Projects

### Overview: This project implements a deep learning model for human pose estimation, achieving superior results on benchmark datasets like COCO and MPII. ### Key features: - **HRNet (High-Resolution Network)**: A novel deep learning architecture that maintains high-resolution representations throughout the entire network. - **Multi-Stage Architecture**: Gradually adds high-to-low resolution subnetworks to form multiple stages, ensuring rich high-resolution representations. - **Parallel Fusions**: Conducts repeated multi-scale fusions to enhance the accuracy of predicted keypoint heatmaps. - **Benchmark Results**: Superior pose estimation results on COCO and MPII datasets compared to existing methods. ### Tech stack: Cuda ### Why it matters: This project provides a powerful tool for accurate and precise human pose estimation, significantly improving the accuracy and spatial precision of pose predictions.

Overview: DeepXDE is a comprehensive library designed for scientific machine learning and physics-informed learning, featuring a variety of algorithms including PINNs, DeepONets, and MFNNs. Key features: - Physics-informed neural networks (PINNs) for solving ordinary, partial, integro-differential, and fractional PDEs. - Residual-based adaptive sampling and gradient-enhanced PINNs for improving PINN accuracy. - (Physics-informed) deep operator network (DeepONet) for learning operators in physics-informed learning. - Multifidelity neural network (MFNN) for handling data with different levels of fidelity. Tech stack: Python Why it matters: DeepXDE provides a powerful toolkit for researchers and practitioners in scientific machine learning and physics-informed learning, offering a wide range of algorithms and techniques to address various computational challenges in these fields.

### Overview: RubyLLM is a comprehensive Ruby-based AI framework designed to integrate AI features seamlessly into Ruby applications. It provides models, tools, agents, and a consistent API, making it battle-tested and widely used in various industries. ### Key Features: - **Code Execution:** Run Ruby, Python, JavaScript, and shell scripts in a sandbox environment. - **Skills:** Teach workflows with Agent Skills format. - **MCP:** Connect agents to systems using server classes. - **Agents:** Reusable assistants. - **The Agentic Loop:** Drive AI with `ask_later`, `step`, and `complete?`. - **Evaluations:** Reusable datasets and model criteria. - **Chat:** Conversational AI with `RubyLLM.chat`. - **Vision:** Analyze images and videos. - **Audio:** Transcribe speech with `RubyLLM.transcribe` and generate it with `RubyLLM.speak`. - **Documents:** Ask questions about PDFs, text files, and other formats. - **OCR:** Turn documents into markdown. - **Image Generation:** Create images with `RubyLLM.paint`. - **Video Generation:** Create videos with `RubyLLM.animate`. - **Embeddings:** Generate embeddings. - **Reranking:** Order retrieval candidates by relevance. - **Moderation:** Flag content, categorize, and score. - **Judgments:** Probabilities, choices, and scores. - **Tools:** Call Ruby methods. - **Tool Approval:** Wait for human approval before running a tool. ### Tech Stack: Ruby

Overview: LMMs-Eval is a unified evaluation toolkit designed to provide reproducible, efficient, and trustworthy evaluations across various multimodal tasks, including text, image, video, and audio. Key features: - Multiple tasks across 8 domains - Efficient evaluation pipeline - Trustworthy evaluation methodology - Support for 2 new model backends - Agentic task evaluation - Video I/O overhaul - Safety/red-teaming baselines - Efficiency metrics - Streamlined flattened JSONL log output Tech stack: Python Why it matters: LMMs-Eval aims to bridge the fragmented evaluation ecosystem, providing consistent and reliable evaluations that can help model teams focus on real improvements rather than chasing noise.

Overview: LLM Foundry is a comprehensive repository for training, finetuning, evaluating, and deploying Large Language Models (LLMs) using Composer and the MosaicML platform. Key features: - Source code for models, datasets, callbacks, and utilities - Scripts for training, inference, and evaluation - Launchable workloads using MCLI and the MosaicML platform - Two DBRX models available for download Tech stack: Python Why it matters: LLM Foundry simplifies the process of experimenting with cutting-edge LLM techniques, making it accessible and efficient for developers and researchers alike.

Overview: img2dataset is a Python library designed to quickly convert large sets of image URLs into image datasets. This tool supports downloading, resizing, and packaging up to 100 million URLs in just 20 hours on a single machine. Key features: - Downloads, resizes, and packages image URLs - Supports downloading, resizing, and packaging 100 million URLs in 20 hours on one machine - Includes support for saving captions for image+caption datasets - Provides examples of datasets to download and their respective download times Tech stack: Python Why it matters: img2dataset simplifies the process of converting image URLs into datasets, making it easier for developers to work with large datasets for machine learning projects. It optimizes performance and supports a variety of datasets, making it a valuable tool for anyone looking to streamline their data processing workflows. Repository: rom1504/img2dataset Primary language: Python GitHub stars: 4458 GitHub description: Easily turn large sets of image urls to an image dataset. Can download, resize and package 100M urls in 20h on one machine.

### Overview: CLUEDatasetSearch 是一个用于搜索和提取中文NLP数据集的项目，涵盖了多种常用和非常用数据集，包括命名实体识别、问答、情感分析、文本分类、文本匹配、文本摘要、机器翻译、知识图谱、语料库和阅读理解等领域的数据集。通过这个项目，开发者可以快速找到和使用这些数据集，进行NLP研究和开发。 ### Key features: - **命名实体识别 (NER)**: 包含多个中文电子病历、人民日报、Boson等数据集。 - **问答 (QA)**: 提供多种问答数据集，如SIGHAN Bakeoff 2005等。 - **情感分析**: 包含多个情感分析数据集，如微软亚研院MSRA等。 - **文本分类**: 提供多种文本分类数据集，如CoNLL-2003等。 - **文本匹配**: 包含多个文本匹配数据集，如CLUE Fine-Grain NER等。 - **文本摘要**: 提供多种文本摘要数据集，如微软亚研院MSRA等。 - **机器翻译**: 包含多个机器翻译数据集，如微软亚研院MSRA等。 - **知识图谱**: 提供多种知识图谱数据集，如微软亚研院MSRA等。 - **语料库**: 包含多种语料库数据集，如微软亚研院MSRA等。 - **阅读理解**: 提供多种

Overview: gptme is a developer-focused AI agent that runs on any terminal, providing local tools for writing code, using the terminal, and browsing the web. It's designed to be provider-agnostic and local-first, making it suitable for a wide range of applications. Key features: - 🛠 Tools: Shell, Python, web, vision, and other local tools. - 🌖 Extensibility: Plugins, skills, and lessons for customization. - 🌐 Integrations: MCP and ACP for seamless integration with other tools. - 🚀 Autonomous Agents: Ships with shell, Python, web, and everything an agent needs. - 🛠️ Guardrails: Built-in security and performance optimizations. - 🛠️ Use Cases: Suitable for developers, coders, and anyone needing a local AI assistant. - 🚀 Getting Started: Detailed documentation and resources for beginners. Tech stack: Python, Shell, Web, Vision, and local tools. Why it matters: gptme is a powerful and versatile AI agent that can be used to enhance productivity and efficiency in various knowledge-work environments. Its local-first approach ensures that it can be easily integrated with existing tools and workflows, making it a valuable addition to any developer's toolkit. Repository: gptme/gptme GitHub stars: 4457 Primary language: Python GitHub description: Your agent in your terminal, equipped with local tools: writes code, uses the terminal, browses the web. Make your own persistent autonomous agent on top!

Overview: This is a production-grade long-term memory plugin for OpenClaw, designed to enhance AI agents with the ability to remember and recall past interactions and decisions. Key features: - Auto-recalls preferences, decisions, and project context from previous sessions. - Uses vector and BM25 full-text search with cross-encoder reranking for hybrid retrieval. - Supports multi-scope isolation for agent, user, and project memory boundaries. - Provides a production-ready CLI, backup, migration, upgrade, and export/import functionality. Tech stack: JavaScript Why it matters: This plugin helps AI agents retain and recall important information, making them more effective and personalized in their interactions with users.

### Overview: VLMEvalKit is an open-source evaluation toolkit designed specifically for large vision-language models (LVLMs). It simplifies the evaluation process by enabling one-command evaluation across various benchmarks, reducing the workload of data preparation across multiple repositories. ### Key Features: - **One-Command Evaluation**: Quickly assess LVLMs on multiple benchmarks with a single command. - **Generation-Based Evaluation**: Utilizes generation-based evaluation methods for accurate results. - **Support for 220+ Models**: Covers a wide range of models, including 80+ benchmarks. - **Customizable Evaluation**: Offers flexibility with options for both exact matching and LLM-based answer extraction. ### Tech Stack: Python ### Why It Matters: VLMEvalKit streamlines the evaluation process for large vision-language models, making it easier to assess their performance across diverse benchmarks and models. This tool is essential for researchers and developers looking to improve the evaluation of LVLMs. ### Repository: Repository: open-compass/VLMEvalKit GitHub Stars: 4445 GitHub Description: Open-source evaluation toolkit of large multi-modality models (LMMs), support 220+ LMMs, 80+ benchmarks ### GitHub Contributors: [![][github-contributors-shield]][github-contributors-link] ### GitHub Forks: [![][github-forks-shield]][github-forks-link] ### GitHub Stars: [![][github-stars-shield]][github-stars-link] ### GitHub Issues: [![][github-issues-shield]][github-issues-link

### Overview: OpenClaw中文用例大全是50个真实场景的合集，涵盖了从新手入门到高级应用的全面内容。它不仅包括了国际用例的中国适配，还包含了国内生态的特色用例和AI助理、内容创作、运维、知识管理和自动化办公等领域的应用。 ### Key Features: - **50个真实场景**：涵盖办公、内容创作、客户服务、个人助理和智能体架构等多个领域。 - **国内生态特色**：提供国内用户友好的用例和解决方案。 - **国际用例适配**：确保国际用例在中国环境中能够正常运行。 - **新手友好**：提供详细的指南和示例，帮助用户快速上手。 ### Tech Stack: 使用OpenClaw、GitHub、Markdown等技术。 ### Why it Matters: 通过这些真实场景，用户可以直观地理解如何将OpenClaw应用到实际工作中，从而提高工作效率和生活质量。

Overview: CrateDB is a distributed and scalable SQL database designed for real-time data storage and analysis, offering PostgreSQL compatibility and Lucene-based query execution. Key features: - PostgreSQL compatibility - Scalable and flexible NoSQL features - Real-time full-text search, time-series data support, geospatial data types - Horizontal scalability, fault tolerance, and self-healing capabilities - Self-rebalancing and auto-replication - Built-in support for dynamic table schemas and queryable objects - Supports standard SQL via wire protocol or HTTP API - Auto-partitioning, auto-sharding, and auto-replication - User-defined functions (UDFs) for extension of functionality Tech stack: Java Why it matters: CrateDB simplifies the storage and analysis of massive amounts of data, offering SQL-like querying capabilities and the ability to handle complex queries in real-time, making it suitable for applications requiring near real-time data processing and analysis. Repository: crate/crate GitHub stars: 4443 GitHub description: CrateDB is a distributed and scalable SQL database for storing and analyzing massive amounts of data in near real-time, even with complex queries. It is PostgreSQL-compatible, and based on Lucene.

### Overview: ADHD is a skill for coding agents that helps generate parallel divergent thoughts under different cognitive frames, scoring, pruning traps, and deepening the survivors. This tool is designed to tackle premature convergence in autoregressive reasoning, treating it as an architectural problem rather than a prompting one. ### Key Features: - **Tree-of-Thought**: Spawns N isolated reasoning processes under deliberately distorted cognitive frames, with zero shared context during divergence. - **Scoring and Pruning**: Runs a separate critic pass to score, cluster, prune traps, and deepen the survivors. - **No-Brainer for Creative and Interdisciplinary Work**: Perfect for design decisions, fuzzy debugging, naming, API surface design, strategy, and any prompt of the shape *"give me a few ways to..."*. ### Tech Stack: TypeScript ### Why It Matters: ADHD treats premature convergence in autoregressive reasoning as an architectural problem, not a prompting one. It spawns N isolated reasoning processes under deliberately distorted cognitive frames, with zero shared context during divergence, then runs a separate critic pass to score, cluster, prune traps, and deepen the survivors. Repository: UditAkhourii/adhd Primary Language: TypeScript GitHub Stars: 4444 GitHub Description: ADHD — a skill for coding agents. Tree-of-thought with pruning, built on the Claude & Codex Agent SDK. Fans out parallel divergent thoughts under different cognitive frames, scores, prunes traps, deepens the survivors. The no-brainer skill for creative and interdisciplinary work. ### Side-by-side: Baseline

Overview: This repository provides technical resources for AI developers to build applications, agents, and systems using Oracle AI Database and OCI services. Key features: - Apps: Demonstrates end-to-end implementations of AI applications, agents, and systems. - FitTracker: Gamified fitness platform. - agentic_rag: Intelligent RAG system with multi-agent Chain of Thought. - finance-ai-agent-demo: Financial services AI agent. - generative-ui-data-chat: Data chatbot with Oracle AI Database. Tech stack: Jupyter Notebook. Why it matters: This repository offers comprehensive examples and documentation to help developers build production-grade AI solutions using Oracle AI Database and OCI services.

Overview: RAGapp simplifies the integration of Agentic RAG into any enterprise, making it as easy to configure as OpenAI's custom GPTs but deployable in your own cloud infrastructure using Docker. Key features: - Configurable with OpenAI or Gemini models - Deployed using LlamaIndex - Supports both hosted and local models Tech stack: TypeScript Why it matters: RAGapp offers a streamlined solution for integrating advanced AI into enterprises, making it accessible and deployable without the complexity of custom GPTs. Repository: ragapp/ragapp GitHub stars: 4444 GitHub description: The easiest way to use Agentic RAG in any enterprise

Overview: Harness engineering is a critical practice that focuses on shaping the environment around AI agents to ensure they operate reliably in real-world workflows, particularly for long-running coding and research tasks. Key features: - Courses & Learning Resources - Foundations - Context, Memory & Working State - Context Design & Delivery - Memory & Knowledge Systems - Constraints, Guardrails & Safe Autonomy - Tool Design & Execution Boundaries - Security, Authorization & Policy - Operational Autonomy & Reliability - Specs, Agent Files & Workflow Design - Instruction Files & Formats - Spec-Driven Development - Operating Principles & Human Oversight - Evals & Observability - Evaluation Design - Verification & Quality Gates - Telemetry, Tracing & Performance - Benchmarks - Coding & Terminal Agents - Web, GUI & Computer Use - Tools, APIs & MCP - Multi-Agent, General & Interactive - Safety, Robustness & Economic Agency - Runtimes, Harnesses & Reference Implementations - Runtime Foundations & Control Layers - Sandboxes & Execution Infrastructure - Coding-Agent Harnesses - Multi-Agent Orchestration - Browser, MCP & Tool Integration - Workflow, Profiles & Asset Management Tech stack: Node.js, Python, JavaScript, and various open-source tools and frameworks. Why it matters: Harness engineering is essential for making AI agents more dependable in real-world applications, especially for tasks that require long-term reliability

Overview: This repository contains Python notebooks designed for Azure Machine Learning Python SDK examples, focusing on machine learning and deep learning applications. Key features: - Machine Learning Examples - Deep Learning Examples Tech stack: Jupyter Notebook Why it matters: This repository provides practical examples and tutorials for developers using Azure Machine Learning Python SDK, ensuring that users can quickly understand and implement various machine learning and deep learning techniques. Repository: Azure/MachineLearningNotebooks Primary language: Jupyter Notebook GitHub stars: 4440 GitHub description: Python notebooks with ML and deep learning examples with Azure Machine Learning Python SDK | Microsoft

### Overview: ALIENS EYE is an AI-powered OSINT engine designed to discover and map 840+ social media identities across the web. Utilizing machine learning and heuristic detection, it scans platforms asynchronously, extracting profile information such as display names, bios, and avatars. ### Key Features: - **840+ Platforms Scanned Asynchronously**: Scan a wide range of social media platforms in seconds. - **ML + Heuristic Detection**: A trained model blended with 30 structural signals for more accurate detection. - **Profile Extraction**: Display names, bios, and avatars pulled from each hit. - **Cross-Site Correlation**: Cluster profiles that look like the same person. - **Recursive Expansion**: Follow linked usernames out of bios and re-scan them. - **Domain Check**: Determine if a username is registered and live. - **Watch Mode**: Re-scan on an interval and alert on changes, optionally to a webhook. - **Resumable Scans**: Checkpoint progress and continue after an interruption. - **Modern Terminal UI**: Live progress, sorted result tables, and summary panels. - **MCP Server**: Expose scanning to LLM agents. - **Proxy & Tor Support**: `--proxy socks5://...` or just `--tor`. ### Tech Stack: Python ### Why It Matters: ALIENS EYE is a powerful tool for discovering and mapping social media identities across the web, making it invaluable for security researchers, ethical hackers, and anyone looking to uncover hidden information. Its advanced features and efficient scanning

Overview: TensorFlow Probability is a library that integrates probabilistic methods with deep networks, providing tools for statistical analysis and machine learning. Key features: - Joint Distributions - Probabilistic Layers - Markov chain Monte Carlo - Variational Inference - Optimizers Tech stack: TensorFlow Why it matters: TensorFlow Probability simplifies the implementation of probabilistic models and inference algorithms, making it easier to build and deploy machine learning solutions that incorporate uncertainty and statistical reasoning.

### Overview: This repository is an extensive collection of resources and projects related to the field of machine learning and artificial intelligence. It includes system designs, tutorials, papers, and more, providing a comprehensive guide for developers and researchers in the field. ### Key Features: - **System Design Tutorials**: Video tutorials on various system design topics. - **Machine Learning Infrastructure**: Resources for machine learning training, inference, and infrastructure. - **Large Language Model (LLM) Infrastructure**: Papers and resources for training and serving LLMs. - **Domain-Specific Infra**: Papers and resources for specific domains like video systems, automated machine learning (AutoML), edge AI, and more. - **Conference Contributions**: Papers and resources from major conferences like OSDI, SOSP, SIGCOMM, NSDI, MLSys, and others. ### Tech Stack: - Python, TensorFlow, PyTorch, and other relevant tools and frameworks. ### Why It Matters: This repository serves as a valuable resource for anyone looking to learn about and implement machine learning and AI systems. It covers a wide range of topics from basic tutorials to advanced research papers, making it a go-to resource for developers and researchers alike. ### Repository Details: - **Repository Name**: HuaizhengZhang/AI-Infra-from-Zero-to-Hero - **Stars**: 4435 - **Description**: A curated list of research in machine learning systems. Links to code if available are also present. It includes videos, papers, and more, with a team maintaining the project. A new website