Projects

### Overview: This Specialization on Generative Adversarial Networks (GANs) from deeplearning.ai covers three courses designed to provide an accessible introduction to image generation with GANs. The courses are foundational to advanced techniques and cover social implications, including bias in ML and privacy preservation. ### Key Features: - **Build Basic Generative Adversarial Networks (GANs)**: Learn about GANs and their applications, understand the intuition behind their components, and build your own GAN using PyTorch. - **Build Better Generative Adversarial Networks (GANs)**: Build a more sophisticated GAN using convolutional layers, learn about useful activation functions, batch normalization, and transposed convolutions. - **Apply Generative Adversarial Networks (GANs)**: Train your own model using PyTorch, use it to create images, and evaluate advanced GANs. ### Tech Stack: Jupyter Notebook ### Why It Matters: - **Introduction to GANs**: Understand the basics of GANs and their applications. - **Foundation in GANs**: Build a comprehensive knowledge base and gain hands-on experience in GANs. - **Advanced Techniques**: Train your own model using PyTorch, apply it to create images, and evaluate advanced GANs. - **Social Implications**: Learn about the social implications of GANs, including bias in ML and privacy preservation. ### GitHub Stars: 482 ### Repository: amanchadha/coursera-gan-specialization ### Primary Language: Jupyter Notebook

### Overview: FROMAGe is a model designed to ground language models to images for multimodal inputs and outputs, as presented in the ICML 2023 paper "Grounding Language Models to Images for Multimodal Inputs and Outputs". The repository includes the code and model weights for the FROMAGe model, which is trained on the Conceptual Captions dataset and can be used for image retrieval and dialogue tasks. ### Key Features: - **Model Architecture**: The FROMAGe model is designed to integrate language models with images, enabling multimodal processing. - **Pretrained Checkpoints**: The repository provides precomputed embeddings for Conceptual Captions images, which can be used for image retrieval. - **Pretrained Checkpoints (Strong Visual Model)**: A stronger visual model with 4 visual tokens is available, which generally performs better on dialogue settings and requires less tuning for inference time. - **Inference Examples**: The repository includes examples of how to call the model for inference, demonstrating its capabilities on various datasets. ### Tech Stack: - **Programming Languages**: The code is written in Jupyter Notebook. ### Why It Matters: FROMAGe is significant because it addresses the challenge of integrating language models with images, enabling applications that can better understand and respond to multimodal inputs and outputs. This model is particularly useful in scenarios where language models need to be able to interpret and generate images, such as in chatbots or virtual assistants. ### Repository: - **Repository**: kohjingyu/fromage - **GitHub Stars**: 482 - **Primary Language**:

Overview: Geti™ is a powerful platform designed to accelerate the development of computer vision AI models, offering a seamless workflow from dataset preparation to deployment. It includes features such as interactive model training, smart annotations, and optimized deployment across Intel XPU. Key features: - Interactive Model Training - Smart Annotations Tech stack: TypeScript Why it matters: Geti™ simplifies the process of building and deploying computer vision models, making it accessible to developers and researchers alike. Repository: open-edge-platform/geti_v2 Primary language: TypeScript GitHub stars: 482 GitHub description: Legacy repository for Geti v2.x. For Geti v3.0+, visit https://github.com/open-edge-platform/geti. This repository contains the source code and history for Geti v2.13 and earlier only. It is no longer under active development and is maintained for historical and archival purposes. For new features, bug fixes, documentation, and releases, please use the new repository: [https://github.com/open-edge-platform/geti](https://github.com/open-edge-platform/geti).

### Overview: 基于Spring Boot+Vue+ElementUI的人力资源管理系统，旨在提升企业人力资源管理效率与员工体验。 ### Key features: - **业务流程深度解耦与纯计算领域设计**：考勤核算、加班计算、薪资核算等流程无框架依赖，单测覆盖与维护确定性高。 - **Flowable 8.0 流程引擎原生适配**：深度整合Spring Boot 3.4与Flowable引擎，流程表独立存放，业务状态与流程推进共享本地事务。 - **双 Token 自动无感续期机制**：基于httpOnly Cookie的双 Token（Access Token + Refresh Token）鉴权体系，前端拦截器透明处理并发无感刷新。 - **混合持久化与企业级RAG知识检索**：采用MySQL、Redis、PostgreSQL/pgvector、MinIO的多引擎存储架构，内置文档切块、向量化ETL流水线，助力企业政策与制度高精度语义召回。 - **企业知识库与智能问答**：搭载兼容DashScope/OpenAI规范的智能问答服务，提供基于企业知识库的“即问即答”。 ### Tech stack: - 后端核心架构: - **Java**: 17 - **Spring Boot**: 3.4.4 - **MyBatis-Plus**: 3.5.10 - **Flowable**: 8.0.0 - **Spring AI**: 1.0.0 (

Overview: This project automates the process of detecting license plates using YOLOv8, a state-of-the-art object detection model. Key features: - Automatic detection of license plates in videos. - Utilizes a pre-trained YOLOv8 model for efficient plate detection. - Includes a licensed plate detector for additional accuracy. - Available model for download. Tech stack: Python. Why it matters: This project streamlines the detection process, making it easier and faster to identify license plates in various video applications, enhancing security and efficiency.

### Overview ManimCat is a project that leverages AI to generate mathematical animations from natural language descriptions. It is designed to be a versatile tool for classroom explanations, worked examples, and visual reasoning tasks. ManimCat supports both static and dynamic visual outputs, making it useful for a wide range of educational purposes. ### Key Features - **Natural Language Generation**: ManimCat allows users to describe mathematical concepts and generate animations based on these descriptions. - **Interactive Workflow**: Supports direct workflow generation and agent-driven studio collaboration. - **High-Quality Rendering**: ManimCat uses LaTeX support and auto-code fixing to ensure high-quality animations. - **Integration with ManimGX and Matplotlib**: ManimCat integrates with ManimGX for animation and timeline-based storytelling, and Matplotlib for static mathematical visuals. - **Standalone Studio**: Manim Studio is the animation-oriented Studio path, while Plot Studio is the more mature Studio path for static visual work and iterative editing. ### Technology Stack - **Frontend**: TypeScript - **Backend**: ManimGX, React, Node.js - **Dependencies**: Manim, Matplotlib, ManimVideoGenerator - **License**: AGPL-3.0-or-later ### Why It Matters ManimCat simplifies the process of creating mathematical animations, making it accessible and efficient for educators and students alike. By integrating natural language generation with advanced AI tools, ManimCat transforms mathematical explanations into engaging, interactive animations, enhancing the learning experience. ### Deployment ManimCat is designed to be deployed on a server or a

Overview: ProcTHOR is a Python-based framework for generating procedurally generated 3D houses that are interactive, diverse, and semantically plausible, aiming to scale embodied AI. Key features: - Procedural generation of interactive, diverse, and semantically plausible houses - Compatible with AI2-THOR - Supports AI2's 3D environment for embodied AI research Tech stack: Python Why it matters: ProcTHOR provides a scalable solution for generating realistic and diverse 3D environments, which is crucial for advancing embodied AI research and creating more realistic testbeds for AI systems.

Overview: This repository contains a comprehensive list of AI residency programs across various years, including internships, bootcamps, and research opportunities. Key features: - All year internships: MILA, DeepkaphaAI, VinAI AI Residency - All year bootcamps (hands-on supervised projects): Fellowship.AI, PiCampus School - 2023 programs: OpenAI Residency-Research, OpenAI Residency-Software Engineering, Meta AI Residency, Apple The 2023 AI/ML Residency Program, DeepLabCut AI Residency, Salesforce AI Residency Program, Toyota Research Institute AI Resident, Google X AI Residency, Google AI Resident (Health), OpenAI 2020 Winter Scholars - 2022 programs: OpenAI Residency, Anthropic, Hugging Face, NVIDIA AI Research Residency, Novartis AI for Life Residency Program, Allen AI Predoctoral Young Investigator - 2021 programs: Apple AI Residency Program, Google AI Resident (Health), IBM AI Residency Program, Shell.ai Residency Programme, Google AI Resident (Health), OpenAI 2020 Winter Scholars - 2020 programs: Google X AI Residency, Google AI Resident (Health), OpenAI 2020 Winter Scholars, AI Residency Applied to Industry for professionals, AI Residency Applied to Industry for companies, Uber AI Residency Program, Shell AI Residency Program Tech stack: None specified Why it matters: These

Overview: MIRA is a best-effort approximation of a continuous digital entity designed to provide long-horizon persistence through asynchronous conversation processing and self-directed context window manipulation. It features a blend of REM sleep-like conversation processing and active self-directed context window manipulation, ensuring believable persistence within an ephemeral framework. Key features: - Continuous conversation history with active conversation history staying live while relevant, older material collapsing into first-person memories, and active conversation segments being compressed into rolling continuations. - First-person memory generation, where summaries are generated based on previous five summaries, allowing for a continuously advancing narrative in the conversation history. - Memory management designed to decay naturally through use-days, with explicit references, links, and temporal relevance to prevent decay. - Domaindoc_tool for stable encrypted documents, allowing collaboration on stable encrypted documents with version history, sharing, pinned sections, and nested subsections. - Document expansion and collapse capabilities to mitigate token explosion in longform content. Tech stack: Python Why it matters: MIRA is a comprehensive best-effort approximation of a continuous digital entity, designed to provide long-horizon persistence through asynchronous conversation processing and self-directed context window manipulation. Its unique features, such as continuous conversation history, first-person memory generation, and stable encrypted documents, make it a valuable tool for developers looking to build believable persistence within an ephemeral framework. GitHub stars: 481

Overview: COPE is a modern IDE designed specifically for writing classical Chinese poetry, including 5-character and 7-character Jueju and Lüshi, with built-in dictionaries and tone pattern recognition. Key features: - A tool for writing, editing, and managing user poems. - Automatically recognizes tone patterns and detects errors. - Supports all 4 variations of 5-character and 7-character Jueju and Lüshi. - Built-in dictionaries for both the "平水韵" and "中华新韵" as well as the "康熙字典". - Provides insights into user poems using data analysis. Tech stack: JavaScript Why it matters: COPE simplifies the process of composing and editing classical Chinese poetry, offering a user-friendly interface and comprehensive features that enhance the quality of the final work.

### Overview: RAGNotebook 是一个基于LangChain、LangGraph、FastAPI和React构建的智能笔记助手项目。它集成了笔记管理、知识图谱、RAG与AI写作辅助功能，旨在解决个人知识管理的痛点问题。 ### Key features: - **笔记管理**: 支持 Markdown 编辑器、笔记自动写入知识图谱、智能标签、语义搜索、Markdown 导出。 - **知识图谱**: 通过LLM抽取实体与关系存入Neo4j，提供可视化画布浏览，检索时沿图扩展证据。 - **RAG与AI写作辅助**: 会话持久化、向量检索、AI联机补全、续写/扩写/摘要生成、关联笔记推荐、间隔重复回顾。 - **多语言支持**: 前端i18n，中英文界面切换。 ### Tech stack: Python ### Why it matters: RAGNotebook 是一个集成了笔记管理、知识图谱、RAG与AI写作辅助功能的智能笔记助手，旨在帮助用户高效管理知识和笔记，解决个人知识管理的痛点问题。

### Overview: Listen, Think, and Understand is a repository containing the official implementation, pretrained checkpoints, and datasets for the first generation of audio and speech large language model, LTU, and the second generation model, LTU-AS. These models bridge audio/speech perception with understanding and achieve state-of-the-art performance on various closed-ended tasks. The repository is designed for developers and researchers to easily access and utilize these models for their projects. ### Key features: - Official PyTorch implementation of LTU and LTU-AS - Pretrained checkpoints available for download - OpenAQA and OpenASQA datasets for training and evaluation - Interactive demos for demonstration of model capabilities ### Tech stack: Python ### Why it matters: These models offer a significant advancement in audio and speech understanding, enabling models to not only comprehend audio but also answer open-ended questions based on the given audio. They are particularly useful in applications requiring natural language processing of audio data, such as voice recognition, speech-to-text, and more. ### Repository: YuanGongND/ltu ### Primary language: Python ### GitHub stars: 480 ### GitHub description: Code, Dataset, and Pretrained Models for Audio and Speech Large Language Model "Listen, Think, and Understand".

### Overview: 本项目提供了一系列关于MindSpore的在线课程，旨在帮助开发者深入理解大模型技术及其应用。课程涵盖了从Transformer到GPT的多方面内容，包括模型设计、训练方法和实际应用案例。 ### Key features: - **理论实践结合**：通过实践指导，帮助学员掌握大模型开发的核心技能。 - **专家解读**：邀请领域专家进行深入解读，提供多角度理解。 - **开源共享**：课程内容和代码均开源，方便学习和分享。 - **ICT大赛赋能**：通过参与ICT大赛，提升实际项目经验。 - **系列课程**：涵盖大模型专题，持续更新。 ### Tech stack: Jupyter Notebook ### Why it matters: 这些课程为开发者提供了系统的学习路径，帮助他们在实际项目中应用大模型技术，提升开发效率和项目质量。 ### Repository: candle-org/step_into_llm ### GitHub stars: 480 ### GitHub description: MindSpore online courses: Step into LLM

### Overview: Fundus is a simple yet effective news crawler written in Python, designed to help developers and researchers easily extract news articles from various online sources. It's particularly useful for those looking to automate the process of gathering news content for analysis or publication. ### Key Features: - **Pythonic Interface**: Fundus is a straightforward and easy-to-use package that can be integrated into any Python project. - **Scrape Multiple Publishers**: Fundus allows you to scrape articles from multiple news publishers, including those from the CC-NEWS dataset. - **Minimal Dependencies**: Requires only Python 3.8+ to run, making it lightweight and easy to install. ### Tech Stack: - Python 3.8+ ### Why It Matters: Fundus simplifies the process of extracting news articles, making it accessible for developers and researchers without requiring extensive coding skills. Its modular design and ability to integrate with various news sources make it a valuable tool for anyone looking to automate news gathering and analysis. ### Repository: The repository for Fundus is hosted on GitHub, where it has received 479 stars. This indicates a significant amount of interest and adoption in the project. ### Primary Language: The primary language used in the project is Python. ### GitHub Description: A very simple news crawler with a funny name. Fundus is designed to help developers and researchers easily extract news articles from various online sources. ### GitHub Description (Continued): Developed at Humboldt University of Berlin. Fundus is built on the idea of building something together, welcoming contributions to help it grow. ### PyPi Version:

### Overview TorchCP is a Python toolbox designed for conformal prediction research on deep learning models, utilizing PyTorch. It includes methods for classification, regression, graph neural networks, and language models, with strong GPU acceleration. The toolbox is based on the PyTorch Library and is maintained by the Hongxin Wei research group at SUSTech. ### Key Features - Implementations of various conformal prediction methods for classification, regression, and graph neural networks. - Strong GPU acceleration for efficient computation. - Comprehensive documentation and examples provided. ### Tech Stack - Python - PyTorch ### Why It Matters TorchCP provides a practical solution for conformal prediction in deep learning, making it an essential tool for researchers and practitioners in the field of machine learning and artificial intelligence.

### Overview: Prompt Decorators are structured prefixes designed to enhance AI responses, inspired by Python decorators. They simplify interactions by enforcing specific response rules without requiring lengthy instructions, making the output more logical, accurate, and well-organized. ### Key Features: - **Cognitive & Generative Decorators:** Guide reasoning, exploration, and idea development. - **Expressive & Systemic Decorators:** Control output formatting and session management. ### Tech Stack: Structured prefixes for AI responses. ### Why It Matters: Simplifies interactions for users by enforcing specific response rules without requiring lengthy instructions, making the output more logical, accurate, and well-organized. ### Repository: smkalami/prompt-decorators ### GitHub Stars: 480 ### Primary Language: Unknown ### License: Not specified ### Citation: BibTeX Citation: ```tex @misc{smkalami-prompt-decorators, author = {smkalami}, title = {Prompt Decorators Framework}, year = {2023}, url = {https://github.com/smkalami/prompt-decorators} } ```

Overview: This project introduces CLIP Surgery, a method for improving the explainability of CLIP, a pre-trained language-image model, by addressing two key issues: opposite visualization and noisy activations. The technique does not require fine-tuning or additional supervision, making it highly effective for enhancing the explainability of CLIP across various downstream tasks. Key features: - CLIP Surgery addresses opposite visualization and noisy activations in CLIP's raw predictions. - It significantly improves the explainability of CLIP and enhances open-vocabulary tasks such as multi-label recognition, semantic segmentation, and interactive segmentation. - A simple demo is provided for interpretability analysis, and guidance on converting text to point prompts for SAM is included. Tech stack: Jupyter Notebook Why it matters: CLIP Surgery offers a practical solution for enhancing the explainability of CLIP, making it easier to understand and interpret the model's predictions, thereby improving its utility in various applications.

### Overview: Deta_Parser 是一个快速中文分词分析工具，专为文本挖掘、自然语言处理（NLP）、词性标注（POS）、人工智能和深度学习领域设计。它能够处理高达2200万到3000万词汇的中文混合分词任务，具有高精准率和快速神经网络分词包。 ### Key features: - 快速中文分词分析 - 支持标点符号分离 - 多语言支持（12国语言） - 高精准率和快速分词速度 - 可扩展词库和词意分析 - 免费的官方互动展示页地址 ### Tech stack: Java ### Why it matters: Deta_Parser 是一个开源的个人著作权项目，旨在提供高精度的中文分词解决方案，适用于文本挖掘、自然语言处理、词性标注和深度学习等领域的应用。其快速分词速度和高精准率使其成为这些领域的重要工具。

Overview: CogComp's Natural Language Processing Libraries and Demos are a collection of modules designed for various NLP tasks, including lemmatization, named entity recognition, part-of-speech tagging, and more. Key features: - Lemmatizer - Named Entity Recognition (NER) - Part-of-Speech (POS) tagging - Prepositional Phrase Semantic Role Labeling (PP-SRL) - Quantifier extraction - Relation extraction - Similarity comparison - Temporal normalizer - Tokenizer - Transliteration - Verb Sense Disambiguation (VSD) Tech stack: Java Why it matters: These libraries and demos are essential for developers working on NLP projects, providing a comprehensive set of tools for tasks such as text annotation, training NLP models, and feature extraction.

Overview: PaperRobot is a tool designed for generating incremental draft ideas for scientific papers. It leverages the PubMed dataset to predict relationships between entities and improve the quality of scientific writing. Key features: - Predicts relationships between entities in scientific papers - Improves the quality of scientific writing through incremental draft generation - Supports both existing paper reading and new paper writing Tech stack: Python Why it matters: PaperRobot offers a practical solution for researchers looking to enhance the clarity and relevance of their scientific papers by leveraging existing knowledge and improving the predictive power of their writing process.