Projects

### Overview: TensorWatch is a powerful debugging and visualization tool designed for machine learning and data science projects, providing real-time visualizations and interactive querying capabilities. ### Key Features: - Real-time visualizations of machine learning training processes - Flexible and extensible architecture for custom visualizations and dashboards - Ability to execute arbitrary queries against live machine learning training processes - Lazy Logging feature for interactive debugging with Python expressions - Support for Python 3.x and PyTorch 0.4-1.x, with TensorFlow eager tensors support - Built-in support for network diagrams using Graphviz ### Tech Stack: Jupyter Notebook ### Why it Matters: TensorWatch simplifies the debugging and visualization of machine learning models, making it easier to identify and fix issues in real-time. Its extensibility allows for the creation of custom visualizations and dashboards, enhancing the overall data science workflow.

Overview: TextAttack is a Python framework designed for adversarial attacks, data augmentation, and model training in Natural Language Processing (NLP). It provides a comprehensive toolkit for researchers and developers to enhance their NLP models. Key features: - Adversarial attacks on NLP models - Data augmentation techniques - Model training with a single command - Pre-trained models and dataset samples - Customizable transformations and constraints Tech stack: Python Why it matters: TextAttack simplifies the process of researching and developing new adversarial attacks, data augmentation methods, and model training techniques in NLP, making it an essential tool for researchers and developers looking to improve the robustness and performance of their NLP models.

### Overview: This project allows an AI model to interact with a real iPhone using a $7 ESP32 board, eliminating the need for a computer or jailbreak. The agent runs in an iPhone app, using the board's Wi-Fi hotspot and USB HID capabilities to control the mouse and keyboard. ### Key Features: - **Screen Mirroring (Wi-Fi)**: The board appears as "ESP32 Agent" in the phone's Screen Mirroring list, forwarding H.264 stream to the app over BLE. - **USB HID Mouse and Keyboard**: The board functions as a standard USB mouse and keyboard, allowing precise control via the app. - **Vision Model Interaction**: The app decodes the stream, sends screenshots to any OpenAI-compatible vision model, and executes actions like tap, swipe, draw, type, and press keys. - **Dynamic Island**: The app displays a Dynamic Island that shows what it is doing, providing visual feedback. - **Pointer Calibration**: The app tracks the pointer's position and corrects for any inaccuracies, ensuring precise actions. - **Replay and Export**: The app keeps a record of every run, allowing users to replay it step by step, export screenshots, actions, and annotations for analysis. ### Tech Stack: - **Hardware**: ESP32-S3 board - **Software**: C, Python ### Why It Matters: This project simplifies the process of integrating AI with real-world devices, making it accessible without requiring advanced technical skills or expensive hardware. It opens up new possibilities for remote work and collaboration, especially in fields

### Overview: This repository serves as a comprehensive guide for individuals preparing for Data Science and Machine Learning interviews. It includes a wide range of resources and techniques to help candidates succeed in interviews and apply their skills effectively. ### Key Features: - **Interview Tips and Strategies**: Guidance on how to prepare and perform well in interviews. - **Essential Skills**: A detailed list of skills required for Data Science roles, including SQL, programming, mathematics, and machine learning. - **Domain Knowledge**: Resources on domain-specific knowledge and understanding of various industries. - **Technical Resources**: Links to various tools, libraries, and frameworks commonly used in Data Science. ### Tech Stack: - Python, SQL, and other programming languages. - Machine Learning and Data Science libraries such as TensorFlow, PyTorch, and Scikit-learn. ### Why It Matters: - Provides a structured approach to preparing for Data Science interviews. - Helps bridge the gap between theoretical knowledge and practical application. - Includes a mix of theoretical and practical resources to ensure a well-rounded preparation. ### Repository: rbhatia46/Data-Science-Interview-Resources - GitHub Stars: 3470 - Description: A repository listing out the potential sources which will help you in preparing for a Data Science/Machine Learning interview. Resources added frequently. - Primary Language: Unknown - GitHub Description: "A repository listing out the potential sources which will help you in preparing for a Data Science/Machine Learning interview. New resources added frequently." ### How to Get an Interview: - Develop necessary skills and be well-versed with fundamentals

### Overview: DeepCTR-Torch is a PyTorch-based package designed for easy-to-use, modular, and extendable deep-learning-based CTR (Click-through Rate) models. It includes a variety of core components that can be easily integrated into custom models. ### Key Features: - **Easy-to-use**: The package is designed to be straightforward and user-friendly. - **Modular**: Components can be easily added or removed, making it highly customizable. - **Extendible**: The architecture allows for easy integration with other models or custom models. ### Tech Stack: - **Python**: The primary programming language used. ### Why It Matters: DeepCTR-Torch simplifies the process of building and deploying CTR models using advanced deep learning techniques. It provides a robust framework that can handle complex interactions between features, making it suitable for various applications requiring accurate click-through rate predictions. ### Repository: - **Repository**: shenweichen/DeepCTR-Torch - **Stars**: 3468 - **Description**: A PyTorch-based package for deep-learning based CTR models. ### Documentation: - **Status**: The documentation is available on ReadTheDocs. - **CI Status**: Continuous Integration status badge. - **Codecov**: Code coverage badge. - **License**: MIT License. ### Installation: - **Installation**: Use `pip install -U deepctr-torch` to install the package. ### Models List: - **Convolutional Click Prediction Model**: A model for click prediction. - **Factorization-supported Neural Network**: Models for user response prediction. -

Overview: MiniMax-01 and MiniMax-VL-01 are large language models and vision-language models based on Linear Attention, developed by MiniMax AI. These models are designed to be efficient and scalable, making them ideal for various applications. Key features: - Large language models and vision-language models based on Linear Attention - Efficient and scalable models - Developed by MiniMax AI Tech stack: Python Why it matters: These models are designed to be efficient and scalable, making them ideal for various applications. They are based on Linear Attention, which allows for faster processing and better performance compared to traditional models. Repository: MiniMax-AI/MiniMax-01 GitHub stars: 3469 GitHub description: The official repo of MiniMax-Text-01 and MiniMax-VL-01, large-language-model & vision-language-model based on Linear Attention.

### Overview: MTEB is a state-of-the-art evaluation framework designed to assess embeddings across various languages and modalities. It serves as a comprehensive tool for developers and researchers to benchmark and compare different embedding models. ### Key Features: - **Benchmarking**: Evaluates embeddings across multiple languages and modalities. - **Model Compatibility**: Supports a wide range of models, including Sentence Transformers, enabling easy integration. - **Task Flexibility**: Offers a variety of tasks such as classification, clustering, and more. - **Ease of Use**: Provides a simple installation process and straightforward CLI commands for running evaluations. ### Tech Stack: Python ### Why It Matters: MTEB provides a standardized way to evaluate embeddings, making it easier for researchers to compare and validate the performance of different models across different domains and languages. This framework is essential for advancing the field of natural language processing and machine learning. ### GitHub Stars: 3466 ### Repository: - Repository: embeddings-benchmark/mteb - Primary Language: Python - GitHub Stars: 3466

Overview: AI-Optimizer is a comprehensive deep reinforcement learning toolkit designed for researchers and developers. It offers a wide range of algorithms and libraries, including model-free, model-based, single-agent, and multi-agent RL methods, as well as a flexible distributed training framework. Key features: - Model-free RL algorithms - Model-based RL algorithms - Single-agent RL algorithms - Multi-agent RL algorithms - Offline RL algorithms - Transfer and multi-task RL algorithms Tech stack: Python Why it matters: AI-Optimizer is crucial for researchers and developers looking to implement and experiment with advanced reinforcement learning techniques in various applications, including multi-agent systems, self-supervised learning, and offline reinforcement learning. Repository: TJU-DRL-LAB/AI-Optimizer Primary language: Python GitHub stars: 3466 GitHub description: The next generation deep reinforcement learning tookit

Overview: Sail is a drop-in Apache Spark replacement written in Rust, offering unified batch processing, stream processing, and compute-intensive AI workloads on a distributed, multimodal compute engine. Key features: - **Compatible with the Spark Connect protocol**, supporting the Spark SQL and DataFrame API with no code rewrites required. - **100% Rust-native**, with no JVM overhead, delivering memory safety, instant startup, and predictable performance. - **~10× faster** than Spark and **98% cheaper** on infrastructure costs. See derived TPC-H benchmarks. - **Proven on ClickBench**, outperforming Spark, popular Spark accelerators, Databricks, and Snowflake. Tech stack: Rust Why it matters: Sail simplifies Apache Spark integration, reduces infrastructure costs, and offers superior performance and cost-effectiveness compared to Spark and other alternatives.

### Overview: Automatic colorization using deep neural networks, developed by Richard Zhang, Phillip Isola, and Alexei A. Efros at the University of California, Berkeley. This project was presented at ECCV 2016 and later featured in SIGGRAPH 2017. The code supports both real-time user-guided image colorization and automatic colorization. ### Key features: - Real-time user-guided image colorization - Automatic colorization for improved efficiency - Supports both Caffe and PyTorch architectures ### Tech stack: - Python ### Why it matters: This project is significant for its ability to automatically colorize images, making it easier for users to create visually appealing images without manual color adjustments. It has been widely adopted and is referenced in various research papers and academic conferences. ### Repository: - GitHub: richzhang/colorization - GitHub stars: 3464 - Primary language: Python - Caffe branch: https://github.com/richzhang/colorization/tree/caffe - Citation: - @inproceedings{zhang2016colorful, title={Colorful Image Colorization}, author={Zhang, Richard and Isola, Phillip and Efros, Alexei A}, booktitle={ECCV}, year={2016}, } - @article{zhang2017real, title={Real-Time User-Guided Image Colorization with Learned Deep Priors}, author={Zhang, Richard and Zhu, Jun-Yan and Isola,

Overview: MetaClaw is a conversational agent that learns and evolves from every interaction, inspired by brain learning mechanisms. It can be used without requiring a GPU and supports one-click deployment and skills-based training. Key features: - Skills Mode: Enables skills-based training without requiring RL. - RL Mode: Trains the agent using reinforcement learning. - Auto Mode: A default mode that combines skills and scheduled RL training. - Memory: Stores user and project history for context-aware prompts. - Configuration: Allows customization of the agent's behavior through settings. - Skills Mode: Enables skills-based training without requiring RL. - RL Mode: Trains the agent using reinforcement learning. - Auto Mode: A default mode that combines skills and scheduled RL training. - Memory: Stores user and project history for context-aware prompts. - Configuration: Allows customization of the agent's behavior through settings. - Skills Mode: Enables skills-based training without requiring RL. - RL Mode: Trains the agent using reinforcement learning. - Auto Mode: A default mode that combines skills and scheduled RL training. - Memory: Stores user and project history for context-aware prompts. - Configuration: Allows customization of the agent's behavior through settings. - Skills Mode: Enables skills-based training without requiring RL. - RL Mode: Trains the agent using reinforcement learning. - Auto Mode: A default mode that combines skills and scheduled RL training. - Memory: Stores user and project history for context-aware prompts. - Configuration: Allows customization of the agent's behavior through settings. - Skills Mode: Enables skills-based training without requiring RL. - RL Mode

### Overview: Claude Code is a comprehensive guide to Claude Code, covering everything from the basics to advanced features. It includes hands-on lessons and terminal exercises, with a focus on learning through practice. ### Key features: - Beginner lessons with step-by-step checks - Intermediate lessons with capstones - Advanced lessons that work from any page - Comprehensive check exercises - Source-cited facts and explanations ### Tech stack: JavaScript ### Why it matters: Claude Code provides a structured, hands-on approach to learning Claude Code, making it accessible and effective for beginners to advanced users. The guide is designed to be self-contained and independent, ensuring that users can learn Claude Code without relying on external resources.

### Overview: Lynote Humanize Text is an open-source pipeline designed to rewrite AI-generated text into natural human prose. This project leverages advanced data pipelines and preference training to achieve superior human-like output, making it ideal for applications requiring high-quality humanization. ### Key Features: - **Pipeline Architecture:** A structure-first data pipeline that trains the model to rewrite structure rather than just surface style. - **Training Data:** Full implementation details in `training-data-pipeline/`, including policies, tests, and 50 demo records per stage. - **API Integration:** Available as a production-ready humanizer API, `lynote.ai/ai-humanizer-api`, for seamless integration into your applications. - **Community Support:** Includes a Discord server and a product hunt page, making it easy to get involved and share your experiences. ### Tech Stack: - Python 3.10+ - Public Crawl + private pairs - Quality + structure gates - Deterministic stratified split - SFT training pool - Reserve a disjoint KTO pool - 6 candidate chains - Negative selection + structure-first 60/20/20 mix - Human review gate - Balanced KTO train / validation - Controlled prompt diversity ### Why It Matters: This project is a significant advancement in the field of AI text rewriting, offering a transparent and effective solution for improving the quality of AI-generated content. By focusing on structure and preference, it ensures that the rewritten text is not only readable but also maintains the intended meaning and structure of the original text. ###

### Overview: Clean, Robust, and Unified PyTorch implementation of popular Deep Reinforcement Learning (DRL) algorithms such as Q-learning, Duel DDQN, PER, C51, Noisy DQN, PPO, DDPG, TD3, SAC, ASL. ### Key Features: - **Q-learning**: A fundamental algorithm for value-based reinforcement learning. - **Duel Double DQN**: Enhances the Double DQN method for better performance. - **Noisy Duel DDQN on Atari Game**: Improves learning by adding noise to the model. - **Prioritized Experience Replay (PER) DQN/DDQN**: Prioritizes experiences to speed up learning. - **Categorical DQN (C51)**: A variant of DQN that uses categorical sampling. - **NoisyNet DQN**: Adds noise to the neural network for better generalization. - **Proximal Policy Optimization (PPO) for Discrete Action Space**: Optimizes the policy using PPO for discrete actions. - **Proximal Policy Optimization (PPO) for Continuous Action Space**: Optimizes the policy using PPO for continuous actions. - **Deep Deterministic Policy Gradient (DDPG)**: A method for continuous action spaces. - **Twin Delayed Deep Deterministic Policy Gradient (TD3)**: Improves the stability of DDPG. - **Soft Actor Critic (SAC) for Discrete Action Space**: A method for discrete action spaces. - **Soft Actor Critic (SAC) for Continuous Action Space**: A

Overview: AdaNet is a fast and flexible AutoML framework designed for learning high-quality models with minimal expert intervention. It builds on recent AutoML efforts to offer learning guarantees and provides a general framework for both learning a neural network architecture and ensemble learning. Key features: - Adaptive neural architecture search and ensemble learning in a single train call - Support for regression, binary and multi-class classification, and multi-head tasks - TensorFlow Estimator API for training, evaluation, prediction, and serving models - Ability to define subnetworks that change structure over time using `tf.layers` via the `adanet.subnetwork` API - CPU, GPU, and TPU support - Distributed multi-server training using `tf.estimator.Estimator` - TensorBoard integration Tech stack: Python, TensorFlow Why it matters: AdaNet is a lightweight framework for AutoML that offers fast and flexible learning guarantees, making it suitable for researchers and practitioners looking to automate the process of creating high-quality models with minimal human intervention.

### Overview: DecryptPrompt是一个专注于LLM论文总结和AIGC应用资源的项目，旨在为开发者提供最新的LLM资源和应用指南，帮助他们在AI研究和开发中取得成功。 ### Key features: - 提供LLM资源汇总，涵盖开源模型、框架、数据集和应用领域。 - 提供跟着博客读论文的系列教程，涵盖从基础到高级的LLM微调和应用。 - 包含推荐的技能工具合集，帮助开发者提升AI能力。 ### Tech stack: - Python - Markdown - Git ### Why it matters: DecryptPrompt是开发者获取最新AI研究和应用资源的宝贵工具，帮助他们在AI领域不断进步。

Overview: Towhee is a framework designed to simplify the processing of unstructured data using Large Language Models (LLMs). It transforms diverse types of unprocessed data into specific formats like text, images, or embeddings, making it easier to integrate into storage systems. Key features: - Multi-Modality Support: Handles various data types including images, videos, text, audio, and molecular structures. - Flexible LLM Pipeline Orchestration: Adapts to different LLMs and allows local hosting of open-source models. - Rich Operators: Offers over 140 models across five domains (CV, NLP, multimodal, audio, and medical) with functionalities like video decoding, audio slicing, frame sampling, and dimensionality reduction. - Prebuilt ETL Pipelines: Provides ready-to-use ETL pipelines for tasks like Retrieval-Augmented Generation, Text Image search, and Video copy detection. - High Performance Backend: Utilizes Triton Inference Server for efficient model serving on both CPU and GPU, and transforms Python pipelines into Docker containers for easy deployment and scaling. - Pythonic API: Includes a method-chaining API for describing custom data processing pipelines and supports schemas for unstructured data processing. Tech stack: Python Why it matters: Towhee simplifies the process of integrating unstructured data with LLMs, making it easier to extract valuable insights from diverse data types. Its prebuilt ETL pipelines and high-performance backend make it suitable for both prototyping and production environments, while its rich operators and Pythonic API make it accessible for developers of all skill levels.

### Overview: This repository is a curated list of language modeling researches specifically focused on code (and other software engineering activities). It includes surveys, models, benchmarks, and datasets related to language models for code and related software engineering tasks. The repository aims to provide a comprehensive overview of the latest advancements in this field. ### Key features: - Code LLM Surveys: A curated list of surveys on language modeling for code and software engineering. - Models: A collection of language models and pretraining strategies, including Base LLMs, existing LLMs adapted to code, general pretraining on code, instruction fine-tuning, and reinforcement learning on code. - Benchmarks: Comprehensive benchmarks for code search, text-to-code, code-to-text, and code-to-code tasks. - Datasets: Various datasets and resources for research and development in language modeling for code. ### Tech stack: Python, Jupyter Notebook, Markdown, Git, and GitHub. ### Why it matters: This repository is crucial for researchers, developers, and practitioners interested in advancing the field of language modeling for code. It provides a centralized source of information and resources, facilitating collaboration and knowledge sharing. By contributing to this repository, one can help advance the state-of-the-art in language modeling for code and contribute to the development of new applications and tools.

### Overview: Otter is a multi-modal model based on OpenFlamingo, trained on MIMIC-IT and showcasing improved instruction-following and in-context learning ability. It is designed to enhance the model's ability to follow instructions and understand context in response to prompts. ### Key Features: - Improved instruction-following - Enhanced in-context learning - Multi-modal capabilities ### Tech Stack: Python ### Why It Matters: Otter demonstrates significant advancements in model performance, particularly in instruction-following and context understanding, making it a valuable tool for developers looking to improve their models' ability to follow instructions and learn from context. ### Repository: EvolvingLMMs-Lab/Otter ### GitHub Stars: 3444 ### Project Credits: - [Otter Paper](https://arxiv.org/abs/2305.03726) - [OtterHD Paper](https://arxiv.org/abs/2311.04219) - [MIMIC-IT Paper](https://arxiv.org/abs/2306.05425) ### Checkpoints: - [luodian/OTTER-Image-MPT7B](https://huggingface.co/luodian/OTTER-Image-MPT7B) - [luodian/OTTER-Video-LLaMA7B-DenseCaption] ### Disclaimer: The code may not be perfectly polished and refactored, but all opensourced codes are tested and runnable. If you have any

Overview: This repository contains machine learning notebooks designed for educational and research purposes. The notebooks are minimal, easily reusable, and extendable. Key features: - Introduction to Computational Graphs - PyTorch Hello World! - A Gentle Introduction to PyTorch - Counterfactual Explanations - Linear Regression from Scratch - Logistic Regression from Scratch - Concise Logistic Regression Tech stack: Jupyter Notebook Why it matters: These notebooks provide a comprehensive introduction to various machine learning concepts and techniques, making them valuable resources for both learning and research. Repository: dair-ai/ML-Notebooks GitHub stars: 3440 GitHub description: :fire: Machine Learning Notebooks