Projects

Overview: Katna is a comprehensive tool designed to automate the extraction of key frames, video compression, image cropping, and resizing tasks. It simplifies these time-consuming and error-prone processes, making video and image management more efficient and accurate. Key features: - Key frame extraction - Video compression - Image cropping - Smart image resizing Tech stack: Python Why it matters: Katna streamlines the video and image processing tasks, making them more efficient and accurate, thus saving time and reducing errors.

### Overview: This repository is a collection of papers and resources related to image inpainting, a technique used to fill in missing or damaged parts of an image. The inpainting process involves restoring the image by predicting the missing content based on the surrounding context. ### Key Features: - Papers and resources related to image inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and resources related to the task of inpainting. - Papers and

Overview: WP-Autoplugin is a powerful tool that allows developers to quickly create functional plugins from simple descriptions, addressing specific needs without unnecessary bloat. It brings an agentic coding workflow into WordPress, enabling a workflow designed specifically for plugins and themes. Key features: - Quickly create plugins from descriptions - Diagnose and fix installed plugins or themes - Add features to existing code without rebuilding from scratch - Create separate extension plugins using target's hooks - Understand unfamiliar codebases through ongoing conversation - Review, test, package, install, fork, apply, or roll back changes from one workspace Tech stack: PHP Why it matters: WP-Autoplugin streamlines the development process by providing a workflow that is designed specifically for plugins and themes, allowing developers to focus on the specific needs of their projects without the need to manually move code between an editor and WordPress. This tool is completely free and compatible with various coding agents, making it an essential resource for anyone working on WordPress projects.

### Overview: WarrenWen666/AI-Software-Startups 是一个致力于 AI 软件创业公司调研的 GitHub 仓库。该仓库持续更新，涵盖了从 AI 基础设施到应用领域的多个方面，包括开源和非开源公司。它提供了详尽的技术分类、最新的融资信息和多国家覆盖的创业公司列表，帮助开发者和创业者了解和分析 AI 软件创业公司的现状和创新点。 ### Key features: - 持续更新的 AI 软件创业公司列表 - 多层次覆盖 AI 基础设施和应用领域 - 开源和非开源公司分类 - 最新一轮融资金额信息 - 多国家覆盖的创业公司列表 ### Tech stack: - **持续更新**：持续加入更多的 AI 创业公司 - **多层次**：覆盖 AI 的基础设施领域到应用领域 - **技术角度**：从技术的角度来 AI 软件创业公司进行分类 - **多国家**：覆盖中国、美国和以色列等国家 - **商业价值**：提供 AI 创业公司的最新一轮融资金额 ### Why it matters: 通过调研 AI 软件创业公司，开发者和创业者可以了解当前 AI 软件领域的现状和创新点，从而为自己的项目提供灵感和指导。

### Overview: T-NER is a Python library designed for fine-tuning language models on Named Entity Recognition (NER) tasks. It provides an easy-to-use interface for model training and evaluation across various datasets, including cross-domain and multilingual options. The library integrates with Hugging Face and supports the sharing of models and datasets via the T-NER Hugging Face group. The repository has received 399 GitHub stars and was accepted to the EACL 2021 conference. ### Key Features: - **Easy Interface**: T-NER simplifies the process of fine-tuning NER models with a user-friendly API. - **Cross-Domain Evaluation**: Supports evaluation on both cross-domain and multilingual datasets. - **High Coverage**: Integrates with publicly available NER datasets and enables the integration of custom datasets. - **Web App Integration**: Models finetuned with T-NER can be deployed on a web app for visualization. - **Model Sharing**: All models and datasets are available via the T-NER Hugging Face group. - **New Dataset**: Released a new NER dataset based on Twitter, `tweetner7`, which includes fine-tuned models and more details available in the paper and repository. ### Tech Stack: Python ### Why It Matters: T-NER is a valuable tool for researchers and practitioners looking to fine-tune language models for NER tasks. It simplifies the process of model training and evaluation, making it accessible to a wide range of users. The integration with Hugging Face and the ability to share models and datasets via the group

### Overview: This repository contains code and models for creating diffusion models from scratch using PyTorch. It includes training and inference for Normal DDPM, Improved DDPM with cosine scheduler and variance prediction, DDIM for faster inference, and Classifier-Free Guidance to enhance image quality. ### Key Features: - **Normal DDPM**: A standard diffusion model. - **Improved DDPM**: Enhanced with cosine scheduler and variance prediction. - **DDIM**: Faster inference method. - **Classifier-Free Guidance**: Improves image quality. ### Tech Stack: PyTorch ### Why It Matters: This repository provides a comprehensive guide to understanding and implementing diffusion models from scratch, including training and inference methods. It's useful for developers and researchers interested in deep learning and image generation. ### Repository: gmongaras/Diffusion_models_from_scratch ### GitHub Stars: 399 ### GitHub Description: Creating a diffusion model from scratch in PyTorch to learn exactly how they work. ### Primary Language: Python ### GitHub Repository: [https://github.com/gmongaras/Diffusion_models_from_scratch](https://github.com/gmongaras/Diffusion_models_from_scratch)

### Overview: This repository serves as the official source for the "Quick Start Guide to Large Language Models - Second Edition". It contains code snippets and Jupyter notebooks that demonstrate various applications and advanced techniques in working with Transformer models and large language models (LLMs). The guide is designed to help developers and enthusiasts learn and implement LLMs effectively. ### Key features: - Jupyter notebooks for each chapter in the book. - Data sets used in the notebooks. - Images and graphs used in the notebooks. - Code snippets and notebooks for Chapter 2: Semantic Search with LLMs. - Code snippets and notebooks for Chapter 3: First Steps with Prompt Engineering. - Code snippets and notebooks for Chapter 4: The AI Ecosystem: Putting the Pieces Together. - Code snippets and notebooks for Chapter 5: Optimizing LLMs with Customized Fine-Tuning. - Code snippets and notebooks for Chapter 6: Advanced Prompt Engineering. - Code snippets and notebooks for Chapter 7: Customizing Embeddings and Model Architectures. - Code snippets and notebooks for Chapter 9: Moving Beyond Foundation Models. - Code snippets and notebooks for Chapter 10: Advanced Open-Source LLM Fine-Tuning. ### Tech stack: Jupyter Notebook ### Why it matters: This guide is essential for developers looking to understand and implement large language models effectively. It provides practical examples and step-by-step instructions to help readers get started with LLMs and fine-tuning them. The code snippets and notebooks are designed to be beginner-friendly, making it easy for developers to follow along and

Overview: Tensil is an open-source machine learning accelerator toolchain that enables the efficient execution of AI/ML models on various FPGA platforms. Key features: - Compile AI/ML models for specific TCU architectures and FPGA development platforms. - Emulate bit-accurate Tensil models for FPGA development. - Create Verilog RTL for FPGA design. - Run AI/ML models on FPGA using PYNQ and Jupyter notebooks. Tech stack: Scala, Docker, Xilinx Vitis, Vivado, PYNQ, Verilator. Why it matters: Tensil simplifies the process of deploying AI/ML models on FPGAs, making it accessible to developers and researchers without requiring deep expertise in FPGA development.

Overview: Adaptive Chunking automatically selects the best chunking method per document for RAG, optimizing retrieval completeness and answer correctness. Key features: - Evaluates multiple chunking strategies - Automatically selects the best method for each document - Consists of five intrinsic quality metrics Tech stack: Python Why it matters: Adaptive Chunking provides a flexible and modular framework for optimizing chunking methods in RAG pipelines, making it easier to extend to new domains and use cases.

Overview: This repository explains and demonstrates the simplest form of an artificial neural network, inspired by the brain's neurons. Key features: - The simplest artificial neural network explained and demonstrated. - Part of a series on neural networks, including backpropagation and backpropagation-continued. Tech stack: Go Why it matters: This project provides a foundational understanding of neural networks, making it a great starting point for developers looking to learn about artificial intelligence.

Overview: 本项目基于 LangChain/LangGraph 实现的 ReAct Agent 智能客服系统，结合 RAG 检索、多工具调用与 Streamlit 界面，适用于智能客服与报告生成场景。 Key features: - ReAct 范式：Thought → Action → Observation 循环，Agent 自主推理并决定调用哪个工具。 - RAG 检索增强：Chroma 向量库 + DashScope Embedding，支持 txt/pdf 混合加载。 - 多工具调用：天气查询、用户定位、外部数据检索、报告上下文填充。 - 动态提示词切换：Middleware 根据运行时上下文自动切换「普通问答」与「报告生成」两套 System Prompt。 - 流式对话界面：Streamlit 构建，支持流式逐字输出、历史消息留存、Agent 推理过程可见。 - 模块化结构：Agent / RAG / Model / Tools / Middleware 独立模块，YAML 配置驱动。 Tech stack: Python, LangChain, LangGraph, Chroma, Streamlit, YAML 驱动（Agent / RAG / Chroma / Prompts）。 Why it matters: 本项目旨在提供一个高效、智能的智能客服与报告生成解决方案，通过整合 LangChain/LangGraph 的强大功能，结合 RAG 检索增强、多工具调用与 Streamlit 界面，为用户提供精准、快速的

### Overview: ImageIndexer is a Python-based tool that creates an index of images and enhances their metadata using a local AI model. It allows users to add tags and captions to image metadata, making it searchable and organized locally without relying on cloud services. ### Key Features: - **Image Analysis**: Utilizes a local AI model to generate keywords and captions for each image. - **Metadata Enhancement**: Automatically edits image metadata with generated tags. - **Local Processing**: All processing is done locally on your machine. - **Multi-Format Support**: Handles a wide range of image formats, including all major raw camera files. - **User-Friendly GUI**: Includes a GUI and installer. Relies on Koboldcpp, a single executable, for all AI functionality. - **Image View**: View images step through them to review the metadata generation after they are processed. - **Simple Model Selection**: Choose a the state-of-the-art model and it will be automatically downloaded and configured. - **Completely Automatic Backend Configuration**: The AI backend (KoboldCpp) will be downloaded and configured with optimal settings. - **GPU Acceleration**: Will use Apple Metal, Nvidia CUDA, or AMD (Vulkan) hardware if available to greatly speed inference. - **Cross-Platform**: Supports Windows, macOS ARM, and Linux. - **Stop and Start Capability**: Can stop and start without having to reprocess all the files again. - **One or Two Step Processing**: Can do keywords and a simple caption in one step, or keywords and a detailed caption in two steps. - **Highly

Overview: MCP Gateway is an advanced intermediary solution for Model Context Protocol (MCP) servers that centralizes and enhances your AI infrastructure. Key features: - Centralizes and enhances your AI infrastructure - Manages the lifecycle of configured MCP servers - Intercepts requests and responses to sanitize sensitive information - Provides a unified interface for discovering and interacting with all proxied MCPs - Includes a security scanner to analyze server reputation and security risks Tech stack: Python Why it matters: MCP Gateway simplifies the management and security of multiple MCP servers, making it easier for developers to integrate and secure their AI infrastructure.

### Overview: This project is an implementation of an abstractive summarization model using the Transformer and Pointer-generator architecture. The model aims to generate high-quality summaries from long documents, addressing the shortcomings of traditional methods. ### Key features: - **Model Architecture**: Based on the Transformer and Pointer-generator models, this project introduces a new mechanism to improve the coverage mechanism, ensuring better performance on short summaries. - **Training**: The project includes detailed steps for downloading the dataset, setting up training parameters, and running the training process. It also supports multi-gpu training for better performance. - **Evaluation**: The loss curves for Transformer-Pointer generator and Transformer models are provided, showing significant improvements in loss reduction. ### Tech stack: Python ### Why it matters: This project provides a practical solution for generating high-quality abstractive summaries, addressing the limitations of existing methods. The improved coverage mechanism and efficient multi-gpu training make it a valuable tool for researchers and developers in the field of natural language processing.

### Overview: Mustango is an advanced multimodal language model designed for generating controlled music from text prompts. Leveraging Latent Diffusion Model (LDM) and Flan-T5, Mustango excels in creating music that closely matches the specified text descriptions, making it a valuable tool for creative music generation. ### Key Features: - **Text-to-Music Generation**: Converts textual descriptions into musical compositions. - **Latent Diffusion Model (LDM)**: Underlying the model for generating high-quality music. - **Flan-T5**: Used for text processing and context understanding. - **Musical Features**: Incorporates musical elements like rhythm, tempo, and harmony to ensure the generated music is as close as possible to the input text. ### Tech Stack: - **Python**: The primary programming language used for developing and training the model. - **Hugging Face**: Utilized for integrating the model into a demo and training setup. ### Why It Matters: Mustango offers a unique approach to text-to-music generation, providing high-quality music that closely matches the textual descriptions. This model is particularly useful for applications requiring precise control over the musical output, such as soundtrack creation or creative music composition. ### Repository: The project is hosted on GitHub at [https://github.com/AMAAI-Lab/mustango](https://github.com/AMAAI-Lab/mustango). ### GitHub Stars: The repository has been starred by 396 users, indicating its popularity and interest in the community. ### Installation: To get started, clone the repository and install the

### Overview: Feeds Fun is a self-hosted news reader that automatically assigns tags, scores news by tags, and filters news based on user preferences. It is designed to be a comprehensive news aggregator with a focus on user control and efficiency. ### Key Features: - **Tagging and Scoring**: Automatically assigns tags to news entries and allows users to score news by tags. - **Filtering and Sorting**: Filters and sorts news based on user-defined rules. - **Multi-/Single-User Support**: Supports both single and multi-user environments. - **Automatic Feed Management**: Manages news feeds and automatically assigns tags. - **Backend and Frontend**: Both the backend and frontend are available as packages on PyPI and NPM, respectively. - **LLM Integration**: Utilizes LLMs for advanced tag generation and filtering. ### Tech Stack: Python ### Why It Matters: Feeds Fun provides a user-friendly and efficient way to manage and read news, especially for those who want to curate their own news feeds without the need for a paid subscription. ### Repository: Tiendil/feeds.fun ### GitHub Stars: 395 ### Roadmap: [long-term plans](https://github.com/users/Tiendil/projects/1/views/1?pane=info) ### Screenshots:  ### Features: - Multi-/single-user. - Feeds management. - Automatic tag assignment for every news entry. - Rules to score news by tags. - Filter news: exclude news by tags, show

Overview: GPTPortal is a lightweight, self-hosted web app that allows developers to chat with multiple major language model providers, including OpenAI, Anthropic, Google, xAI, DeepSeek, Groq, Moonshot, Mistral, and OpenAI-compatible endpoints. It supports customizable multimodality for voice, images, and files, and is designed for pay-as-you-go billing. Key features: - Seamless switching between different models and providers with a single dropdown. - Customizable multimodality for voice, images, and files. - Pay-as-you-go by design, with no subscription and no middleman. Tech stack: Node.js and Express. Why it matters: GPTPortal is a tool that empowers developers to interact with various language models directly, without the complexity of managing multiple web interfaces or APIs. It simplifies the process of experimenting with different models and providers, making it easier to integrate these models into their projects. Repository: Zaki-1052/GPTPortal

### Overview: This repository is a curated list of Human Preference Datasets for LLM fine-tuning, Reinforcement Learning from Human Feedback (RLHF), and evaluation. It includes datasets such as OpenAI WebGPT Comparisons, OpenAI Summarization, Anthropic Helpfulness and Harmlessness Dataset (HH-RLHF), OpenAssistant Conversations Dataset (OASST1), Stanford Human Preferences Dataset (SHP), Reddit ELI5, Human ChatGPT Comparison Corpus (HC3), HuggingFace H4 StackExchange Preference Dataset, ShareGPT.com, and Alpaca. ### Key Features: - **OpenAI WebGPT Comparisons**: 20k comparisons with human-rated preference scores. - **OpenAI Summarization**: 64k summarization examples. - **Anthropic Helpfulness and Harmlessness Dataset (HH-RLHF)**: 170k human preference comparisons. - **OpenAssistant Conversations Dataset (OASST1)**: 161k conversation corpus. - **Stanford Human Preferences Dataset (SHP)**: 385K preferences over responses. - **Reddit ELI5**: 270k examples. - **Human ChatGPT Comparison Corpus (HC3)**: 60k ChatGPT answers. - **HuggingFace H4 StackExchange Preference Dataset**: 10 million StackOverflow questions and answers. - **Alpaca**: 52k instructions and demonstrations. - **GPT4All**: 1M prompt-response pairs. ### Tech

### Overview: 夫子•明察司法大模型是由山东大学、浪潮云、中国政法大学联合研发，以ChatGLM为大模型底座，基于海量中文无监督司法语料与有监督司法微调数据训练的中文司法大模型。该模型支持法条检索、案例分析、三段论推理判决以及司法对话等功能，旨在为用户提供全方位、高精准的法律咨询与解答服务。 ### Key features: - **基于法条检索回复**：结合相关法条进行回复生成，确保回复基于与问题相关的法律依据。 - **基于案例检索回复**：生成与用户提供的案情相似的案情描述及判决结果，通过真实历史案例辅助生成判决。 - **三段论推理判决**：系统自动分析案情，识别关键事实和法律法规，生成逻辑严谨的三段论式判决预测。 ### Tech stack: Python ### Why it matters: 夫子•明察司法大模型旨在为用户提供全方位、高精准的法律咨询与解答服务，通过结合法条检索、案例分析和三段论推理判决等功能，为用户提供权威、可靠和可信的法律咨询与解答。

### Overview: Axial Attention is a powerful technique for efficient multi-dimensional data attention, implemented in PyTorch. It simplifies the process of attending to multi-dimensional data by incorporating positional encoding and leveraging axial dimensions for efficient computation. ### Key Features: - **Efficient Multi-Dimensional Attention**: Axial Attention efficiently handles multi-dimensional data, such as images, videos, and more, by rotating attention heads across axial dimensions. - **Positional Encoding**: Adds positional information to the data to enhance the model's ability to understand spatial relationships. - **Simple Implementation**: The library provides a straightforward API, making it easy to integrate into existing projects. - **Versatile Applications**: Demonstrated success in applications like weather prediction and all-attention image segmentation. ### Tech Stack: - **Python**: The core programming language used for implementing the library. - **PyTorch**: The deep learning framework used to build and train the models. ### Why It Matters: Axial Attention has shown remarkable performance in various applications, particularly in fields where multi-dimensional data is prevalent, such as image processing and computer vision. Its simplicity and effectiveness have made it a valuable tool for researchers and practitioners alike. ### Repository: - **Repository**: lucidrains/axial-attention - **GitHub Stars**: 396 - **Primary Language**: Python - **GitHub Description**: Implementation of Axial attention - attending to multi-dimensional data efficiently - **Citation**: - **arXiv**: [arXiv:1912.12180](https://ar