Projects

### Overview: This repository is a comprehensive survey of reinforcement learning techniques specifically applied to large reasoning models. It aims to provide an overview of recent advancements and current trends in reinforcement learning, focusing on how these techniques can be effectively utilized to enhance large language models (LLMs) and reasoning models. ### Key Features: - **Reinforcement Learning for Large Reasoning Models**: A detailed survey of recent advancements in reinforcement learning (RL) specifically applied to large reasoning models. - **OpenRSI Release**: The first OpenRSI release includes models like Frontis-MA1 (35B / 30B, with GGUF derivatives), OpenMLE stack (Gym / RL / Evo), and OpenMLE Tasks and OpenMLE SFT Traces datasets. - **Self-Improving Agents in the Era of Experience**: A survey of self-improving agents, focusing on self-to-meta-evolution. - **Memory for Agents**: A comprehensive list of papers covering breakthroughs in Context Management and Learning from Experience. - **RL for LRMs Survey**: A survey of reinforcement learning techniques for large language models. - **SSRL**: An investigation into Agentic Search RL without reliance on external search engines. - **MARTI**: A Framework for LLM-based Multi-Agent Reinforcement Training and Inference. - **TTRL**: An open-source solution for online RL on data without ground-truth labels, particularly test data. ### Tech Stack: The repository is written in LaTeX and uses the TeX language. ### Why It Matters: This survey serves as a valuable resource for researchers

Overview: news-please is an open-source, easy-to-use news crawler that extracts structured information from almost any news website. It can recursively follow internal hyperlinks and read RSS feeds to fetch both most recent and old, archived articles. Key features: - Works out of the box with just a single command - Runs as a library or as a CLI tool - Extracts headline, lead paragraph, main text, image, author(s), publication date, and language - Supports commoncrawl.org news archive for convenient article extraction Tech stack: Python Why it matters: news-please provides a powerful and user-friendly solution for extracting structured information from news articles, making it easy to integrate into various applications and projects.

Overview: WebGradients is a curated collection of 180 CSS3, Figma, Sketch, and PSD gradients, available for free use in personal and commercial projects. It includes structured JSON data for design tools and AI agents. Key features: - 180 stunning gradients in various formats - Free and fast download - Structured JSON dataset for design tools and AI agents - Machine-readable data - Figma plugin for seamless integration Tech stack: CSS, JSON, Figma, Sketch, PSD Why it matters: WebGradients provides a comprehensive set of gradients for designers, developers, and AI agents, making it easier to create visually appealing designs and enhancing the user experience.

Overview: 🤗 Evaluate is a library designed to simplify the evaluation and comparison of machine learning models and datasets. Key features: - Implementations of dozens of popular metrics - Comparisons and measurements for evaluating models and datasets - Easy way to add new evaluation modules to the Hugging Face Hub Tech stack: Python Why it matters: 🤗 Evaluate streamlines the process of evaluating models and datasets, making it easier and more standardized. It includes a variety of metrics for NLP and Computer Vision tasks, and allows for easy comparisons and measurements. The library is actively maintained and has a growing community, making it a valuable tool for researchers and developers in the field of machine learning.

### Overview: CausalNex is a Python library designed to infer causation rather than observing correlation, focusing on simplifying the process of causal reasoning and "what-if" analysis using Bayesian Networks. ### Key features: - **Causal Structure Learning**: Helps in learning causal structures from data. - **Expert Augmentation**: Allows domain experts to enhance the relationships in the graph model. - **Counterfactual Analysis**: Enables assessing the impact of potential interventions and identifying the right intervention. ### Tech stack: Python ### Why it matters: CausalNex simplifies the process of causal reasoning and counterfactual analysis, making it easier for data scientists to develop the right interventions from analytics.

### Overview: Autoformer is a novel time series forecasting model that extends the Transformer architecture to achieve series-wise connection and series-wise auto-correlation, demonstrating significant improvements in accuracy across various applications. ### Key Features: - **Series-wise Auto-Correlation Mechanism:** Employs a series-wise connection to aggregate information at the series level, enhancing inherent log-linear complexity. - **Deep Decomposition Architecture:** Renovates the Transformer to a deep decomposition architecture for progressive trend and seasonal component decomposition. - **State-of-the-Art Performance:** Achieves SOTA on six benchmarks, including 38% relative improvement over previous methods. ### Tech Stack: Jupyter Notebook ### Why It Matters: Autoformer provides a powerful solution for time series forecasting, offering significant improvements in accuracy and efficiency across various applications, making it a valuable tool for real-world applications such as energy, traffic, economics, weather, and disease forecasting.

Overview: This project proposes a method for adapting a single-step diffusion model, such as SD-Turbo, to new tasks and domains through adversarial learning, enabling efficient inference and improved performance compared to existing methods. Key features: - CycleGAN-Turbo - pix2pix-turbo - Supports unpaired and paired image-to-image translation tasks - Performs various image-to-image translation tasks, including sketch2image, day2night, and more Tech stack: Python Why it matters: This project leverages pre-trained diffusion models to achieve efficient image-to-image translation tasks, making it suitable for real-time applications and improving performance compared to existing methods.

Overview: Elia is a snappy, keyboard-centric terminal user interface designed for interacting with large language models, including popular models like ChatGPT, Claude, Llama 3, Phi 3, Mistral, and Gemma. Key features: - Local SQLite database for storing conversations - Support for multiple models including proprietary and local models - Inline and full-screen chat options - Ability to specify models via command line - Running local models via `ollama` or `LocalAI` Tech stack: Python Why it matters: Elia simplifies the interaction with LLMs by providing a keyboard-centric interface, making it efficient and fun to use, especially for those who prefer a terminal-based experience. Repository: darrenburns/elia GitHub stars: 2484 GitHub description: A snappy, keyboard-centric terminal user interface for interacting with large language models. Chat with ChatGPT, Claude, Llama 3, Phi 3, Mistral, Gemma and more.

Overview: A fully functional Pokerbot that adapts to various poker tables, using Open-CV or neural network for table scraping and genetic algorithm-based decision-making for equity calculation. Key features: - Table scraping with Open-CV or neural network - Genetic algorithm for decision-making - Monte Carlo simulation for equity calculation Tech stack: Python Why it matters: This bot is adaptable and can work on multiple poker platforms, making it a valuable tool for players looking to improve their game.

Overview: Graphormer is a general-purpose deep learning package designed for molecule modeling tasks, aiming to accelerate research and applications in AI for molecule science. Key features: - Pre-trained models on PCQM4M and PCQM4Mv2 - Supports PyG, DGL, OGB, and OCP interfaces - Fairseq backbone support - Document online Tech stack: Python Why it matters: Graphormer provides a powerful toolkit for researchers and developers to train custom models for molecule modeling tasks, making it easier to accelerate research and applications in AI for molecule science. Repository: microsoft/Graphormer GitHub stars: 2481 GitHub description: Graphormer is a deep learning package that allows researchers and developers to train custom models for molecule modeling tasks. It aims to accelerate the research and application in AI for molecule science, such as material discovery, drug discovery, etc. [Project website](https://www.microsoft.com/en-us/research/project/graphormer/). Advanced pre-trained versions of Graphormer are available exclusively on [Azure Quantum Elements](https://quantum.microsoft.com/en-us/our-story/quantum-elements-overview).

### Overview: This repository is a curated collection of resources and papers related to Human Pose Estimation, focusing on both 2D and 3D pose estimation, person generation, and real-time pose estimation techniques. ### Key features: - Papers and resources on Human Pose Estimation - 2D Pose Estimation - 3D Pose Estimation - Person Generation - Real-time Pose Estimation ### Tech stack: The repository uses GitHub for version control and documentation. ### Why it matters: This collection serves as a comprehensive guide for beginners and experts in the field of Human Pose Estimation, providing insights into the latest advancements and methodologies in the field.

Overview: Inference turns any computer or edge device into a command center for your computer vision projects, offering self-hosted fine-tuned models, access to the latest foundation models, and the ability to track, count, time, measure, and visualize predictions. Key features: - Fine-tuned models - Latest foundation models - Access to Workflows for common use-cases - Real-time monitoring and analysis - Video processing and camera management - External system and API integration - Extensibility with custom code and models - Production deployment at scale Tech stack: Python Why it matters: Inference simplifies computer vision projects by providing a powerful, self-hosted solution that integrates seamlessly with existing workflows, making it ideal for developers and researchers looking to enhance their projects with advanced computer vision capabilities.

### Overview: This repository is a curated collection of research papers on decision, classification, and regression trees, focusing on implementations from various conferences including NeurIPS, ICML, ICLR, CVPR, ICCV, ECCV, KDD, CIKM, ICDM, SDM, PAKDD, PKDD/ECML, AAAI, AISTATS, ICANN, and UAI. The papers cover a wide range of topics such as explainability, fairness, robustness, and optimization, making it a valuable resource for developers interested in decision tree research. ### Key Features: - **MaxSAT for Efficient Explanations of Tree Ensembles (AAAI 2022)** - **Key Features:** MaxSAT, Explanation, Tree Ensembles - **Why it Matters:** This paper introduces a method to efficiently explain tree ensemble models using MaxSAT, enhancing the interpretability of decision trees. - **FOCUS: Flexible Optimizable Counterfactual Explanations for Tree Ensembles (AAAI 2022)** - **Key Features:** Counterfactual Explanations, Decision Trees - **Why it Matters:** This work provides a flexible and optimal counterfactual explanation method for decision trees, improving model fairness and robustness. - **Explainable and Local Correction of Classification Models Using Decision Trees (AAAI 2022)** - **Key Features:** Local Correction, Decision Trees - **Why it Matters:** This paper presents a technique to locally correct classification models using decision trees

### Overview: This repository contains a collection of GPTs (Generative Pre-trained Transformer) prompts designed to bypass the 25 message limit or to explore GPTs without a Plus subscription. The prompts are categorized into various themes such as memes, negotiation, tech support, and more. These prompts can be used to generate text-based content or automate tasks, making them useful for creative writing, project management, and more. ### Key Features: - **Prompt Creation**: Generate text-based content using GPTs. - **File Listing**: List files in a directory and generate download links. - **Creative Writing**: Help parents understand child development. - **Game Time**: Explain board games or card games. - **Cosmic Dream**: Visualize digital art. - **Node.js Project Builder**: Create project plans and execute tasks. - **React Project Builder**: Outline, pseudocode, and code an app. - **Angular Project Builder**: Similar to Node.js Project Builder. - **Svelte Project Builder**: Similar to Angular Project Builder. - **Poe Bot Creator**: Create chatbots using GPT. - **IsHealthy GPT**: Assist with healthy food choices. - **Habit Coach**: Enhance a healthier lifestyle. - **Business Boss**: Manage business operations. - **Modern Thinker**: Focus on mental aesthetics. - **Visual Weather Artist**: Create weather reports. - **Course Creator**: Design interactive online courses. - **Watercolor Illustrator**: Create minimalist watercolor illustrations. - **What Should I Watch?**: Recommend TV shows and movies. ### Tech Stack:

Overview: Torchmetrics is a collection of machine learning metrics specifically designed for use with PyTorch applications. It provides a set of pre-defined metrics that can be easily integrated into PyTorch models, making it easier to evaluate and optimize models. Key features: - Built-in metrics for various tasks such as classification, regression, and more. - Support for distributed training, making it ideal for large-scale applications. - Compatibility with PyTorch Lightning, a popular framework for building and deploying machine learning models. - Easy to integrate into existing PyTorch projects without requiring significant changes. Tech stack: Python Why it matters: Torchmetrics simplifies the process of evaluating machine learning models by providing a comprehensive set of metrics that are optimized for PyTorch and PyTorch Lightning. This makes it an essential tool for researchers and practitioners working on PyTorch-based projects, enabling them to quickly and accurately assess model performance across different tasks.

Overview: `yek` is a fast Rust-based tool designed to serialize text-based files in a repository or directory for efficient use with Large Language Models (LLMs). Key features: - Automatically skips unwanted files using `.gitignore` rules. - Infers important files based on Git history. - Detects and ignores binary, large, and other files. - Handles piped output and writes to a temporary directory. - Supports processing multiple directories and individual files. - Configurable via a `yek.yaml` file. Tech stack: Rust Why it matters: `yek` is essential for developers looking to efficiently prepare their repositories for use with LLMs, ensuring that only the most relevant content is processed and easily consumed by AI models. Repository: mohsen1/yek GitHub stars: 2476 GitHub description: A fast Rust based tool to serialize text-based files in a repository or directory for LLM consumption.

Overview: PIKE-RAG is an innovative specialized knowledge and rationale augmented generation method designed to enhance the capabilities of Large Language Models (LLMs) by focusing on extracting, understanding, and applying domain-specific knowledge while building coherent reasoning logic. Key features: - PIKE-RAG integrates specialized knowledge extraction and rationale generation to improve the accuracy and efficiency of LLMs. - It utilizes context-aware segmentation, automatic term label alignment, and multi-granularity knowledge extraction techniques to enhance factual information retrieval and reasoning. - The framework includes document parsing, knowledge extraction, knowledge storage, retrieval, organization, reasoning, and task decomposition and coordination. Tech stack: Python Why it matters: PIKE-RAG addresses the limitations of traditional RAG systems by focusing on extracting and applying domain-specific knowledge, thereby improving the accuracy and efficiency of LLMs in real-world industrial applications.

### Overview: Pixel2Motion is an open-source Codex and Claude skill designed for logo animation, SVG animation, and AI-assisted brand motion. It converts raster logos into smooth minimal SVG, generating animated logo HTML, GIF/video previews, and motion QA evidence. The skill is suitable for various design and development scenarios, including animated logos, SVG logo reveals, and motion design projects. ### Key Features: - **Logo Animation**: Converts raster logos into smooth minimal SVG. - **SVG Animation**: Generates SVG animations for logos. - **HTML Motion Demo**: Creates interactive HTML motion demos. - **GIF/Video Previews**: Provides animated logo GIFs and videos. - **Motion QA Evidence**: Offers motion QA evidence for quality assurance. ### Tech Stack: Python ### Why It Matters: Pixel2Motion simplifies the process of converting raster logos into motion-ready SVGs, making it easier for designers and developers to create engaging and scalable brand motion projects. The skill's open-source nature ensures accessibility and collaboration across various projects and teams. ### Repository: nolangz/pixel2motion ### GitHub Stars: 2462 ### Primary Language: Python ### GitHub Description: AI logo animation skill: turn raster logos into smooth SVG animation, animated HTML demos, GIF/video previews, and motion QA evidence. The skill is designed for logo animation, SVG animation, and AI-assisted brand motion. It converts raster logos into smooth minimal SVG, generating animated logo HTML, GIF/video previews, and motion QA evidence. Use it for animated logos, SVG logo reveals, logo motion design, pixel

Overview: Writing Tools is a system-wide grammar assistant that optimizes writing across all languages, featuring AI-powered features like proofreading, rewriting, summarization, and more. It supports Windows, Linux, and macOS, and has been featured in 30+ global publications. Key features: - Proofreading - Rewriting - Summarization - Custom instructions Tech stack: Swift Why it matters: Writing Tools is the world's most intelligent system-wide grammar assistant, working seamlessly across any language and operating system, with features that have been recognized in 30+ global publications.

### Overview: Locally Uncensored is a free, open-source desktop application that combines AI chat, image generation, and coding capabilities into a single, user-friendly interface. The app runs entirely on your local machine without the need for Docker, terminal, or cloud services. ### Key Features: - **AI Chat**: Engage in conversations with AI-powered chatbots. - **Image Generation**: Create images and art directly within the application. - **Coding Agent**: Get instant feedback on code changes and suggestions. - **Integrated Interface**: All features are seamlessly integrated into a single, intuitive workspace. ### Tech Stack: - TypeScript - WebAssembly (Wasm) for performance and compatibility ### Why It Matters: Locally Uncensored simplifies the process of experimenting with AI and coding by providing a local, accessible environment. It eliminates the need for Docker, terminal, or cloud services, making it ideal for developers and AI enthusiasts who prefer a hands-on approach. ### Repository: - **Repository**: [PurpleDoubleD/locally-uncensored](https://github.com/PurpleDoubleD/locally-uncensored) - **GitHub Stars**: 2450 - **GitHub Description**: The all-in-one local AI studio for your desktop: chat, image and video generation and a coding agent in one free, open source app. Windows and Linux. No Docker, no terminal, no cloud required. ### Download: - **Windows 10 and 11**: Available as an NSIS installer or MSI installer. - **Linux**: Available as an AppImage, Debian package