Projects

**Overview:** DINO is an official implementation of the paper "DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection". This project is a significant contribution to the field of object detection, offering enhanced anchor boxes and improved performance on various datasets. **Key features:** - Improved DeNoising Anchor Boxes - Enhanced Performance on COCO Test-Dev - Supports DETR, Deformable DETR, Conditional DETR, DAB-DETR, DN-DETR, and DINO - Code and Checkpoints Available **Tech stack:** Python **Why it matters:** DINO is a crucial advancement in object detection, offering improved anchor boxes and better performance on COCO datasets, making it a valuable tool for researchers and developers working on end-to-end object detection models.

Overview: ELI5 is a Python library designed to help developers debug and explain the predictions of machine learning classifiers. It supports various frameworks and packages, including scikit-learn, Keras, xgboost, LightGBM, CatBoost, and lightning, and provides tools for text and image data. Key features: - Supports scikit-learn, Keras, xgboost, LightGBM, CatBoost, and lightning for explaining predictions and feature importances. - Provides text-based, HTML embeddable, pandas DataFrame, and JSON versions of explanations. - Implements algorithms for inspecting black-box models, including LIME for text classifiers and permutation importance for non-text classifiers. Tech stack: Python, Jupyter Notebook Why it matters: ELI5 simplifies the process of debugging and explaining machine learning models, making it easier for developers to understand and trust their models' predictions.

### Overview: ReadAny is an AI-powered cross-platform e-book reader designed for desktop and mobile devices. It offers semantic search, RAG chat, local vector store, notes management, text-to-speech functionality, and WebDAV sync, making it a versatile tool for enhancing reading experience. ### Key Features: - **AI Chat with Semantic Search**: Ask questions about the book and get answers grounded in the context of your reading. - **Local Vector Store**: Keeps your books, highlights, and notes offline-capable. - **RAG Chat**: Connects OpenAI, Claude, Gemini, Ollama, or custom-compatible providers. - **Text-to-Speech**: Converts text to speech for better accessibility. - **WebDAV Sync**: Syncs your reading across devices. ### Tech Stack: TypeScript ### Why It Matters: ReadAny provides a seamless and efficient way to enhance your reading experience by integrating AI and local storage, making it easier to manage and access your reading materials. ### GitHub Stars: 2849 ### Repository: codedogQBY/ReadAny

Overview: SetFit is an efficient few-shot learning framework that utilizes Sentence Transformers to achieve high accuracy with minimal labeled data. Key features: - No prompts or verbalizers: SetFit generates rich embeddings directly from text examples, eliminating the need for handcrafted prompts. - Fast training: SetFit is significantly faster to train and run inference compared to other few-shot learning methods. - Multilingual support: Works with any Sentence Transformer checkpoint available on the Hugging Face Hub. Tech stack: Python, Sentence Transformers, Hugging Face Hub, scikit-learn, PyTorch. Why it matters: SetFit provides a practical solution for few-shot fine-tuning, especially when dealing with limited labeled data, by leveraging the power of Sentence Transformers and being compatible with multilingual models.

Overview: Graph of Thoughts (GoT) is an official implementation of a framework designed to solve complex problems by modeling them as a Graph of Operations (GoO) and executing them with a Large Language Model (LLM). This framework aims to provide a flexible and extensible solution for various problem-solving scenarios. Key features: - Solves complex problems using a Graph of Operations (GoO) modeled with a Large Language Model (LLM). - Flexible and extensible, allowing for both new GoT approaches and modifications of previous methods like CoT or ToT. - Supports both CoT-like and GoT approaches for problem-solving. Tech stack: Python Why it matters: This framework offers a novel approach to problem-solving by leveraging the power of large language models to handle complex tasks, making it particularly useful in scenarios where traditional methods may be insufficient or impractical.

Overview: This repository provides comprehensive tutorials for TensorFlow, offering detailed explanations and source code in an easy-to-understand format. Key features: - Detailed explanation of TensorFlow concepts in .ipynb format - Source code in .py format for hands-on practice Tech stack: Python Why it matters: This project aims to simplify the learning process of TensorFlow, addressing common issues and providing a practical, easy-to-follow tutorial approach.

Overview: Gse is a Go implementation of the Jieba library, featuring support for multiple languages including English, Chinese, Japanese, and others, as well as advanced NLP features such as POS tagging, named entity recognition, and support for Elasticsearch and Bleve. Key features: - Support for common, search engine, full mode, precise mode, and HMM mode word segmentation. - User and embed dictionary support, POS tagging, segment info analysis, stop words, and word trimming. - Multilingual support for English, Chinese, Japanese, and other languages. - Traditional Chinese support. - HMM text segmentation using Viterbi algorithm. - NLP capabilities through TensorFlow (under development). - Support for Elasticsearch and Bleve. - JSON RPC service for easy integration. Tech stack: Go Why it matters: Gse offers efficient multilingual NLP and text segmentation capabilities, making it a valuable tool for developers working with text processing and natural language understanding in multiple languages. Repository: go-ego/gse Primary language: Go GitHub stars: 2843 GitHub description: Go efficient multilingual NLP and text segmentation; support English, Chinese, Japanese and others. README excerpt: Gse is a Go implementation of the Jieba library, featuring support for multiple languages including English, Chinese, Japanese, and others, as well as advanced NLP features such as POS tagging, named entity recognition, and support for Elasticsearch and Bleve. [](https

Overview: WhyLogs is an open-source data logging library designed for machine learning models and data pipelines. It provides visibility into data quality and model performance over time, ensuring safety and robustness through privacy-preserving data collection. Key features: - Real-time data logging - Model performance monitoring - Data quality analysis - Privacy-preserving data collection Tech stack: Jupyter Notebook Why it matters: WhyLogs empowers developers and data scientists to track and analyze data quality and model performance, ensuring the reliability and safety of their machine learning projects.

### Overview: This project is an end-to-end automatic speech recognition system implemented in TensorFlow, designed for both Madarian and English languages. It supports TensorFlow version 1.0 and beyond, including features like dropout for dynamic RNNs, shell script support, automatic evaluation after training epochs, and improvements in data preprocessing and training efficiency. ### Key Features: - **Support TensorFlow r1.0**: The project began with TensorFlow version 1.0 support. - **Support dropout for dynamic RNN**: Enhances the model's performance by adding dropout layers. - **Support running in shell file**: Enables easy execution of the model. - **Support evaluation every several training epochs automatically**: Facilitates continuous model improvement. - **Fix bugs for character-level automatic speech recognition**: Corrects issues in character-level recognition. - **Improve some function APIs for reusable**: Enhances the modularity of the code. - **Add scaling for data preprocessing**: Enables better data handling. - **Add reusable support for LibriSpeech training**: Simplifies the training process. - **Add simple n-gram model for random generation or statistical use**: Provides basic language models. - **Improve some code for pre-processing and training**: Enhances the overall training process. - **Replace TABs with blanks and add nist2wav converter script**: Simplifies the script for converting data. - **Add some data preparation code**: Helps in preparing the data for further processing. - **Add WSJ corpus standard preprocessing by s5 recipe**: Ensures consistent preprocessing for the WSJ corpus. - **Restructuring

### Overview: OpenPipe is an open-source fine-tuning and model-hosting platform designed to transform expensive prompts into cheaper, fine-tuned models. It allows developers to integrate powerful but expensive LLMs with smaller and more affordable models, enabling them to switch between OpenAI and fine-tuned models with a single line of code. ### Key Features: - **Easy Integration**: Seamless integration with OpenAI's Python SDK and TypeScript SDK. - **OpenAI-Compatible Chat Completions**: Supports chat completions using OpenAI's API. - **Fine-Tuning Capabilities**: Fine-tune models like GPT-3.5, Mistral, and Llama 2, hosted on the platform or downloaded weights. - **Query Logs**: Access powerful built-in filters for querying logs. - **Dataset Import**: Import datasets in OpenAI-compatible JSONL files. - **Duplicate Text Pruning**: Prune large chunks of duplicate text. - **Model Switching**: Easily switch between OpenAI and fine-tuned models with minimal changes. ### Tech Stack: TypeScript ### Why It Matters: OpenPipe simplifies the process of fine-tuning models, making it accessible to developers without requiring extensive knowledge of the underlying technology. By providing a platform for fine-tuning and hosting models, it democratizes access to powerful AI capabilities, making it easier for developers to experiment and create tailored solutions.

Overview: ZoeDepth is a project that combines relative and metric depth estimation, designed for metric depth estimation from a single image. It is based on the MiDaS repository and is now maintained by Intel. Key features: - ZoeD_N: Relative depth estimation model - ZoeD_K: Metric depth estimation model - ZoeD_NK: Combined relative and metric depth estimation model Tech stack: Jupyter Notebook Why it matters: ZoeDepth provides a practical solution for depth estimation in various applications, combining relative and metric depth to offer better accuracy and flexibility. Repository: isl-org/ZoeDepth GitHub stars: 2835 GitHub description: Metric depth estimation from a single image

Overview: This repository is a collection of resources spanning key areas of Computer Science, including algorithms, data structures, computer architecture, operating systems, computer graphics, programming languages, and more. It serves as a comprehensive guide for anyone looking to learn about computer science. Key features: - Algorithms & Data Structures - Computer Architecture - Operating Systems - Mathematics - Regular Expressions - Physics - Signals and Systems - Information Theory - Cloud Computing - Quantum Computing - Computer Networks - Computer Graphics - Virtual Reality Tech stack: Markdown Why it matters: This collection of resources is essential for anyone interested in computer science, providing a wealth of knowledge across various fields of the discipline.

Overview: AdapterHub's _Adapters_ library simplifies parameter-efficient and modular transfer learning by integrating 10+ adapter methods into 20+ advanced Transformer models, enabling efficient fine-tuning and modular transfer learning. Key features: - Integrates 10+ adapter methods into 20+ Transformer models - Supports full-precision and quantized training (e.g., Q-LoRA, Q-Bottleneck Adapters, Q-PrefixTuning) - Allows adapter merging via task arithmetics and composition of multiple adapters - Enables advanced research in parameter-efficient transfer learning for NLP tasks Tech stack: Python Why it matters: AdapterHub's _Adapters_ library simplifies the process of parameter-efficient and modular transfer learning by providing a unified interface for fine-tuning and inference, making it easier and more efficient for researchers and developers to experiment with advanced techniques in NLP tasks.

Overview: Scikit-Optimize, or "skopt", is a library designed for minimizing expensive and noisy black-box functions. It uses sequential model-based optimization methods and is built on top of NumPy, SciPy, and Scikit-Learn. The library aims to be accessible and easy to use in various contexts. Key features: - Sequential model-based optimization - Supports gradient-based optimization with "scipy.optimize" - Built on NumPy, SciPy, and Scikit-Learn - Accessible and easy to use in many contexts Tech stack: Python Why it matters: Scikit-Optimize is essential for anyone looking to minimize expensive and noisy black-box functions, making it particularly useful in fields like machine learning, physics simulations, and engineering design where computational resources are limited or expensive.

### Overview: DataChain is a Python library designed to transform unstructured data files into versioned, typed datasets, making them queryable at warehouse speed. It leverages S3, GCS, and Azure for storage and offers features like parallel computing, versioning, and efficient querying capabilities. ### Key Features: - **Compute Engine**: Utilizes parallel Python processing over files, distributed on Studio, for efficient computation. - **Dataset DB**: Pydantic schemas, versioning, and automatic lineage tracking for datasets. - **Sub-second Filter, Join, and Group_by**: Local and distributed operations on millions of records, with vector search capabilities. - **Optional Agent Skill**: Knowledge Base and Agent Harness for skillful workflows, including Claude Code, Cursor, Codex, GitHub Copilot, and Pi. ### Tech Stack: Python ### Why it Matters: DataChain simplifies the process of turning unstructured data into queryable datasets, optimizing storage and computation. It ensures that data remains in your storage, reducing recomputation, and provides a robust, scalable solution for developers and data scientists.

### Overview: This repository is designed to provide comprehensive instructions, code examples, and solutions for the *Hands-On* and *Exercise* portions of each chapter in the book "Introduction to Artificial Neural Networks and Deep Learning: A Practical Guide with Applications in Python". The repository includes PDF and ebook versions of the book from [Leanpub](https://leanpub.com/ann-and-deeplearning). ### Key Features: - **Hands-On Examples**: Detailed code examples and solutions for each chapter. - **Exercise Solutions**: Solutions to exercises and practical problems. - **Chapters Covered**: - 01 - Introduction - 02 - The Perceptron - 03 - Optimizing Cost Functions with Gradient Descent - 04 - Logistic Regression and Softmax Regression - 05 - From Softmax Regression to Multilayer Perceptrons - 06 - Cross Validation and Performance Metrics - 07 - Regularization in Neural Networks - 08 - Learning Rates and Weight Initialization - 09 - Convolutional Neural Networks - 10 - Recurrent Neural Networks - 11 - Autoencoders - 12 - General Adversarial Neural Networks - 13 - Deep Generative Models - 14 - Reinforcement Learning ### Tech Stack: Jupyter Notebook ### Why it Matters: This repository is essential for anyone looking to implement deep learning algorithms in Python, as it provides practical,

Overview: Knock Knock is a Python library designed to notify developers when their deep learning training is complete or crashes, providing automatic alerts for both scenarios. Key features: - Integrates with multiple platforms for notifications - Supports email, Slack, Telegram, Microsoft Teams, Text Message, Discord, Desktop notifications, Matrix, Amazon Chime, DingTalk, RocketChat, and WeChat Work - Minimal code modification required for seamless integration Tech stack: Python Why it matters: Knock Knock simplifies the process of setting up notifications for deep learning training, making it easier for developers to monitor their models' progress and ensure their projects run smoothly without manual intervention. Repository: huggingface/knockknock GitHub stars: 2825

Overview: LeptonAI is a Python library and command-line tool designed to simplify the operation of the NVIDIA DGX Cloud Lepton platform. Key features: - A `lep` command-line tool for creating and managing endpoints, Dynamo graph deployments, batch jobs, dev pods, Ray clusters, fine-tuning jobs, storage, secrets, and more. - A `Client` to call deployed endpoints as native Python functions. - Pythonic configuration specifications that are easily shipped to the cloud. - Skills that enable agents to operate the Lepton platform. Tech stack: Python Why it matters: LeptonAI simplifies the complex process of deploying and managing AI services on the NVIDIA DGX Cloud Lepton, making it accessible and easy to use for developers.

Overview: This project aims to use deep learning to crack captcha verification codes. It involves building a convolutional neural network using Keras to recognize captcha images. Key features: - Uses deep learning techniques for captcha recognition - Utilizes Keras for building and training the model - Supports both image and text captcha types - Includes a data generator for efficient data generation - Provides visualization code for better understanding and debugging Tech stack: TensorFlow, Keras Why it matters: This project is valuable for developers looking to automate captcha recognition, enhancing security and user experience in applications that use captcha verification. Repository: ypwhs/captcha_break Primary language: Jupyter Notebook GitHub stars: 2821 GitHub description: Crack captcha verification codes using deep learning.

### Overview: PromptBench is a unified evaluation framework designed for large language models (LLMs). It provides a comprehensive set of tools and methods to evaluate and understand the performance of various LLMs, including prompt engineering techniques and adversarial prompt evaluation. ### Key Features: - **Prompt Engineering:** Implementing efficient methods like Few-shot Chain-of-Thought, Emotion Prompt, and Expert Prompting. - **Adversarial Prompt Evaluation:** Integrating prompt attacks to assess the robustness of LLMs. - **Unified Evaluation:** Facilitating easy setup and evaluation of LLMs across different datasets and models. ### Tech Stack: - **Python:** The primary language used for development and evaluation. ### Why It Matters: PromptBench simplifies the evaluation process for researchers, making it easier to compare and understand the performance of different LLMs. Its unified approach ensures that evaluations are consistent and comparable across various datasets and models, facilitating better insights into model performance and improvement strategies. ### Repository: - **Repository:** microsoftarchive/promptbench - **GitHub Stars:** 2820 - **GitHub Description:** A unified evaluation framework for large language models. ### Installation: - **Dependencies:** Python 3.8+ and PyTorch 1.10+ are required. - **Setup:** Clone the repository and install the necessary dependencies. ### Usage: - **Quick Model Performance Assessment:** Use the provided APIs to quickly build, load datasets, and evaluate model performance. - **Prompt Engineering:** Implement prompt engineering methods to enhance model performance and understanding. ### Supported D