Projects

### Overview: This course is designed to teach developers how to build a production-ready LLM and RAG system using best practices from the LLMOps framework. ### Key Features: - Learn to architect and build an end-to-end production-grade LLM system. - Gain hands-on experience with microservices, vector databases, and MLOps best practices. - Build and deploy your own LLM twin using LLMOps good practices. - Data collection, pipeline, feature extraction, fine-tuning, and inference using Python microservices. ### Tech Stack: Python ### Why it Matters: This course provides a comprehensive guide to building a production-ready LLM and RAG system, ensuring developers can confidently integrate LLMs and RAGs into their projects.

### Overview: LimiX is a large foundation model designed for structured data modeling, specifically tailored for generalist intelligence. It leverages contextual mechanism networks (CMN) to learn context-dependent representations of data generation, outperforming existing models on various benchmarks. ### Key features: - **Contextual Mechanism Network (CMN)**: A novel paradigm for learning context-dependent representations. - **Synthetic Datasets**: Pretrained with context-conditional masked modeling (CCMM) using structural causal models (SCMs). - **Tabular Predictive Performance**: Performs classification, regression, and missing-value imputation in one forward pass. - **Lightweight Model**: LimiX-2M, a lightweight version, reduces GPU memory usage and inference time without compromising quality. ### Tech stack: Python ### Why it matters: LimiX-2 and LimiX-2M are pivotal advancements in the field of large foundation models for structured data, providing a unified framework for generalist intelligence. Their ability to learn context-dependent representations and their causal-aware feature attention make them highly valuable tools for applications requiring structured data modeling.

### Overview: Flint is a visualization language designed for AI agents, enabling them to create expressive and polished charts from simple, human-editable specifications. Flint translates these specifications into native Vega-Lite, ECharts, Chart.js, and Plotly formats, making it easier for AI agents to generate high-quality visualizations without needing to hardcode complex parameters. ### Key Features: - **Semantic Chart Specs:** Flint uses 70+ semantic types to capture the meaning of each field, ensuring accurate representation of data. - **Automatic Layout:** Flint adjusts sizing, spacing, labels, marks, and legends based on the data cardinality, chart design, and canvas constraints, ensuring optimal visual representation. - **Formal Visual Themes:** Flint allows defining layout behavior, semantic presentation, and visual identity once, then applying them across a chart library with a preset, custom `ThemeSpec`, or inherited theme. - **Multiple Backends:** Flint compiles one input into backend-native output across Vega-Lite, ECharts, Chart.js, and Plotly formats, supporting a wide range of visualization libraries. ### Tech Stack: - JavaScript/TypeScript ### Why It Matters: Flint simplifies the process of creating visually appealing charts by abstracting away the complexities of charting parameters, allowing AI agents to focus on the data and design. This makes Flint particularly useful in scenarios where automation and efficiency are crucial, such as in data analysis and AI-driven decision-making processes. Repository: microsoft/flint-chart Primary Language: TypeScript GitHub Stars: 4386 GitHub Description: 🪄

### Overview: LLPlayer is a media player designed specifically for language learning, featuring dual subtitles, AI-generated subtitles, real-time translation, and more. It supports multiple subtitle formats and engines, making it an essential tool for language enthusiasts and learners. ### Key Features: - **Dual Subtitles:** Two subtitles can be displayed simultaneously, supporting text and bitmap subtitles. - **AI-generated subtitles (ASR):** Real-time automatic subtitle generation using Whisper, powered by OpenAI Whisper and faster-whisper. - **Real-time Translation:** Supports Google, DeepL, Ollama, LM Studio, and OpenAI engines. - **Context-aware Translation:** Highly accurate translation using language models. - **Real-time OCR subtitles:** Converts bitmap subtitles to text subtitles in real time using Tesseract OCR. - **Subtitles Sidebar:** Supports text and bitmap subtitles, with seek and word lookup capabilities. - **Instant word lookup:** Word lookup and browser searches can be performed on subtitle text. - **Customizable Browser Search:** Customizable search sites can be used from the context menu. - **Plays online videos:** Integrates with yt-dlp for real-time video playback with AI subtitle generation and word lookups. - **Flexible Subtitles Size/Placement Settings:** Adjustable subtitle sizes and positions. - **Subtitles Seeking for any format:** Supports any subtitle format for subtitle seek. - **Built-in Subtitles Downloader:** Supports opensubtitles.org. - **Integrate with browser extensions:** Compatible with Yomitan and 10ten. - **Customizable Dark Theme:** Based on black with

Overview: Sacred is a comprehensive tool designed to help developers configure, organize, log, and reproduce experiments efficiently at IDSIA. Key features: - Config Scopes - Config Injection - Command-line Interface - Observers - Automatic seeding Tech stack: Python Why it matters: Sacred simplifies the tedious work involved in setting up and running experiments, ensuring reproducibility and ease of use for researchers and developers alike.

**Overview:** This repository serves as a comprehensive guide for deploying deep learning models in production, covering various aspects such as converting PyTorch models, integrating PyTorch with other technologies, and serving PyTorch models on different platforms. The repository includes numerous tutorials, articles, and examples that are essential for developers looking to deploy their models in a production environment. **Key features:** - PyTorch Production Level Tutorials - PyTorch 1.0 tracing JIT and LibTorch C++ API - Deploying PyTorch models on Flask - Serving PyTorch models on AWS Lambda with Caffe2 & ONNX - FastAI PyTorch in Production - OpenMMLab Model Deployment Framework - PyTorch Native Architecture Optimization: torchao - PyTorch C++ API - PyTorch via C++ **Tech stack:** Python, PyTorch, Flask, AWS Lambda, C++, ONNX, FastAI, OpenMMLab, WebDNN, LitServe, PyTorch Serve, PyTorch Native Architecture Optimization, AUTOGRADPP **Why it matters:** This repository is crucial for developers who need to deploy their deep learning models in a production environment. It provides a wealth of practical knowledge and examples that are essential for anyone looking to integrate PyTorch models into their production systems. The repository covers a wide range of topics, from model conversion to deployment on various platforms, making it a valuable resource for anyone working on deploying deep learning models in production.

### Overview: This project is a comprehensive adaptation of the book "Dive Into Deep Learning" by Aston Zhang, Zachary C. Lipton, Mu Li, Alex J. Smola, and the community contributors. It has been ported from the original [Dive Into Deep Learning](https://d2l.ai/) book to PyTorch, making it accessible and user-friendly for developers. ### Key Features: - Adapted code from MXNet into PyTorch - Includes chapters on installation, introduction, data manipulation, linear regression, softmax regression, multilayer perceptrons, and model selection - Contains detailed explanations and practical examples ### Tech Stack: - Jupyter Notebook ### Why It Matters: This project is essential for anyone looking to understand and implement deep learning concepts, as it provides a clear and concise guide to the book's content using modern, user-friendly tools like PyTorch.

Overview: neurojs is a JavaScript framework designed for deep learning and reinforcement learning, focusing on reinforcement learning but versatile for any neural network-based task. It includes visual demos like a 2D self-driving car and supports uniform and prioritized replay buffers, advantage learning, and deep Q-networks and actor-critic models. Key features: - Full-stack neural-network based machine learning framework - Extended reinforcement-learning support - Binary import and export of network configurations - High-performance Tech stack: JavaScript Why it matters: neurojs is a valuable tool for developers looking to implement deep learning and reinforcement learning in the browser, offering neat demos and a community-driven approach. Repository: janhuenermann/neurojs Primary language: JavaScript GitHub stars: 4367 GitHub description: A JavaScript deep learning and reinforcement learning library. README excerpt: [](https://github.com/janhuenermann/neurojs) Key features: - Implements a full-stack neural-network based machine learning framework - Extended reinforcement-learning support - Binary import and export of network configurations - High-performance Running the examples: ```bash npm install npm start ``` Open `http://localhost:8080/examples/` in your browser and select the demo you want to run.

Overview: ModLens is a plugin for DeepSeek Harness that allows text-only models to read and understand images, providing structured JSON evidence for various DeepSeek models. Key features: - Reads images pasted into the chat interface. - Supports multiple DeepSeek, GLM, and MiMo Pro models. - Automatically discovers and adds entry points for eligible text-only models. - Provides structured JSON evidence for the pasted images. Tech stack: TypeScript Why it matters: ModLens simplifies the integration of visual information into text-based models, making them more versatile and powerful.

Overview: PocketSphinx is a small, efficient speech recognizer developed by Carnegie Mellon University, dating back to the 1970s but still useful for many applications due to its compactness and efficiency. Key features: - Compact and efficient - Live speech recognition - Command-line interface - Python bindings - C library Tech stack: C Why it matters: PocketSphinx is a valuable tool for speech recognition applications, providing a lightweight solution for continuous speech recognition that can be easily integrated into various projects.

Overview: TorchServe is a tool designed to serve and scale PyTorch models in production environments, providing a flexible and easy-to-use solution for deploying models in real-world applications. Key features: - Real-time model serving and scaling - Token authorization enabled by default for enhanced security - Model API control disabled by default to prevent potential malicious code introduction - Support for both nightly builds and Docker containers - Quick start options for both Python and conda environments - Quick start guide for a smooth deployment experience - Quick start guide for deploying with the VLLM engine - Quick start guide for deploying with the TRT-LLM engine Tech stack: Python, PyTorch, Docker, and Hugging Face's Transformers Why it matters: TorchServe simplifies the process of deploying PyTorch models in production, ensuring models can be served and scaled efficiently, while also providing enhanced security features to protect against unauthorized API calls and potential malicious code introduction.

Overview: This project offers a suite of AI agent skills designed for social media research, leveraging ScrapeCreators to analyze TikTok, Instagram, YouTube, Reddit, Twitter, LinkedIn, Facebook, Threads, Bluesky, Pinterest, Rumble, and more. It includes skills for finding outliers, mining comments, competitor research, ad library analysis, and trend discovery. Key features: - Outlier post finder - Transcript intelligence - Comment mining - Competitor social research - Ad library teardown - Trend discovery - Influencer prospecting - Audience research - Social listening brief - Product demand research - Creator profile teardown - Content repurposing - ScrapeCreators API Tech stack: Python Why it matters: This project provides a comprehensive set of research skills for AI agents, enabling them to run complete research workflows across various social media platforms, making it ideal for creators, marketers, and researchers looking to leverage social media data effectively. Repository: ScrapeCreators/social-media-research-skills Primary language: Python GitHub stars: 4037 GitHub description: AI agent skills for social media research, powered by ScrapeCreators.

### Overview: This repository contains a comprehensive set of Tensorflow tutorials, covering basic to advanced concepts. The tutorials are designed to help developers build neural networks and understand various architectures and techniques. ### Key features: - **Tensorflow Basic**: Session, Placeholder, Variable, Activation, Regression, Classification, Save and Reload, Optimizers, Tensorboard, Dataset. - **Advanced Neural Networks**: CNN, RNN, AutoEncoder, DQN Reinforcement Learning, GAN, Transfer Learning. - **Others (WIP)**: Dropout, Batch Normalization, Visualize Gradient Descent, Distributed Training. ### Tech stack: Python ### Why it matters: These tutorials provide a structured learning experience for developers, covering both foundational concepts and advanced techniques in Tensorflow. The repository includes video and text tutorials in Chinese, making it accessible for a wide audience. ### Repository: - **Repository**: MorvanZhou/Tensorflow-Tutorial - **Stars**: 4335 - **Description**: Tensorflow tutorial from basic to hard, 莫烦Python 中文AI教学 ### Primary language: - **Primary language**: Python ### GitHub description: - **Description**: Tensorflow tutorial from basic to hard, 莫烦Python 中文AI教学

Overview: 53AI Hub is an open-source AI portal and knowledge base designed for managing enterprise knowledge, AI agents, prompts, and AI tools. It seamlessly integrates with Coze, Dify, FastGPT, RAGFlow, and 53AI Studio, supporting seamless integration with cloud platforms like Aliyun, Tencent Cloud, and Baidu Cloud. Key features: - Supports integration with mainstream agent development platforms, cloud services, and large language model platforms. - Provides full lifecycle management for AI agents, prompts, and AI tools. - Supports the operation of both registered users and internal users. - Supports one-click deployment on both cloud and local environments, and binding to a custom domain name. - Provides a customizable interface with multiple styles. - Offers enterprise-grade access permissions and enterprise-level features. Tech stack: Go Why it matters: 53AI Hub simplifies the process of building and operating AI portals, making it accessible to developers and enterprises alike. It supports integration with popular AI platforms and cloud services, reducing the complexity of AI implementation.

### Overview: Finance is a comprehensive Python toolkit designed for quantitative finance, including stock analysis, technical indicators, strategy backtesting, portfolio optimization, and financial modeling. It supports calculations using explicit inputs, small dependencies, and tested execution conventions. ### Key Features: - Stock analysis and technical indicators - Strategy backtesting and optimization - Portfolio optimization and simulation - Financial modeling and forecasting - Data retrieval and analysis - Sentiment analysis and market insights ### Tech Stack: Python 3.12 or newer ### Why It Matters: Finance is a powerful tool for quantitative finance professionals, enabling robust market data analysis, strategy research, and optimization. It simplifies complex financial calculations and provides a solid foundation for quantitative trading and investment strategies. ### Repository: shashankvemuri/Finance - Primary language: Python - GitHub stars: 4333 - GitHub description: Python toolkit for quantitative finance: stock analysis, technical indicators, strategy backtesting, portfolio optimization, and financial modeling. - README excerpt: The repository includes core functionalities for market data, technical indicators, financial analysis, stock screening, strategy research, backtesting, portfolios, statistical models, and more. It supports explicit inputs, small dependencies, and tested execution conventions. Examples are provided for various features, including backtesting, optimization, and research tools. ### Install: ```bash git clone https://github.com/shashankvemuri/Finance.git cd Finance python -m venv .venv source .venv/bin/activate # Windows: .venv\Scripts

Overview: LightLLM is a lightweight Python-based LLM inference and serving framework designed for high-speed performance and easy scalability. Key features: - Lightweight design - Easy scalability - High-speed performance Tech stack: Python Why it matters: LightLLM is optimized for large language models, providing efficient inference and serving capabilities with minimal resource consumption.

### Overview This section contains the papers that overview the general trends in recent natural language processing with big (pretrained) models. 1. **OpenPrompt: An Open-source Framework for Prompt-learning.** Preprint. - **Keywords Convention:** T5, Continuous_Template, Generation, Analysis - **Tech stack:** Python, PyTorch, Transformers - **Why it matters:** OpenPrompt provides a comprehensive framework for prompt learning, enabling researchers to easily adapt pre-trained models for various tasks without needing to manually design prompts. 2. **Pre-Trained Models: Past, Present and Future.** Preprint. - **Keywords Convention:** T5, Continuous_Template, Generation, Analysis - **Tech stack:** Python, PyTorch, Transformers - **Why it matters:** This paper offers a comprehensive survey of prompt learning methods in NLP, highlighting their strengths and weaknesses, and providing a roadmap for future research. 3. **Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in Natural Language Processing.** Preprint. - **Keywords Convention:** T5, Continuous_Template, Generation, Analysis - **Tech stack:** Python, PyTorch, Transformers - **Why it matters:** This survey provides a systematic overview of prompting methods in NLP, covering various aspects such as task selection, prompt design, and evaluation metrics, making it a valuable resource for researchers. 4. **Paradigm Shift in Natural Language Processing.** Machine Intelligence Research. - **Keywords Convention:** T5,

Overview: OptiLLM is an optimizing inference proxy designed to enhance the accuracy and performance of large language models (LLMs) without requiring additional training. Key features: - Instant improvements of 2-10x better accuracy on reasoning tasks. - Drop-in replacement for any OpenAI-compatible API endpoint. - 20+ optimization techniques, including MARS, CePO, AutoThink, LongCePO, and MOA. - Zero training required, using proxying of existing API calls. - Production-ready, used by companies and researchers worldwide. - Supports multiple providers, including OpenAI, Anthropic, Google, Cerebras, and 100+ models via LiteLLM. Tech stack: Python. Why it matters: OptiLLM leverages advanced optimization techniques to achieve significant improvements in LLM accuracy and performance on reasoning tasks, making it a valuable tool for developers and researchers looking to enhance model performance without the need for extensive training. Repository: algorithmicsuperintelligence/optillm GitHub stars: 4319.

Overview: Lemo-Opuscar is a collection of 51 short films, each featuring a unique film style created entirely in code. The films are accompanied by a style prompt and a demo film, showcasing 51 different styles ranging from 1927 to the present day. Key features: - 51 film styles, each with a short film made entirely in code. - A guide to creating your own style and getting a 15-30 second test film. - 8 new evidence films that film real material, every number measured or sourced. - A style prompt JSON file for easy reuse. Tech stack: JavaScript Why it matters: Lemo-Opuscar provides a unique way to explore film styles through code, offering a hands-on experience for those interested in the technical aspects of filmmaking. Repository: lemomo-ai/lemo-opuscar GitHub stars: 4323

Overview: Deep Daze is a simple command-line tool designed for text-to-image generation using OpenAI's CLIP and Siren, inspired by the work of Ryan Murdock. This tool aims to democratize the creation of psychedelic and abstract images through a straightforward interface. Key features: - Text-to-image generation - Supports CLIP and Siren techniques - Optimized for GPU performance - Compatible with both Python and Colab environments Tech stack: Python Why it matters: Deep Daze simplifies the process of generating psychedelic and abstract images, making it accessible to artists, designers, and anyone interested in exploring the boundaries of creativity through text-to-image techniques. Repository: lucidrains/deep-daze GitHub stars: 4315 GitHub description: Simple command line tool for text to image generation using OpenAI's CLIP and Siren (Implicit neural representation network). Technique was originally created by https://twitter.com/advadnoun [colab-badge]: <https://colab.research.google.com/assets/colab-badge.svg>