Projects

Overview: PentestGPT is an automated penetration testing agent powered by large language models, designed to streamline and enhance the process of ethical hacking. Key features: - Multi-stage pipeline for comprehensive testing - Autonomous agent that runs tools without human intervention - Session persistence for continuous testing sessions - Support for multiple categories including web, crypto, reversing, forensics, and privilege escalation - Live walkthrough and real-time feedback - Extensible architecture for future enhancements Tech stack: Python Why it matters: PentestGPT leverages advanced AI reasoning to perform penetration testing and CTFs, making it a valuable tool for security professionals looking to automate their testing processes. Repository: GreyDGL/PentestGPT Primary language: Python GitHub stars: 15,998 GitHub description: Automated Penetration Testing Agentic Framework Powered by Large Language Models GitHub link: https://github.com/GreyDGL/PentestGPT

### Overview: Nano-vLLM is a lightweight implementation of the vLLM model, designed for fast offline inference and clean codebase. ### Key Features: - **Fast Offline Inference**: Comparable inference speeds to vLLM. - **Readable Codebase**: Clean implementation in ~1,200 lines of Python code. - **Optimization Suite**: Prefix caching, Tensor Parallelism, Torch compilation, CUDA graph, etc. ### Tech Stack: Python ### Why it Matters: Nano-vLLM offers a fast and efficient alternative to vLLM, with a clean and optimized codebase, making it suitable for various applications requiring high inference speeds and performance optimization.

Overview: Outlines is a Python library designed to generate structured outputs directly from any language model, ensuring consistent and valid data across different models and APIs. Key features: - Works with any model - Simple integration - Guarantees valid structure - Provider independence Tech stack: Python Why it matters: Outlines simplifies the process of generating structured outputs, making it easier to ensure data quality and consistency across different LLMs and models. Repository: dottxt-ai/outlines Primary language: Python GitHub stars: 15951 GitHub description: Structured Outputs README excerpt: <div align="center" style="margin-bottom: 1em;"> <img src="./docs/assets/images/logo-light-mode.svg#gh-light-mode-only" alt="Outlines Logo" width=300></img> <img src="./docs/assets/images/logo-dark-mode.svg#gh-dark-mode-only" alt="Outlines Logo" width=300></img> 🗒️ *Structured outputs for LLMs* 🗒️ Made with ❤👷️ by the team at [.txt](https://dottxt.co) <br>Trusted by NVIDIA, Cohere, HuggingFace, vLLM, etc. <!-- Project Badges --> [![PyPI Version][pypi-version-badge]][pypi] [![Downloads][downloads-badge]][pypistats] [![Stars][stars-badge]][stars] <!-- Community Badges --> [![Discord][discord-badge]][discord] [![Blog][dottxt-blog-badge

Overview: ms-swift is a comprehensive framework for training, inference, evaluation, quantization, and deployment of large language models (LLMs) and multimodal large models. It supports 600+ text-only and 400+ multimodal models, including Qwen3, Qwen3.5, InternLM3, GLM4.5, Mistral, DeepSeek-R1, Llama4, and more. The framework accelerates training using Megatron parallelism techniques and reinforcement learning algorithms, and supports a wide range of tasks such as preference learning, embedding, reranking, and sequence classification. Key features: - Training (pre-training, fine-tuning, human alignment) - Inference - Evaluation - Quantization - Deployment Tech stack: Python Why it matters: ms-swift provides a robust and efficient solution for training, deploying, and evaluating large language models and multimodal models, making it a valuable tool for researchers and developers in the field of AI and natural language processing.

### Overview: This repository contains the lecture slides and course description for the "Deep Natural Language Processing" course offered in Hilary Term 2017 at the University of Oxford. The course focuses on recent advances in analysing and generating speech and text using recurrent neural networks, covering topics such as understanding sequential language modelling, conditional language models for transduction tasks, and advanced applications involving other mechanisms. ### Key Features: - Lecture slides and course description - Motivation for studying language processing using Deep Learning techniques - Introduction to basic machine learning concepts - Word embeddings as a practical and scalable solution for representing and learning word meanings ### Tech Stack: - Python - TensorFlow - PyTorch ### Why It Matters: This course is essential for anyone interested in the advancement of natural language processing, as it provides a comprehensive understanding of how deep learning techniques can be applied to improve the analysis and generation of language data. The practical implementation of these models on both CPU and GPU hardware is also discussed, making it a valuable resource for both theoretical and practical learners.

### Overview: Unstructured is an open-source ETL solution designed to transform complex documents into clean, structured formats, making them easily usable by language models. This project aims to simplify the process of converting unstructured data into structured data, enhancing the efficiency and accuracy of data processing for various applications. ### Key Features: - **Document Conversion**: Converts PDF, DOCX, and other document formats into structured data. - **Structured Output**: Provides clean, structured data that can be easily processed by language models. - **Partitioning and Enrichments**: Supports partitioning and enrichments to handle large datasets efficiently. - **Integration**: Integrates seamlessly with existing data pipelines and workflows. ### Tech Stack: Unstructured is built with Python and utilizes the following libraries and tools: - **Unstructured**: The main library for document conversion and structured output. - **Pandas**: For data manipulation and analysis. - **Dask**: For parallel processing of large datasets. - **PyTorch**: For machine learning models that process structured data. ### Why It Matters: Unstructured is crucial for anyone looking to integrate unstructured data into their language model workflows. By converting documents into structured data, it significantly enhances the accuracy and efficiency of data processing, making it easier to extract meaningful insights and improve the performance of language models.

Overview: Kaldi is the official location of the Kaldi project, a comprehensive speech recognition toolkit. It provides a robust and flexible framework for building and optimizing speech recognition systems. Key features: - Comprehensive speech recognition toolkit - Robust and flexible framework - Official location of the Kaldi project Tech stack: C++ and Shell Why it matters: Kaldi is a critical tool for researchers and developers working on speech recognition systems, offering a powerful and reliable solution for building high-quality speech recognition applications.

### Overview: ChatGLM2-6B 是一个开源的双语对话语言模型，基于 ChatGLM 初代模型开发，性能大幅提升，支持中英双语。它在多个数据集上表现出色，包括 MMLU、CEval、GSM8K 和 BBH 等，与初代模型相比有显著提升。此外，ChatGLM2-6B 的上下文长度扩展到 32K，推理速度和显存占用效率更高，支持多种新功能如 System Prompt、Function Call、Retrieval 和 Web_Search。 ### Key features: - **性能提升**：基于 GLM 混合目标函数，性能大幅提升。 - **上下文长度扩展**：将基座模型的上下文长度从 2K 扩展到 32K。 - **推理速度提升**：在官方模型实现下，推理速度提升 42%，INT4 量化下，6G 显存支持的对话长度由 1K 提升到 8K。 - **新功能支持**：支持 System Prompt、Function Call、Retrieval 和 Web_Search 等新功能。 ### Tech stack: Python ### Why it matters: ChatGLM2-6B 是一个强大的开源双语对话语言模型，性能显著提升，支持多种新功能，为开发者和研究者提供了一个高效、准确的对话模型。

Overview: This repository contains essential cheat sheets for deep learning and machine learning engineers, covering topics such as TensorFlow, Keras, Neural Networks Zoo, Numpy, Scipy, Pandas, Matplotlib, Seaborn, PySpark, and more. Key features: - TensorFlow - Keras - Neural Networks Zoo - Numpy - Scipy - Pandas - Matplotlib - Seaborn - PySpark - Dask Tech stack: Python, Jupyter Notebook, and Markdown. Why it matters: These cheat sheets provide a quick reference for common tasks and concepts in deep learning and machine learning, making it easier for researchers and engineers to get up to speed quickly.

Overview: Vosk is an offline speech recognition toolkit that supports 20+ languages, enabling continuous large vocabulary transcription and zero-latency responses with streaming API. Key features: - Supports 20+ languages including English, Indian English, German, French, Spanish, Portuguese, Chinese, Russian, Turkish, Vietnamese, Italian, Dutch, Catalan, Arabic, Greek, Farsi, Filipino, Ukrainian, Kazakh, Swedish, Japanese, Esperanto, Hindi, Czech, Polish, and more. - Small models (50 Mb) with continuous large vocabulary transcription. - Zero-latency response with streaming API. - Reconfigurable vocabulary and speaker identification. - Speech recognition bindings for Python, Java, Node.js, C#, C++, Rust, Go, and others. - Scalable from small devices to big clusters. Tech stack: Python, Java, Node.js, C#, C++, Rust, Go, Python, Java, Node.js, C#, C++, Rust, Go, Python, Java, Node.js, C#, C++, Rust, Go, Python, Java, Node.js, C#, C++, Rust, Go, Python, Java, Node.js, C#, C++, Rust, Go, Python, Java, Node.js, C#, C++, Rust, Go, Python, Java, Node.js, C#, C++, Rust, Go, Python, Java, Node.js, C#, C++, Rust, Go, Python, Java, Node.js, C#, C++, Rust, Go, Python, Java, Node.js, C#, C++, Rust, Go, Python

Overview: Hunyuan3D-2 is a high-resolution 3D assets generation system that leverages large-scale diffusion models to create immersive and realistic virtual worlds. Key features: - Open-source and simulation-capable 3D world generation model - Fully open-sourced with PBR model, VAE encoder, and training code - Turbo paint model and multiview texture generation pipeline - Texture enhancement module for high-definition textures - Blender addon for easy integration into 3D workflows Tech stack: Python Why it matters: Hunyuan3D-2 provides a powerful tool for generating high-resolution 3D assets, enabling creators to bring their imagination to life in immersive virtual environments.

### Overview Easy Dataset is a powerful tool designed for creating datasets specifically for Large Language Models (LLM), Retrieval-Augmented Generation (RAG), and model performance evaluation. It includes an intuitive interface and built-in document parsing tools, intelligent segmentation algorithms, and data cleaning capabilities. Users can easily convert domain-specific documents into high-quality structured datasets suitable for fine-tuning, RAG, and performance evaluation. ### Key Features - **Document Processing & Data Generation** - **Intelligent Document Processing**: Supports multiple formats like PDF, Markdown, DOCX, TXT, EPUB, and more. - **Intelligent Text Splitting**: Multiple algorithms (Markdown structure, recursive separators, fixed length, code-aware chunking) with customizable visual segmentation. - **Intelligent Question Generation**: Auto-extract relevant questions from text segments, with question templates and batch generation. - **Domain Label Tree**: Builds global domain label trees based on document structure, with auto-tagging capabilities. - **Answer Generation**: Uses LLM API to generate comprehensive answers and Chain of Thought (COT), with AI optimization. - **Data Cleaning**: Cleans text to remove noise and improve data quality. - **Multiple Dataset Types** - **Single-Turn QA Datasets**: Standard question-answer pairs for basic fine-tuning. - **Multi-Turn Dialogue Datasets**: Customizable roles and scenarios for conversational format. - **Image QA Datasets**: Generates visual QA data from images, with multiple import methods (directory, PDF, ZIP). - **Data Distillation**:

Overview: Virgilio is an open-source initiative designed to mentor and guide individuals in the world of Data Science. It aims to provide a clear path and resources for beginners to get started and learn new skills. Key features: - Free mentorship and guidance - Comprehensive resources and tutorials - Easy-to-use Jupyter Notebook interface - Clear and concise explanations Tech stack: Jupyter Notebook Why it matters: Virgilio simplifies the learning process by providing structured guidance and curated resources, making Data Science accessible to anyone regardless of their background. Repository: virgili0/Virgilio GitHub stars: 15,000 GitHub description: Your new Mentor for Data Science E-Learning.

Overview: Botpress is an open-source platform designed to build and deploy chatbots and assistants powered by GPT/LLM models, including those based on OpenAI. Key features: - Integrations: Public integrations on the Botpress Hub maintained by Botpress. - Devtools: All Botpress Cloud dev tools (CLI, SDK, API Client). - Bots: Example bots built using the Botpress SDK and CLI. - Plugins: Coming soon, all plugins on the Botpress Studio. Tech stack: TypeScript. Why it matters: Botpress simplifies the creation and deployment of chatbots and assistants, making it easier for developers to integrate cutting-edge AI models into their projects. GitHub stars: 14,956. Repository: botpress/botpress

### Overview: `nlp-tutorial` is a comprehensive tutorial series for deep learning researchers studying Natural Language Processing (NLP) using PyTorch. The repository contains detailed, step-by-step implementations of various NLP models, including Neural Network Language Models (NNLM), Word2Vec, FastText, TextCNN, TextRNN, TextLSTM, Bi-LSTM, and Seq2Seq with Attention. These models are implemented with less than 100 lines of code, making them accessible and easy to understand. ### Key Features: - **NNLM (Neural Network Language Model)**: Predicts the next word in a sequence. - **Word2Vec (Skip-gram)**: Embeds words and visualizes the relationship between words and phrases. - **TextCNN (Convolutional Neural Network)**: Classifies binary sentiment. - **TextRNN (Recurrent Neural Network)**: Predicts the next step in a sequence. - **TextLSTM (Long Short-Term Memory)**: Provides autocompletion capabilities. - **Bi-LSTM (Bidirectional LSTM)**: Predicts the next word in a long sentence. - **Seq2Seq (Sequence-to-Sequence)**: Changes words and translates between languages. - **Seq2Seq with Attention**: Translates text by aligning and translating sequences. ### Tech Stack: - **Programming Languages**: Python (Jupyter Notebook) ### Why It Matters: - The tutorial series is designed for deep learning researchers and provides a practical, step-by-step approach to implementing NLP models using PyTorch

Overview: This repository offers state-of-the-art deep learning examples that are easy to train and deploy, achieving high accuracy and performance on enterprise-grade infrastructure using NVIDIA CUDA-X software stack. Key features: - NVIDIA GPU Cloud (NGC) Container Registry - Latest NVIDIA examples, contributions, and software libraries - Monthly release notes for NVIDIA optimized containers Tech stack: NVIDIA CUDA-X software stack Why it matters: These examples provide a comprehensive set of deep learning scripts, making it easy to train and deploy models with high reproducibility and performance on NVIDIA GPUs.

### Overview: This project provides a Python implementation of reinforcement learning concepts as explained in Sutton & Barto's book "Reinforcement Learning: An Introduction (2nd Edition)". The code is based on the book and includes detailed explanations and examples for various reinforcement learning algorithms. ### Key Features: - Tic-Tac-Toe - Bandit problems - Epsilon-greedy action-value methods - Optimistic initial action-value estimates - UCB action selection - Gradient bandit algorithm - Gridworld examples - Blackjack problem - Monte Carlo ES - Weighted importance sampling - Ordinary importance sampling - Random walk - Batch updating - Sarsa applied to the windy grid world - Cliff problem ### Tech Stack: Python ### Why it Matters: This project serves as a practical guide for understanding and implementing reinforcement learning algorithms. It provides a clear and detailed explanation of the concepts, making it accessible to both beginners and advanced learners. The inclusion of examples and detailed code makes it an excellent resource for anyone interested in reinforcement learning.

Overview: OmniVoice is a high-quality voice cloning TTS model that supports over 600 languages, making it the most comprehensive zero-shot TTS model available. It features voice cloning, voice design, fine-grained control, and fast inference, all within a state-of-the-art architecture. Key features: - 600+ Languages Supported - Voice Cloning - Voice Design - Fine-grained Control - Fast Inference (RTF 0.025) Tech stack: Python Why it matters: OmniVoice stands out due to its broad language support, advanced voice cloning and design capabilities, and efficient inference, making it a valuable tool for developers and creators working with multilingual voice synthesis. Repository: k2-fsa/OmniVoice GitHub stars: 14,720 GitHub description: High-Quality Voice Cloning TTS for 600+ Languages README excerpt: # OmniVoice 🌍 <p align="center"> <img width="200" height="200" alt="OmniVoice" src="https://zhu-han.github.io/omnivoice/pics/omnivoice.jpg" /> </p> <p align="center"> <a href="https://huggingface.co/k2-fsa/OmniVoice"><img src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Model-FFD21E" alt="

Overview: NLTK is a comprehensive suite of Python modules, data sets, and tutorials for Natural Language Processing (NLP), designed to support research and development in the field. Key features: - Comprehensive suite of Python modules - Extensive data sets and tutorials - Supports research and development in NLP Tech stack: Python Why it matters: NLTK is essential for anyone working in NLP, providing a powerful toolkit for text analysis, sentiment analysis, named entity recognition, and more. Repository: nltk/nltk Primary language: Python GitHub stars: 14742 GitHub description: NLTK Source README excerpt: # Natural Language Toolkit (NLTK) [](https://pypi.python.org/pypi/nltk)  NLTK -- the Natural Language Toolkit -- is a suite of open source Python modules, data sets, and tutorials supporting research and development in Natural Language Processing. NLTK requires Python version from 3.10 up to the latest 3.14. For documentation, please visit [nltk.org](https://www.nltk.org/).

Overview: waooAI/waoowaoo is the first industrial-grade全流程 AI 影视生产平台，提供从短片到真人动作的全流程创作流程，采用好莱坞标准的工作流。 Key features: - Work with the right-hand Assistant to develop a brief and create assets in an ongoing project. - Upload reference material, organize files, and inspect generated images and videos on the canvas. - Generate with model-aware aspect ratios, durations, resolutions, and reference roles. - Use supported first-frame, first/last-frame, or reference-image modes; availability depends on the selected model. - Refine an existing result into a new version while retaining the original. Tech stack: TypeScript Why it matters: waooAI/waoowaoo is the industry's first AI creative workspace for images and videos, allowing developers to create and refine their work in real-time. Repository: waooAI/waoowaoo Primary language: TypeScript GitHub stars: 14,574 GitHub description: 首家工业级全流程 AI 影视生产平台。Industry-first professional AI Agent platform for controllable film & video production. From shorts to live-action with Hollywood-standard workflows. README excerpt: [English](README.md) · [简体中文](README_zh.md) · [日本語](README_ja.md) · [한국어](README_ko.md) > [!IMPORTANT] > **Preview release:** we are iterating rapidly, and some bugs and rough edges remain. Join the community to share feedback and follow updates.