub, and various GitHub features for documentation and collaboration. ### Why It Matters: This repository serves as a valuable resource for developers looking to improve their technical interview skills. It provides a structured approach to preparing for interviews across different domains, from algorithms to coding practices. The curated nature of the content ensures that users can find relevant and useful information quickly. ### Repository: The Awesome Interview repository

Overview: Gollama is a macOS/Linux tool designed for managing Ollama models, offering a user-friendly interface for listing, inspecting, deleting, copying, and pushing models. Key features: - List available models - Display metadata such as size, quantization level, model family, and modified date - Edit/Update a model's Modelfile - Sort models by name, size, modification date, quantization level, family etc - Select and delete models - Run and unload models - Inspect model for additional details - Calculate approximate vRAM usage for a model - Copy / rename models - Push models to a registry - Show running models - Has some cool bugs Tech stack: Go Why it matters: Gollama simplifies the management of Ollama models, making it easier for developers to work with and maintain their models efficiently. Repository: sammcj/gollama Primary language: Go GitHub stars: 1842 GitHub description: Go manage your Ollama models GitHub URL: https://github.com/sammcj/gollama

### Overview: This repository provides a guide on how to use pre-trained large language models for source code. It includes models like PolyCoder, which can be used with the Huggingface library, and detailed instructions on how to set up, generate code, and evaluate the models. ### Key features: - **PolyCoder Model**: A pre-trained model for source code generation. - **Huggingface Integration**: Easy integration with the Huggingface library for model usage. - **Code Generation**: Generate code snippets using the PolyCoder model. - **Evaluation**: Evaluate the performance of the models through various metrics. ### Tech stack: Python ### Why it matters: This guide is useful for developers who want to leverage pre-trained models for improving their code generation tasks, making it easier to generate high-quality code snippets and improving the efficiency of their development processes.

Overview: This repository contains the official implementation of a text-to-image diffusion model called RPG, which was accepted at ICML 2024. RPG utilizes proprietary or open-source multimodal language models (MLLMs) as prompt recaptioners and region planners, complemented by regional diffusion, achieving state-of-the-art text-to-image generation and editing capabilities. Key features: - Text-to-image diffusion model - Utilizes proprietary or open-source MLLMs as prompt recaptioners and region planners - Complementary regional diffusion for flexibility and generality - Capable of generating high-resolution images Tech stack: Jupyter Notebook Why it matters: RPG offers a powerful training-free paradigm that leverages advanced multimodal language models to generate high-quality images, making it a significant advancement in the field of text-to-image diffusion.

### Overview: Magenta RealTime 2 is a state-of-the-art open-weights model for real-time music generation, featuring a Python library, C++ inference engine, and various example applications. ### Key Features: - **Python Library (`magenta-rt`)**: Inference with JAX and MLX backends. - **C++ Inference Engine (`magentart::core`)**: Efficient streaming audio generation on Apple Silicon MacBooks. - **Example Applications**: - **AUv3 Plugin for DAWs**: All-in-one AUv3 plugin for digital audio workstations. - **DAW App**: All-in-one standalone macOS app. - **Note Control App**: App for exploring note control. - **Prompt Space App**: App for exploring the prompt space. ### Tech Stack: Python and C++. ### Why it Matters: Magenta RealTime 2 offers a powerful tool for real-time music generation, making it ideal for creating DAW plugins, embedding the model into new applications, and exploring the prompt space through various apps. It supports real-time streaming on Apple Silicon Macs and offers offline inference capabilities on NVIDIA GPUs. ### Repository: The repository is hosted on GitHub at `https://github.com/magenta/magenta-realtime`, with 1841 GitHub stars. It includes documentation, models, and a variety of example applications. ### Hardware Requirements: Real-time streaming requires Apple Silicon (M-series) devices. The model sizes `mrt2_small` and `mrt2_base` support real-time streaming on

### Overview: This repository is dedicated to collecting resources related to BERT, including papers, applications, and GitHub projects. It serves as a comprehensive collection of BERT-related materials, covering various aspects of BERT and its applications. ### Key features: - BERT-related papers and applications - GitHub repositories implementing BERT - Official TensorFlow code and pre-trained models for BERT - PyTorch implementations of BERT - Gluon-NLP for pre-training and fine-tuning - UER-py toolkit for pre-training and fine-tuning - Kashgari framework for multilingual NLP tasks - Fast-BERT library for easy BERT-based NLP models - BERT-keras implementation in Keras - Bert-chainer framework for BERT in Chainer ### Tech stack: - Python - TensorFlow - PyTorch - Gluon-NLP - UER-py - Kashgari - Chainer ### Why it matters: BERT has revolutionized the field of natural language processing (NLP) by providing state-of-the-art performance on various downstream tasks. This repository serves as a valuable resource for researchers, developers, and practitioners looking to leverage BERT for their own projects. It provides a comprehensive set of tools and resources to help users understand, implement, and improve upon BERT models.

### Overview: Repo2txt is a web-based tool designed to convert the contents of GitHub repositories into a single formatted text file. It supports public and private repositories, local files, and zip uploads, making it accessible and efficient for developers. ### Key Features: - **Multiple Sources**: Supports GitHub, local files, and GitLab. - **Smart Filtering**: Extensive file type selection, Gitignore support, and custom patterns. - **Performance**: Virtual scrolling, lazy loading, and efficient memory usage. - **Modern UX**: Dark mode, responsive design, real-time token counting, and file statistics. - **Privacy First**: Browser-based, no server uploads, and secure token handling. ### Tech Stack: - **Framework**: React 19 + TypeScript - **Build Tool**: Vite 5 - **Styling**: Tailwind CSS 3 - **State Management**: Zustand - **File Handling**: JSZip - **Tokenization**: gpt-tokenizer - **Virtual Scrolling**: TanStack Virtual - **Testing**: Vitest + Playwright - **CI/CD**: GitHub Actions ### Why It Matters: Repo2txt simplifies the process of converting repository contents into a text format, which is essential for generating LLM prompts. It offers a fast, browser-based solution that is accessible and efficient, making it a valuable tool for developers working with GitHub repositories.

Overview: Spirula Studio is a powerful tool for training 3D Gaussian Splatting models, seamlessly converting raw video or photos into detailed textures for meshes. It supports cross-vendor GPUs and runs efficiently on Windows, Linux, and macOS, making it accessible across various platforms. Key features: - Cross vendor support via Vulkan compute - One strategy combining MCMC/IGS+/MRNF for improved results - VRAM efficiency with quantized training - Native support for 360° cameras and equirectangular images - Built-in SfM, AI masking, and frame extraction from videos - Lightning-fast sparse reconstruction and feature detection Tech stack: C++ Why it matters: Spirula Studio simplifies the process of converting video or photos into detailed textures for 3D models, making it an essential tool for anyone working with 3D splatting or similar applications. Its cross-vendor support and efficient VRAM usage make it a versatile solution for both professional and hobbyist users.

### Overview: AI as Workspace is an elegant AI chat client designed for a seamless and consistent experience across multiple platforms, including Windows, Linux, Mac OS, Android, and web (PWA). It supports multiple AI providers, including OpenAI, Anthropic, Google, DeepSeek, xAI, Azure, and more, ensuring a versatile and powerful chat environment. The client features a user-friendly interface, allowing for modifications and regenerations of messages as branches, customizable keyboard shortcuts, and support for multiple workspaces. Data is stored locally first, offering offline capabilities, and supports cloud synchronization for cross-device collaboration. The design includes features such as text file attachments, code pasting, and responsive data synchronization across multiple tabs in the same browser. ### Key Features: - **Consistent Experience Across All Platforms**: Supports Windows, Linux, Mac OS, Android, and web (PWA) platforms. - **Multiple AI Providers**: Integrates with OpenAI, Anthropic, Google, DeepSeek, xAI, Azure, and more. - **Conversation Interface**: User input preview, modifications presented as branches, customizable keyboard shortcuts, and quick scrolling to the beginning/end of a message. - **Multiple Workspaces**: Create multiple workspaces to separate conversations by themes, group workspaces into folders, and create multiple assistants within a workspace or global assistants. - **Data Storage**: Data is stored locally first, accessible offline, and loads instantly. Cloud synchronization is available after login for cross-device syncing. - **Design Details**: Supports text files as attachments, allows pasting code from VSCode without cluttering

### Overview: This repository is a curated list of data mining papers focused on fraud detection, covering various conferences and topics such as network science, data science, natural language processing, and artificial intelligence. It provides a comprehensive overview of recent research in fraud detection, including methods for graph-based fraud detection, label-free approaches, and dynamic neighborhood modeling. ### Key Features: - **Network Science**: Papers on fraud detection using network science concepts. - **Data Science**: Papers on fraud detection using data science techniques. - **Natural Language Processing**: Papers on fraud detection using natural language processing methods. - **Artificial Intelligence**: Papers on fraud detection using artificial intelligence techniques. ### Tech Stack: The repository uses Python for its implementation and documentation. ### Why It Matters: This curated list is essential for researchers and practitioners in the field of fraud detection, providing a valuable resource for staying updated with the latest research and techniques in the field.

### Overview: This project showcases a machine learning algorithm applied to the classic Flappy Bird game, utilizing neural networks and a genetic algorithm to optimize the bird's flight path. ### Key features: - Implementing a neural network architecture for the Flappy Bird game. - Utilizing a genetic algorithm to train artificial neural networks. - Optimizing the bird's flap actions to maximize the distance traveled through barriers. - Using HTML5 with Phaser framework and Synaptic Neural Network library for the implementation. ### Tech stack: JavaScript, HTML5, Phaser framework, Synaptic Neural Network library. ### Why it matters: This project demonstrates the potential of machine learning in video games, showcasing how algorithms can be used to enhance gameplay mechanics.