Projects

Overview: This repository contains code for a tutorial on Transfer Learning in Natural Language Processing, presented at NAACL 2019. The tutorial focuses on modern transfer learning methods and their applications in NLP. Key features: - Pretraining model - Pretraining training script - Fine-tuning model - Fine-tuning training script - Pretrained models and datasets hosted Tech stack: Python Why it matters: This tutorial provides a practical introduction to transfer learning methods in NLP, showcasing how these techniques can improve model performance across various tasks, including text classification and language modeling.

### Overview: This project introduces MDLM, a Masked Discrete Diffusion Language Model that simplifies the diffusion process by parameterizing the absorbing state with a novel substitution-based approach. This results in state-of-the-art perplexity on LM1B and OpenWebText while maintaining competitive zero-shot perplexity with state-of-the-art AR models across various datasets. ### Key Features: - **Substitution-based Parameterization**: Simplifies the absorbing state diffusion loss to a mixture of classical masked language modeling losses. - **Simplified Loss Calculation**: Achieves SOTA perplexity on LM1B and OpenWebText while maintaining competitive zero-shot perplexity with SOTA AR models. - **Efficient Implementation**: Supports ancestral sampling, analytic sampling, and an efficient sampler that is **~3-4x** faster than existing diffusion models, enabling semi-autoregressive generation. ### Tech Stack: Python ### Why It Matters: This model offers a significant improvement in perplexity and efficiency compared to existing diffusion models, making it a valuable tool for language modeling and diffusion techniques.

Overview: Infinite Monitor is an AI-powered dashboard builder that allows developers to describe what they want and an AI agent writes, builds, and deploys it in real time. Each widget is a full React app running in an isolated iframe, providing a secure and customizable dashboard experience. Key features: - Real-time dashboard building - Drag, resize, and organize widgets on an infinite canvas - Supports various domains including cybersecurity, OSINT, trading, and prediction markets - Uses AI to write and rebuild React code in seconds Tech stack: TypeScript, React, Next.js, SQLite, Secure Exec's isolated-vm, and various AI providers Why it matters: Infinite Monitor simplifies the process of creating custom dashboards, enabling developers to focus on their core responsibilities while AI handles the technical aspects, ensuring a secure and customizable solution.

Overview: AI Memory Vault is a free, open-source system that turns an Obsidian vault into your AI's persistent memory, ensuring everything is remembered across sessions and outside the model with no size ceiling. It includes a boot config, a folder structure for projects, daily notes that write themselves, and a profile that updates as the AI learns about you. Key features: - Boot config for Claude Code sessions - Vault integration with Obsidian - Self-maintaining system - AI learning and remembering tasks - Templates for filling in notes Tech stack: Markdown and Obsidian Why it matters: This system provides a reliable, persistent memory for AI assistants, ensuring consistent and accurate results across sessions and tasks.

### Overview: This repository is a curated list of data science and AI guided projects designed to help developers build their portfolio. It includes projects ranging from building conversational chatbots with Langchain and large language models to fine-tuning LLMs for various applications. ### Key Features: - Building a Conversational Chatbot with Langchain and Large Language Models - Automating YouTube Script Writing with Langchain and Streamlit - Building a PDF Chat App using Langchain and OpenAI API - Building a Falcon 4 0BChat Web Application using HuggingFace & Gardio - Building a GPT Banker Using LLaMA 2 70B - Finetuning Llama 2 on Your Local Machine Using HuggingFace Autotrain - Building a Lex Fridman Podcast Summarization App with Whisper Jax, Azure OpenAI, and Langchain - Creating a Veterinary Chatbot using Llama 2: Harnessing Gen AI for Pet Care - Deploying Llama 2 on AWS SageMaker using DLC (Deep Learning Containers) - Article/Blog Generation App using Llama2, Langchain, and Pexels ### Tech Stack: Python, Langchain, Hugging Face, Streamlit, OpenAI API, LLaMA, Gardio, Whisper Jax, Azure OpenAI, Langchain, Pexels, AWS SageMaker, DLC (Deep Learning Containers) ### Why it Matters: These projects offer practical solutions for developers looking to enhance their skills in data science and AI. They provide a hands-on approach to

### Overview: Orpheus-FASTAPI is a high-performance Text-to-Speech server optimized for RTX GPUs. It leverages OpenAI's API for compatibility, supports 8 voices in multiple languages, and features emotion tags for enhanced realism. The server is designed for modern web interfaces and comes with comprehensive multilingual support, including 24 voices across 7 languages, and optimized models for various languages. ### Key Features: - **OpenAI API Compatible**: Seamless integration with OpenAI's speech API. - **8 Voices**: Support for 8 distinct voices in multiple languages. - **Emotion Tags**: Enhances the realism of the output with emotion tags. - **Modern Web UI**: Clean, user-friendly interface for easy interaction. - **GPU Optimization**: Optimized for RTX GPUs for better performance. ### Tech Stack: Python ### Why It Matters: Orpheus-FASTAPI is designed to provide a high-quality, efficient, and modern solution for text-to-speech applications, making it ideal for developers looking to integrate speech synthesis into their projects. Its compatibility with OpenAI's API and support for multiple languages make it a versatile tool for a wide range of applications, from gaming to educational software. ### Repository: Repository: Lex-au/Orpheus-FastAPI GitHub Stars: 721 GitHub Description: High-performance Text-to-Speech server with OpenAI-compatible API, multilingual support with 24 voices, emotion tags, and modern web UI. Optimized for RTX GPUs. ### Changelog: - **v1.3.

### Overview: PaddleMIX 是一个支持主流多模态任务的多模态集成和探索框架，包括大规模多模态预训练模型和扩散模型工具包。它具备高性能和灵活性，适用于各种应用场景。 ### Key features: - 多模态模型库 - 全流程开发体验 - 高性能分布式训练推断能力 - 特色模型与工具 ### Tech stack: Python ### Why it matters: PaddleMIX 是一个强大的工具，适用于多模态任务的开发和应用，提供了一站式的解决方案。

Overview: This repository contains a collection of CVPR papers from various years, including 2022, 2021, 2020, 2019, 2018, and 2017. The papers are available in PDF format, and the repository is updated regularly with the latest research findings. Key features: - CVPR 2022/2021/2020/2019/2018/2017 papers - Available in PDF format - Updated regularly with the latest research findings Tech stack: Unknown Why it matters: This repository serves as a comprehensive resource for researchers and developers interested in the latest advancements in computer vision and related fields. It provides access to cutting-edge research papers, making it an invaluable tool for staying updated with the latest developments in the field.

### Overview: DataFoundry is an open-source AI workbench designed for data analysis. It unifies data sources, knowledge, tools, and agent runtime into a governed workspace for interactive analytics. The system reads business definitions through unified semantics, runs complex multi-table, multi-step analysis inside read-only boundaries, and keeps every step auditable and replayable, ensuring every conclusion is trustworthy. ### Key Features: - **28 Datasource Types Out of the Box**: Connects to PostgreSQL, MySQL, Snowflake, BigQuery, ClickHouse, MongoDB, Redis, and Elasticsearch. - **Enterprise Semantics and Context Organization**: Manages schema, metric definitions, and field relationships for accurate data analysis. - **Self-Hosted Deployment**: Runs inside your own boundary, supporting any OpenAI-compatible provider for model deployment. - **Safe by Default, Auditable Throughout**: Ensures read-only queries, credential isolation, field masking, row limits, and timeouts by default, with SQL, tool calls, and event streams fully persisted and replayable. - **Deep Optimization for Complex Data Tasks**: Built for multi-table, multi-field, long-horizon analysis and multi-step reasoning, ensuring complex questions are efficiently handled. ### Tech Stack: TypeScript ### Why It Matters: DataFoundry addresses the common issue of natural-language analytics not being trustworthy in the enterprise by integrating the agent into a semantic, policy-aware, evidence-preserving data task system. This approach ensures that every step of the analysis is auditable and verifiable, making the results reliable and verifiable.

Overview: The Hailo Model Zoo is a comprehensive repository that includes pre-trained models and a full building and evaluation environment for high-performance deep learning applications. It provides pre-compiled models in ONNX and TensorFlow formats, as well as the Hailo Executable Format (HEF) for faster development and optimized performance. Key features: - Pre-trained models in ONNX and TensorFlow formats - Pre-compiled Hailo Executable Format (HEF) for device execution - Full precision accuracy measurement - Quantized accuracy using the Hailo Emulator - Ability to generate Hailo Executable Format (HEF) binary files - Models for different devices: Hailo-15H, Hailo-10H, Hailo-15L, Hailo-8, and Hailo-8L - Compatibility with Hailo Dataflow Compiler v3.x for Hailo-10 and Hailo-15 devices Tech stack: Python, TensorFlow, CUDA, Hailo Dataflow Compiler, HailoRT Why it matters: The Hailo Model Zoo simplifies the process of model development and deployment by providing pre-trained models and a comprehensive evaluation environment. It ensures high performance and accuracy, making it ideal for real-world applications requiring high-speed and low-latency deep learning solutions.

### Overview: iLab CONJURE 是一个面向 GPT-image-2 的 AI 图片生成工作台，支持 Codex Responses 和 OpenAI 兼容 API 接入，内置公用图库、多类型 Chip 快捷引用、提示词模板、多任务并发和本地队列管理。 ### Key features: - 支持 GPT Image 与 Gemini，统一模型目录支持官方协议或 OpenAI-Compatible 中转站。 - Codex 工作流继续支持默认 Codex Image 通道、Codex Responses 兼容通道。 - 多任务并发、本地队列状态、分页历史库、缩略图和结果归档。 - 停止任务会中断本地请求及等待中的图片槽位，保留已生成图片和历史记录。 - 生成页按需加载最近任务和媒体，隐藏图库与模板仅在打开后渲染。 - 输出参数支持一键锁定并以只读摘要展示，避免连续生成或浏览历史任务时误改设置。 - 独立 `/history` 页面支持 SQLite 分页、搜索、筛选、网格/列表视图和懒加载详情。 - 历史任务可收藏并添加多个标签，最多可一次整理 300 个已选任务。 - 单个或多个历史任务可导出成一个 ZIP，支持仅图片或图片＋提示词。 - 生成页与历史库共用顶部工具栏、小兔子 Logo、返回入口和跟随系统/浅色/

Overview: The Adapt Intent Parser is a flexible framework for parsing natural language text into structured intents, enabling programmatic invocation of actions. Key features: - Parses natural language text into structured intents - Supports extensible intent definitions - Provides a modular design for easy integration and customization Tech stack: Python Why it matters: The Adapt Intent Parser is essential for developers looking to integrate natural language processing into their applications, providing a robust and extensible solution for intent-based interactions.

### Overview: Frontman is an AI coding agent designed to streamline frontend fixes by editing source files directly in the browser. It leverages live DOM, computed CSS, and component trees to provide non-technical team members with precise and accurate fixes without the need for an IDE. ### Key Features: - **Live DOM Editing**: Frontman edits source files in real-time as you describe changes in the browser, ensuring the correct files are updated. - **Source Maps Integration**: It reads computed CSS and component tree data to understand the impact of your edits, ensuring the right files are modified. - **Contextual Fixes**: Frontman provides context by reading the rendered output, making it easier for non-technical team members to make informed decisions. ### Tech Stack: - **ReScript**: The primary language used for development. ### Why It Matters: Frontman offers a solution for teams where non-technical stakeholders need to make frontend fixes quickly and accurately, without waiting for developers to open an IDE. This tool bridges the gap between design and development, ensuring that changes can be made in a collaborative and efficient manner. ### Repository: The repository for Frontman is located at [frontman-ai/frontman](https://github.com/frontman-ai/frontman). ### GitHub Stars: The repository has received 719 GitHub stars, indicating a growing interest and adoption of Frontman by the developer community. ### Primary Language: The primary language used in the Frontman project is ReScript. ### Why It Matters: Frontman is designed to streamline the process of making frontend fixes, making it easier for non

### Overview: MindChat是一个旨在帮助人们缓解心理压力和解决心理困惑的心理大模型，旨在从心理咨询、心理评估、心理诊断、心理治疗四个维度提供支持。 ### Key features: - **心理咨询**: 提供心理咨询和心理支持，帮助用户解决心理问题。 - **心理评估**: 进行心理评估和测评，帮助用户了解自己的心理状态。 - **心理诊断**: 提供心理诊断服务，帮助用户识别心理问题的严重程度。 - **心理治疗**: 提供心理治疗方案，帮助用户改善心理问题。 ### Tech stack: Python ### Why it matters: MindChat通过其强大的心理评估和治疗功能，旨在帮助用户提高心理健康水平，缓解心理压力，促进心理健康发展。

Overview: This repository contains a collection of AI algorithm job resume templates, suitable for developers. The templates are designed to be concise and easy to understand, focusing on key features and technical stack. Key features: - Resume templates for AI algorithm jobs - Word version of the templates - Detailed instructions on formatting and content Tech stack: Markdown and LaTeX Why it matters: These templates are useful for developers looking to present their skills and experience effectively in the field of AI algorithm jobs. Repository: amusi/AI-Job-Resume GitHub stars: 719

Overview: MeTA is a modern C++ data sciences toolkit designed for text analysis, featuring advanced features like text tokenization, inverted and forward indexes, ranking functions, topic models, classification algorithms, graph algorithms, and more. Key features: - Text tokenization with deep semantic features - Inverted and forward indexes with compression and caching - Ranking functions for efficient search - Topic models - Classification algorithms - Graph algorithms - Language models - CRF implementation (POS-tagging, shallow parsing) - Wrappers for liblinear and libsvm - UTF8 support for analysis on various languages - Multithreaded algorithms Tech stack: C++ Why it matters: MeTA is a comprehensive toolkit for modern text analysis, making it an essential tool for researchers and developers in the field of natural language processing and data sciences.

Overview: DeepRetrieval is a novel approach that leverages reinforcement learning to train Large Language Models (LLMs) for query generation, enhancing information retrieval performance. This method eliminates the need for supervision and significantly outperforms traditional methods across various retrieval tasks. Key features: - No supervision required - Significantly outperforms previous state-of-the-art methods - Excels in diverse retrieval tasks including literature search, evidence-seeking retrieval, classic information retrieval, and SQL database search - Achieves superior results with only 3B parameters, outperforming larger models Tech stack: Python, VLLM, Hugging Face, Wandb Why it matters: DeepRetrieval provides a powerful and efficient solution for information retrieval, making it an attractive choice for developers looking to improve their search engine capabilities without the need for extensive human annotation or model fine-tuning. Repository: pat-jj/DeepRetrieval

### Overview: Horizon is a GPU-accelerated terminal board that seamlessly integrates all your sessions into an infinite canvas, offering a structured workflow and customizable panels for a dynamic and productive workspace. ### Key features: - **Infinite Canvas**: Workspaces group related panels and sessions on an infinite 2D surface. - **Command Palette**: Quickly jump to the command palette for quick access to commands. - **Preset Panels**: Easily launch predefined panels like SSH, Grok, or Code Agents. - **Workspace Management**: Organize workspaces with color-coded clusters and customizable layouts. ### Tech stack: Rust ### Why it matters: Horizon transforms how you manage multiple terminal sessions, providing a clean, structured workflow and a dynamic workspace that adapts to your needs.

Overview: The Prompt Fuzzer is an open-source tool designed to assess the security of GenAI applications' system prompts against various dynamic LLM-based attacks. It provides a security evaluation based on simulated attacks, helping developers harden their system prompts. Key features: - Supports 16 different LLM providers - Includes 16 different attacks - Offers interactive and CLI modes - Supports multi-threaded testing Tech stack: Python Why it matters: The Prompt Fuzzer helps developers ensure their GenAI applications are secure by simulating and evaluating potential security threats, making them stronger against generative AI attacks. Repository: prompt-security/ps-fuzz Primary language: Python GitHub stars: 719 GitHub description: Make your GenAI Apps Safe & Secure :rocket: Test & harden your system prompt README excerpt: <h1 align="center"> <img src="resources/prompt-icon.svg" alt="prompt-icon"> Prompt Fuzzer <img src="resources/prompt-icon.svg" alt="prompt-icon"> </h1> <h2 align="center"> The open-source tool to help you harden your GenAI applications <br> <br> [](https://opensource.org/licenses/MIT)   ![Last

### Overview: This repository is a curated list of projects, works, people, articles, and resources focused on creating art using machine learning. It includes projects in music, visual art, and more, with a particular emphasis on generative and interactive art. ### Key Features: - **Projects**: A diverse collection of projects ranging from generative art to interactive installations. - **People**: Profiles of notable artists and researchers in the field of machine learning and art. - **Libraries**: Various open-source libraries and tools used in the projects. - **TODO**: A placeholder for future additions or updates. ### Tech Stack: Python, JavaScript, TensorFlow, PyTorch, and other relevant machine learning frameworks. ### Why it Matters: This repository serves as an inspiring resource for artists and developers interested in exploring the intersection of machine learning and art. It provides practical examples and insights into how machine learning can be used to create innovative and visually engaging art.