Projects

Overview: ofxCv is an alternative approach to interfacing with OpenCV from openFrameworks, providing a more user-friendly and efficient way to work with OpenCV functions within the openFrameworks framework. Key features: - Simplifies complex OpenCV functions into more manageable, intuitive wrappers. - Offers a smart interface for dynamic memory management, improving performance. - Provides in-place and out-of-place alternatives for handling data. - Presents the power of OpenCV in an easily understandable manner through clear class methods. - Facilitates seamless interoperability between openFrameworks and OpenCV, making it easier to integrate OpenCV functions directly. Tech stack: C++ Why it matters: ofxCv simplifies the process of working with OpenCV within the openFrameworks framework, making it more accessible and efficient for developers.

### Overview: This repository is a curated list of decision models, also known as System One models or typed decision models. It includes hosted APIs, open-weight models, runtimes, SDKs, applications, benchmarks, and papers related to these models. The models and runtimes accept Noul, Choice, and Score questions over a text state in one request, with keys from the console and served through various platforms. ### Key Features: - **Hosted APIs**: APIs-only models, oldest first. - **Open Models**: Models hosted by others, such as Clef and pplx-decider. - **Inference Techniques**: Various inference methods and techniques. - **Runtimes & Platforms**: Platforms and runtimes that support these models. - **SDKs & Clients**: SDKs and clients for interacting with the models. - **Applications**: Applications and games that use these models. - **Demos & Games**: Demos and games showcasing the models. - **Agent Tools**: Tools for agents or decision-making processes. - **Benchmarks & Evaluations**: Benchmarks and evaluations of the models. - **Papers**: Research papers related to the models. - **Articles**: Articles and blog posts discussing the models. - **Related**: Links to related projects and resources. - **Contribute**: Information on how to contribute to the project. ### Tech Stack: The repository is written in Python and hosted on GitHub. ### Why It Matters: This curated list provides a comprehensive overview of decision models, their applications, and the tools used to interact with them, making it useful for

### Overview: Craw4LLM is an efficient web crawling framework designed for large language model (LLM) pretraining. The repository contains the code for the paper "Craw4LLM: Efficient Web Crawling for LLM Pretraining" by [Author Name]. ### Key features: - **Efficient Crawling**: The crawler is optimized for large datasets, leveraging SSDs for faster data access. - **Scoring Methods**: Documents are scored using various models, including DCLM fastText, to determine their relevance. - **Configuration**: Users can customize the crawling process with different configurations, including seed documents, output directories, and selection methods. ### Tech stack: Python ### Why it matters: Craw4LLM provides a robust solution for pretraining large language models, making it essential for researchers and developers working on advanced natural language processing tasks.

### Overview: LongLM: Self-Extend LLM Context Window Without Tuning is a project that introduces a method for extending the context window of large language models (LLMs) without the need for retraining. This approach allows for more efficient and effective use of model memory and computation resources. ### Key Features: - **Self-Extend Implementation**: A method to extend the context window of LLMs without retraining, improving performance and memory usage. - **LLM Support**: Supports various LLM models, including LLaMa-3 and Qwen1.5. - **Flash Attention**: Implements a flash attention mechanism for self-extend, enhancing computational efficiency. - **Google Gemma Integration**: Provides a demonstration of the method on the Google I/O session. - **Legacy Support**: Maintains compatibility with older versions of transformers and flash attention. ### Tech Stack: Python ### Why It Matters: This project demonstrates the practical application of extending LLM context windows, showcasing the potential for more efficient model utilization and improved performance without the need for retraining.

Overview: Photo Agents is a Python package designed to create autonomous self-evolving agents that operate your computer. It leverages a perceive, reason, and act framework based on vision-grounded layered memory and self-written skills. Key features: - A streaming agent loop that drives any tool-calling LLM through a perceive → reason → act cycle. - A multi-provider LLM router with support for Anthropic Claude, OpenAI GPT, and a mixin failover session. - A physical-execution toolset including file I/O, sandboxed code execution, browser automation, and a layered memory system. - Pluggable clients for a polished Streamlit web app, PyQt desktop app, desktop companion, and ready-to-run bots for Telegram, QQ, Feishu, WeCom, and DingTalk. - Optional observability via Langfuse and a cron-style scheduler. Tech stack: Python 3.10+. Why it matters: Photo Agents provides a robust and autonomous solution for LLM-driven agents that operate your computer, ensuring you retain ownership of your data and keys. Repository: jmerelnyc/Photo-agents GitHub stars: 666 GitHub description: Autonomous self-evolving agents. Vision-grounded layered memory and self-written skills for LLM agents that operate your computer.

Overview: This project explores the use of large language models, such as ChatGPT and GPT-4, as re-ranking agents in Information Retrieval (IR). The research aims to evaluate the performance of these models in relevance ranking and contribute to the field of natural language processing. Key features: - Novel evaluation dataset, NovelEval, with novel search questions and passages. - Support for multiple LLMs, including Azure, Claude, Cohere, and Llama2 via LiteLLM. - Instruction distillation to simplify complex ranking instructions and achieve state-of-the-art performance with open-source LLMs. - Code for the paper "[Is ChatGPT Good at Search? Investigating Large Language Models as Re-Ranking Agent](https://arxiv.org/abs/2304.09542)". Tech stack: Python. Why it matters: This project provides valuable insights into the capabilities of generative LLMs in relevance ranking, contributing to the advancement of information retrieval techniques and natural language processing.

Overview: FoleyCrafter is an AI-based video-to-audio generation framework designed to add lifelike and synchronized sound effects to silent videos, enhancing the overall audiovisual experience. Key features: - Realistic sound effects - Synchronized with videos - Improves immersion and realism Tech stack: Python Why it matters: FoleyCrafter is a valuable tool for filmmakers, game developers, and content creators looking to elevate the audio quality of their silent videos, making them more immersive and engaging for the audience. Repository: open-mmlab/FoleyCrafter GitHub stars: 665

### Overview: Draw.io AI Kit is a tool designed to teach AI agents how to draw correct, beautiful diagrams using the Draw.io diagramming software. It provides a declarative catalog, design principles, and a structural validator to ensure the generated diagrams are structurally precise and aesthetically standardized, optimized for cloud environments like AWS, Azure, and GCP. ### Key Features: - **Declarative Catalog**: Maps Draw.io stencil IDs to their respective taxonomies and canonical color palettes. - **Design Principles**: Codified architectural and layout rules. - **Structural Validator**: Validates diagram XML to ensure stencil references are valid and design principles are satisfied prior to serialization. ### Tech Stack: TypeScript ### Why It Matters: Draw.io AI Kit helps mitigate common AI agent hallucinations such as generating non-existent stencil IDs, ensuring the diagrams are structurally precise and aesthetically standardized, making them suitable for cloud environments and other complex architectures.

### Overview: This repository contains the implementation of the paper "End-to-end Lane Detection through Differentiable Least-Squares Fitting" by Wouter Van Gansbeke et al., presented at the ICCV 2019 Workshop. The project focuses on end-to-end lane detection for self-driving cars, showcasing a novel approach that integrates weight maps for more accurate predictions. ### Key features: - **End-to-End Architecture**: Implements an end-to-end method for lane detection, combining a network for weight maps with a regression layer for final lane line coordinates. - **Differentiable Least-Squares Fitting**: Uses a regression approach to directly optimize the lane line coordinates, improving accuracy over traditional segmentation methods. - **Multi-Lane Detection**: Includes experiments on the TuSimple dataset, demonstrating the effectiveness of the method on complex scenarios. ### Tech stack: - Python - PyTorch 1.1 - OpenCV - Scikit-learn - torchvision - NumPy - Matplotlib - JSON/ujson - Pillow ### Why it matters: This project provides a significant advancement in lane detection for self-driving cars by offering a more accurate and efficient solution through end-to-end optimization. The ability to directly optimize for lane line coordinates and the inclusion of multi-lane detection experiments on the TuSimple dataset make it a valuable contribution to the field of autonomous driving.

### Overview: This repository is a curated list of pre-trained language models in scientific domains, covering various model sizes and modalities. It includes models for mathematics, physics, chemistry, materials science, biology, medicine, geoscience, and more. ### Key Features: - **SciBERT:** A pre-trained language model for scientific text (EMNLP 2019) - **BERT-Base:** 110M parameters - **BERT-Large:** 340M parameters - **Other Models:** Models with different sizes and modalities (e.g., language, graph, vision, table, molecule, protein, genome, climate time series) ### Tech Stack: Python ### Why It Matters: This repository is part of a survey paper that provides a comprehensive overview of scientific large language models and their applications in scientific discovery. It serves as a valuable resource for researchers and practitioners in the field, offering a curated list of models that can be used for scientific text analysis and discovery.

Overview: FinBERT is a pre-trained BERT model specifically designed for financial communication text, achieving state-of-the-art performance across various financial NLP tasks such as sentiment analysis, ESG classification, and forward-looking statement (FLS) classification. Key features: - FinBERT-Pretrained: A large-scale financial text pre-trained model. - FinBERT-Sentiment: Sentiment classification model. - FinBERT-ESG: ESG classification model. - FinBERT-FLS: Forward-looking statement (FLS) classification model. Tech stack: Jupyter Notebook Why it matters: FinBERT is a powerful tool for enhancing financial NLP research and practice, providing superior performance on various financial tasks, including sentiment analysis, ESG classification, and forward-looking statement classification.

### Overview: Heeler is a native iOS companion app for Herdr, an agent-first terminal runtime. It provides a dashboard of every coding agent running on your machines, sorted by who needs you. Open an Agent to read and steer its live terminal, draft with the full iOS keyboard, and send complete messages via SSH. ### Key features: - **Console**: A status-sorted list of every Agent on every machine, filterable by Host, and updated live. - **Attach**: The Agent's real terminal rendered by libghostty, featuring native scrollback and touch scrolling for full-screen TUIs. - **Chat**: Real-time chat with full messages sent over plain SSH. - **Skills**: Suggestions for Composer skills on the Agent's terminal. - **Live Activity**: Real-time tracking of Agents on the lock screen. ### Tech stack: Swift ### Why it matters: Heeler streamlines the management and interaction with Herdr agents, providing a seamless and efficient way to monitor and control your coding environments.

Overview: Dress Code is a high-resolution multi-category virtual try-on dataset introduced at the European Conference on Computer Vision (ECCV 2022). This dataset consists of over 50,000 high-resolution image pairs categorized into three categories: dresses, upper-body clothes, and lower-body clothes. Key features: - Over 50,000 high-resolution images - 3 categories: dresses, upper-body clothes, and lower-body clothes - 1024 x 768 image resolution - Includes keypoints, skeletons, human label maps, and dense poses Tech stack: Python Why it matters: Dress Code provides a comprehensive dataset for image-based virtual try-on, enabling researchers and developers to evaluate and improve virtual try-on techniques effectively.

### Overview: The **dart_openai** package is an unofficial SDK for the OpenAI API, providing a Dart/Flutter client for comprehensive access to OpenAI's services. It supports responses, chat completions, real-time sessions, videos, batch processing, fine-tuning, and more. The package is compatible with various providers such as Azure, DeepSeek, LM Studio, and Ollama, ensuring seamless integration with different environments. ### Key Features: - **Responses API**: Access to OpenAI's response endpoints. - **Chat Completions**: Real-time chat completions with tools and reasoning parameters. - **Conversations**: Facilitates conversations and interactions. - **Audio**: Supports speech, transcription, translation, and voice services. - **Images**: Generates, edits, and variates images, with streaming and partial results. - **Embeddings**: Provides vector embeddings for text data. - **Files**: Handles file uploads and multipart sessions. - **Batch**: Supports batch processing of requests. - **Vector Stores**: Manages vector stores and file batches. - **Models / Moderation**: Access to models and moderation features. - **Evals**: Provides evaluation tools. - **Graders**: Facilitates grading and validation. - **Fine-tuning jobs**: Supports fine-tuning jobs. - **Videos**: Generates, remixes, and downloads videos. - **Realtime sessions**: Enables real-time sessions and command line interfaces. ### Tech Stack: - **Language**: Dart - **Platforms**: Android, iOS, macOS, Linux, Windows, web, and server-side

### Overview: Dưới đây là một cuốn sách hướng dẫn sâu sắc về học sâu học, tập trung vào các công nghệ được sử dụng trên 300 trường đại học trên 55 quốc gia, bao gồm MIT, Stanford, Harvard, và Cambridge. ### Key features: - **Dive into Deep Learning**: Một cuốn sách hướng dẫn sâu sắc về học sâu học. - **GitHub Repository**: Cập nhật liên tục với 664 GitHub stars. - **Language**: Primary language is Python. - **Why it matters**: Đây là một tài liệu học sâu học chất lượng cao, được sử dụng rộng rãi trong các trường đại học hàng đầu trên thế giới. ### Tech stack: Python ### Why it matters: Dưới đây là một cuốn sách học sâu học chất lượng cao, được sử dụng rộng rãi trong các trường đại học hàng đầu trên thế giới.

### Overview: DFloat11 is a lossless compression framework designed to reduce the size of Large Language Models (LLMs) and diffusion models like FLUX.1, Qwen-Image, and HiDream-I1-Full by approximately 30%, while maintaining bit-for-bit identical outputs. This compression method enables efficient GPU inference on resource-constrained hardware, significantly reducing GPU memory usage without compromising accuracy. ### Key Features: - **Lossless Compression**: DFloat11 reduces the size of models without losing any accuracy. - **Efficient GPU Inference**: It enables GPU inference on resource-limited hardware, reducing peak GPU memory usage. - **Hardware-Aware Decompression**: The framework decompresses weights on-the-fly directly on the GPU, minimizing memory footprint. ### Tech Stack: - **Python**: The primary programming language used for development. - **PyTorch**: Used for implementing the compression and decompression algorithms. ### Why It Matters: DFloat11 is crucial for reducing the computational and memory overhead of running large language models on resource-constrained devices, such as mobile phones or low-end servers. By compressing models without sacrificing accuracy, it allows for more efficient and faster inference, making it ideal for real-time applications and edge computing environments.

Overview: cvpods is a versatile and efficient codebase designed for various computer vision tasks such as classification, segmentation, detection, and more. It aims to streamline experiments and facilitate smooth task switching. Key features: - Model Zoo: A collection of pre-trained models across multiple datasets and tasks. - Multi-node training: Support for distributed training on multiple GPUs or machines. - Inference and training capabilities: Easy to use and integrate into existing projects. Tech stack: Python, PyTorch, OpenCV. Why it matters: cvpods provides a comprehensive solution for computer vision tasks, making it easier to experiment and deploy models efficiently. Repository: Megvii-BaseDetection/cvpods GitHub stars: 663 GitHub description: All-in-one Toolbox for Computer Vision Research.

Overview: Datefinder is a Python module designed to extract dates from unstructured text, making it useful for parsing dates in various formats and mixed formatting documents. Key features: - Extract dates from text - Support for explicit calendar dates and relative expressions - Multiple date formats handling - Anchor relative parsing to a reference date - Compatibility with structured match objects Tech stack: Python Why it matters: Datefinder simplifies the process of extracting and parsing dates from text, making it invaluable for applications dealing with unstructured data like emails, support conversations, and legal documents.

Overview: This repository is a comprehensive collection of APIs designed for building autonomous AI agents, covering Agents, AI Models, and MCP Servers. It aims to streamline the development process by providing production-ready APIs, making it easier to create and deploy autonomous systems. Key features: - 2,207 production-ready APIs across Agents, AI Models, and MCP Servers - 582 APIs for execution layers, orchestration, and autonomous task handling - 1,266 APIs for AI model generation, reasoning, and transformation - 359 APIs for MCP server integrations and tool connections Tech stack: JavaScript Why it matters: This repository simplifies the process of building autonomous AI agents by providing a set of APIs that focus on the essential layers required for autonomous systems, eliminating the need to build infrastructure from scratch. It streamlines the development process, allowing developers to focus on the core functionality of their agents without the hassle of setting up complex infrastructure.

### Overview: This repository is a curated list of resources for Natural Language Processing (NLP) in Korean. It includes tools, datasets, blogs, papers, lectures, journals, conferences, institutes, and online communities related to Korean NLP. ### Key Features: - **Tools**: - **Morpheme/형태소 분석기 + Part of Speech(PoS)/품사 Tagger**: Hannanum (Java, C), Kkma (Java), Komoran (Java), Mecab-ko (C++), Twitter (Scala, Java), dparser (REST API), UTagger, Arirang (Lucence, Java), Rouzeta, seunjeon (Scala, Java), RHINO (라이노), KTS, 깜짝새 - **Named Entity(NE) Tagger**: annie - **Spell Checker**: PNU Spell Checker, Naver Spell Checker, Daum Spell Checker, hunspell-ko - **Syntax Parser**: dparser (REST API), NLP HUB - **Sentimental Analysis**: OpenHangul (오픈한글) - **Translator**: Naver NMT, OpenNMT, Google Translator - **Packages**: KoNLP (R), KoNLPy (Python), KoalaNLP (Scala), NLTK (Python), gensim (Python), FastText (C) ### Tech Stack: Python, Java, Scala, C++, R, C ### Why it Matters: This repository serves