Projects

### Overview: Instrukt is a terminal-based AI integrated environment designed for developers. It allows users to create and instruct AI agents, generate document indexes for question-answering, and attach tools to agents for enhanced functionality. ### Key Features: - **Custom AI Agents**: Design and share custom Python packages as agents. - **Document Indexes and Question-Answering**: Automatically index and retrieve data for question-answering. - **Tool Attachments**: Attach pre-defined tools or create your own to tailor workflows. - **Prompt Console**: Integrated REPL for quick interactions and feedback. ### Tech Stack: Python ### Why It Matters: Instructs developers on how to build, test, and integrate AI agents in a terminal-based environment, making it accessible and efficient for developers working in various programming languages and environments. ### Repository: blob42/Instrukt ### GitHub Stars: 328 ### GitHub Description: Integrated AI environment in the terminal. Build, test and instruct agents.

### Overview: minGRU is an implementation of the proposed minGRU model in Pytorch, focusing on the log-space numerically stable version. This model is designed to be efficient and stable, making it suitable for a wide range of applications. ### Key Features: - Log-space numerically stable version of the minGRU model. - Supports both parallel and sequential processing for efficient computation. - Includes a simple and effective model for small language models. ### Tech Stack: Pytorch ### Why it Matters: minGRU is an important contribution to the field of language modeling, offering a more stable and efficient alternative to traditional RNNs. Its log-space version ensures numerical stability, making it suitable for a variety of applications, including text generation and sequence modeling.

Overview: OpenUE is a lightweight toolkit designed for knowledge graph extraction, offering a variety of tasks such as entity relationship extraction, event extraction, and slot and intent extraction. It leverages pre-trained language models like BERT and Roberta for efficient and accurate extraction. Key features: - Entity Relationship Extraction - Event Extraction - Slot and Intent Extraction - More tasks to come - Training and Testing Interfaces - Fast Deployment for NLP Models Tech stack: Python Why it matters: OpenUE simplifies the process of knowledge graph extraction by providing a user-friendly toolkit that can be easily integrated into various NLP projects, making it an essential tool for researchers and practitioners in the field. Repository: zjunlp/OpenUE Primary language: Python GitHub stars: 328 GitHub description: [EMNLP 2020] OpenUE: An Open Toolkit of Universal Extraction from Text README excerpt: [**中文说明**](https://github.com/zjunlp/OpenUE/blob/main/README.md) | [**English**](https://github.com/zjunlp/OpenUE/blob/main/README_EN.md) <p align="center"> <a href="https://github.com/zjunlp/OpenUE/blob/main/imgs/logo.png" width="400"/></a> </p> <p align="center"> <strong>OpenUE is a lightweight toolkit for knowledge graph extraction.</strong> </p> <p align="center"> <a href="https://badge.fury.io/py/openue"> <img src

### Overview: UltraEval-Audio is the world's first open-source framework designed for comprehensive evaluation of audio foundation models. It aggregates 34 authoritative benchmarks across four major domains—speech, sound, medicine, and music—covering 10 languages and 12 task categories. The framework supports direct replication of popular models, one-click benchmark management, built-in evaluation tools, powerful and flexible features, and seamless integration with custom datasets and existing systems. ### Key Features: - **Direct Replication of Popular Models**: Detailed replication documentation and commands for popular models. - **Isolated Runtime**: Model-specific dependencies managed automatically, ensuring dependency-free inference. - **Specialized Model Evaluation Support**: Audio models for TTS, ASR, and Audio Codec, expanding evaluation coverage. - **Powerful and Flexible**: Supports preview testing, random sampling, error retries, and resume-from-breakpoint, ensuring flexibility and efficiency. - **Seamless Integration of Custom Datasets**: Supports both public benchmarks and custom dataset integration, facilitating rapid application in various scenarios. ### Tech Stack: Python ### Why It Matters: UltraEval-Audio provides a unified benchmark framework that simplifies the evaluation process for large audio models, ensuring transparency, reproducibility, and efficiency. It supports direct replication of popular models, one-click benchmark management, and seamless integration with custom datasets, making it an indispensable tool for researchers and developers in the field of audio evaluation.

Overview: The LLMRank project investigates the capability of large language models as zero-shot rankers for recommender systems, aiming to enhance the ranking performance of recommendation systems. Key features: - Utilizes LLMs as ranking models in an instruction-following paradigm. - Uses specific promptings to trigger LLMs to perceive the order of historical user behaviors. - Evaluates ChatGPT's zero-shot ranking abilities on the ML-1M dataset. Tech stack: Python Why it matters: The project contributes to the advancement of zero-shot ranking techniques in recommendation systems, potentially improving the performance of recommendation models without explicit training on the target dataset.

Overview: Test-Model-Thing is a small proof-of-concept language model designed to incorporate latent-space prediction, internal state using recurrent trace units, and byte-by-byte output. Built with MLX, it aims to be a foundational model for further development and experimentation. Key features: - Latent-space prediction - Internal state + recurrent trace units (RTUs) - Byte input/output - Continuous data streaming - Test-time training Tech stack: MLX Why it matters: This model represents a promising starting point for exploring advanced language modeling techniques, offering a foundation for future improvements and applications. Repository: jrz97619761/test-model-thing Primary language: Python GitHub stars: 326 GitHub description: A small proof-of-concept language model (not an LLM) that aims to incorporate latent-space prediction, internal state using recurrent trace units, and byte-by-byte output, built with MLX.

Overview: This project demonstrates the use of the STRIPS library in Node.js, offering a practical example of AI planning with PDDL problems. Key features: - Solves planning problems using STRIPS and PDDL - Includes examples from the Blocks World domain - Demonstrates the AI planning process with a simple domain and problem Tech stack: Node.js Why it matters: This project serves as a practical demonstration of AI planning techniques, showcasing how STRIPS can be used to automate the planning process for various domains, including the Blocks World domain, which is particularly interesting due to its complexity and the need for multiple actions and states.

Overview: Dialogbot is a comprehensive dialogue model technology that integrates search-based, task-based, and generative dialogue models. It supports networked search, domain knowledge questions, task-driven dialogue, and casual conversation, providing a user-friendly and effective solution for various applications. Key features: - Local search-based dialogue bot - Networked search-based dialogue bot - Task-based dialogue bot - Generative dialogue bot Tech stack: Python 3.6+ Why it matters: Dialogbot offers a robust and versatile solution for building intelligent dialogue systems, making it suitable for a wide range of applications such as customer service, virtual assistants, and more. Repository: shibing624/dialogbot Primary language: Python GitHub stars: 327 GitHub description: dialogbot, provide search-based dialogue, task-based dialogue and generative dialogue model.

Overview: Troll is a tool for performing sentiment analysis on text blocks and classifying users as trolls using a neural network trained on a basic dataset. Key features: - Sentiment analysis on arbitrary text blocks - User classification as trolls using a neural network - Redis-based data storage - Training and classification functions Tech stack: JavaScript Why it matters: Troll provides a simple yet effective way to identify users as trolls based on their language sentiment, making it useful for various applications, such as content moderation or social media monitoring. Repository: thisandagain/troll Primary language: JavaScript GitHub stars: 327

### Overview: DF-GAN is a simple and effective baseline for text-to-image synthesis, presented at CVPR 2022. It achieves comparable results to large pretrained diffusion models while being training-efficient and inference-efficient. ### Key features: - **Simple Architecture**: DF-GAN uses a straightforward architecture for text-to-image synthesis. - **Training-Efficient**: Requires only 3% of the training data and 6% of learnable parameters compared to large pretrained models. - **Inference-Efficient**: Can be inferred on CPU, significantly lowering hardware requirements. - **Compared to Large Pretrained Models**: Achieves comparable results to large pretrained diffusion models but with lower training and inference requirements. ### Tech stack: Python 3.8, Pytorch 1.9, NVIDIA GPU ### Why it matters: DF-GAN provides a practical and efficient solution for text-to-image synthesis, making it accessible to a wider range of users. Its training and inference efficiency make it suitable for real-world applications where hardware resources are limited. Repository: [tobran/DF-GAN](https://github.com/tobran/DF-GAN) GitHub stars: 327 GitHub description: [CVPR2022 oral] A Simple and Effective Baseline for Text-to-Image Synthesis

### Overview: qa_match是一款基于深度学习的问答匹配工具，旨在通过结合领域分类和意图匹配模型，为用户提供准确的问答结果。它支持一层和两层结构知识库问答，同时具备无监督预训练功能，提升基于知识库问答等下游任务的效果。 ### Key features: - 支持一层和两层结构知识库问答。 - 通过意图匹配模型识别用户意图。 - 融合领域分类模型和意图匹配模型的结果，提供多样化的回答方式。 - 提供无监督预训练功能，提升下游任务效果。 ### Tech stack: Python ### Why it matters: qa_match为用户提供了一种高效且准确的问答解决方案，适用于各种知识库结构和应用场景，尤其在无监督预训练方面表现出色，有效提升了基于知识库的问答任务性能。

Overview: Seedance 2.5 API, prompts, and examples for video generation. Key features: - Seedance 2.5 API documentation - Prompts for text-to-video, image-to-video, and character consistency - Camera controls and multimodal workflows - 30+ curated production-ready prompt examples Tech stack: Python Why it matters: Provides comprehensive resources for developers interested in integrating Seedance 2.5 into their projects, including API documentation, prompts, and examples for various video generation tasks.

### Overview: This repository is a curated collection of papers on Interactive World Models, organized by key research challenges. It serves as a comprehensive resource for researchers and developers interested in the field. ### Key features: - **Interactive Webpage**: Browse and search through papers in an interactive format. - **Open Source**: Papers are available for public use, including code, data, model weights, demos, and project pages. - **Not Open Source**: Papers that do not provide public resources. ### Tech stack: Python ### Why it matters: This repository provides a structured and accessible way to explore the latest advancements in Interactive World Models, making it an invaluable resource for researchers and developers in the field.

Overview: This is an CPU-only inference implementation for the DeepSeek family of large language models written in C++, based on Yet Another Language Model (YALM). Key features: - CPU inference for DeepSeek models - Utilizes YALM for efficient model loading and inference - Supports various quantization levels (Q2_K, Q3_K, Q4_K, F8E5M2, F8E4M3, FP16, BF16, FP32) Tech stack: C++ Why it matters: This implementation is a small, lean codebase compared to other inference engines, making it extra hackable and suitable for testing single-batch DeepSeek decoding performance on CPU.

### Overview: This repository contains implementations for key components of robot perception as part of MIT-Princeton's winning entry in the Amazon Robotics Challenge 2017. The system features a robust approach to robotic grasping and recognition, capable of handling a wide range of object categories and recognizing novel objects in cluttered environments. ### Key Features: - **Suction-Based Grasping**: Implements fully convolutional neural networks (FCNs) for predicting suction-based grasping affordances from RGB-D images. - **Baseline Algorithm**: Predicts suction-based grasping affordances by computing the variance of surface normals of a 3D point cloud, where lower variance indicates higher affordance. - **Parallel-Jaw Grasping**: Implements FCNs for predicting parallel-jaw grasping affordances from heightmaps created from RGB-D images. - **Baseline Algorithm**: Detects anti-podal parallel-jaw grasps by detecting "hill-like" geometric structures over a 3D point cloud. - **Image Matching**: Implements two-stream convolutional neural networks for matching observed images of grasped objects to their product images for recognition. ### Tech Stack: Torch and Lua. ### Why It Matters: This system demonstrates the capability to handle a wide range of object categories without requiring any task-specific training data for novel objects, making it highly versatile and practical for real-world applications.

Overview: VeritasGraph is an open-source Knowledge Graph and GraphRAG framework designed for building multi-hop reasoning, ontology-aware retrieval, and verifiable attribution over local or cloud-based data. It leverages nodes, edges, RDF, and linked-data to run locally or in the cloud. Key features: - Multi-hop reasoning - Ontology-aware retrieval - Verifiable attribution Tech stack: Python Why it matters: VeritasGraph provides a robust solution for knowledge graph applications, enabling accurate and verifiable reasoning over structured data, making it valuable for applications requiring high precision and accountability. Repository: bibinprathap/VeritasGraph GitHub stars: 326

Overview: This repository contains a collection of advanced GPT/LLM tricks and techniques, designed to showcase the capabilities and limitations of large language models. Key features: - Asynchronous Requests with asyncio - ColdMailer - GPT4-Compressor - Find-Waldo - Run an Local LLM - Function Calling Capability Tech stack: Python Why it matters: These tricks demonstrate the cutting-edge capabilities and potential pitfalls of modern large language models, providing valuable insights into their performance and limitations.

### Overview: OpenAI C++ is a lightweight, header-only modern C++ library that provides convenient access to the OpenAI API. This library simplifies the integration of OpenAI's services into applications written in C++. ### Key features: - **OpenAI API Integration**: Access to OpenAI's models, completions, embeddings, and more. - **Minimal Header Files**: Only two header files, reducing dependency on external libraries. - **Community-Maintained**: Regular updates and contributions from the community. ### Tech stack: C++ ### Why it matters: OpenAI C++ is essential for developers looking to integrate OpenAI's cutting-edge AI services into their C++ applications. The library simplifies the process of using the OpenAI API, making it easier to build sophisticated AI-driven applications.

Overview: Alice is a voice-first desktop AI assistant application built with Vue.js, Vite, and Electron. It features advanced memory systems, function calling capabilities, and support for local and cloud LLMs, making it a comprehensive tool for both local and remote use. Key features: - 🗣️ Voice Interaction: Fast, VAD-powered voice recognition with OpenAI/Google TTS and optional local multilingual text-to-speech. - 🧠 Memory & Context: Short-term context in Hnswlib vector DB, structured long-term facts in local DB, summarization, and emotion awareness. - 🎨 Vision & Visual Output: Screenshot interpretation, image generation, and animated video states. - 🪄 Computer Use Tools: File system browsing, shell command execution, command approvals, and local RAG. - 🚀 Function Calling: Web search, Google Calendar & Gmail integration, torrent search & download, time & date awareness, clipboard management, task scheduler, open applications & URLs, image generation, MCP server support. - 💬 Wake Word Support: Auto language detection, wake-up word support with customizable settings. - 🛠️ Custom Tools: Supports custom tools defined in JSON scripts. Tech stack: TypeScript. Why it matters: Alice offers a user-friendly interface, advanced AI capabilities, and a wide range of tools, making it a versatile and useful companion for both local and remote use.

### Overview: VecturaKit is a Swift-based vector database designed for on-device apps, offering local vector storage and retrieval using MLTensor and MLX Embedders. It provides a pluggable embedding provider system, allowing for easy integration with various embedding models and accelerators. ### Key Features: - **Pluggable Embeddings:** Supports any provider that conforms to `VecturaEmbedder`. - **Auto-Dimension Detection:** Automatically detects embedding dimensions from the configured embedder. - **On-Device Storage:** Stores and manages vector embeddings locally. - **Hybrid Search:** Combines vector similarity with BM25 text search for relevant search results. - **Pluggable Search Engines:** Implement custom search algorithms by conforming to the `VecturaSearchEngine` protocol. - **Batch Processing:** Indexes documents in parallel for faster data ingestion. - **Persistent Storage:** Automatically saves and loads document data, preserving the database integrity. ### Tech Stack: Swift 6.0+, iOS 18.0+, macOS 15.0+, watchOS 11.0+, tvOS 18.0+, visionOS 2.0+. ### Why it Matters: VecturaKit simplifies the process of integrating vector databases into iOS and macOS apps, providing a robust foundation for on-device vector storage and retrieval. Its flexibility and ease of use make it an ideal choice for developers looking to enhance their app's search functionality.