Projects

### Overview: This repository contains the implementation of UVR-NMT, a neural machine translation model that utilizes universal visual representation for improved performance. The project was presented at ICLR 2020 and is based on the fairseq library. ### Key Features: - Universal Visual Representation (UVR) for improved translation quality - Neural Machine Translation (NMT) model with fairseq framework - Segmented training sets for multi30k and NMT datasets - Pre-extracted visual features for indexing in image embedding layer - Script for data preprocessing, including BPE code segmentation and dictionary generation - Script for generating a lookup table for NMT model training ### Tech Stack: Python ### Why it Matters: UVR-NMT leverages UVR to enhance translation accuracy, making it a significant advancement in the field of neural machine translation. The model's ability to incorporate visual information into the translation process is particularly noteworthy, as it addresses the limitations of purely text-based models.

Overview: 这是一个用于小红书ASR/FireRedASR的webUI及API项目，旨在提供中文语音转文字功能的封装。 Key features: - 实现中文语音转文字功能 - 支持多种模型，包括小红书本身的识别模型和Qwen2-7B-Instruct模型 - 提供API接口兼容OpenAI格式 Tech stack: Python Why it matters: 该项目为开发者提供了一个便捷的工具，用于将语音转换为文字，适用于需要实时语音转文字应用的场景，如文本记录、会议记录等。 Repository: jianchang512/fireredasr-ui Primary language: Python GitHub stars: 92 GitHub description: 一个中文语音转文字项目，封装自FireRedASR

Overview: This repository provides a research-friendly PyTorch Lightning toolkit for training, fine-tuning, and benchmarking AutoencoderKL for Stable Diffusion, FLUX, and beyond. It supports various VAE models with Vector Quantization (VQ), Residual Vector Quantization (RVQ), Finite Scalar Quantization (FSQ), and Lookup-Free Quantization (LFQ), and offers practical recipes and guidance for training high-quality autoencoders. Key features: - Support for training and fine-tuning VAE models for Stable Diffusion and FLUX - Support for VQ-VAE and VQGAN models with Vector Quantization, Residual Vector Quantization, Finite Scalar Quantization, and Lookup-Free Quantization - Evaluation metrics for reconstruction quality: rFID, PSNR, SSIM, and LPIPS - Easy-to-use codebase that can be modified and extended for custom architectures, losses, and training strategies Tech stack: PyTorch Lightning Why it matters: This repository offers a simple and easy-to-use training script for VAE models, making it accessible for researchers and developers alike. The toolkit is research-friendly and supports various VAE models, making it valuable for projects involving Stable Diffusion, FLUX, and beyond. Repository: lavinal712/AutoencoderKL Primary language: Python GitHub stars: 92

### Overview: 这个项目旨在提供机器智能领域的相关书籍、经典论文和实验代码，涵盖自动机器学习、情感分类、语音识别、声纹识别和语音合成等多个主题。 ### Key features: - **书籍和论文资料**：涵盖自动机器学习、情感分类、语音识别、声纹识别和语音合成等主题的书籍和经典论文。 - **开源工具**：提供Auto-Keras、Auto-sklearn、NNI-v0.8、TransmogrifAI等自动机器学习工具的链接和简要介绍。 - **商业化平台**：包括Google Cloud AutoML、阿里云PAI Studio、百度EasyDL、智立方DarwinML、智铀科技小智、第四范式先知AutoML、探智科技深思平台等商业平台的介绍。 - **实验代码**：提供情感分析、语音识别和声纹识别的实验代码示例。 ### Tech stack: Jupyter Notebook ### Why it matters: 这个项目为开发者提供了丰富的资源和工具，帮助他们在机器智能领域进行学习和研究，涵盖了从理论到实践的多个方面，包括书籍、论文、工具和实验代码，旨在促进技术的交流和学习。

Overview: Whisper Overlay is a Wayland overlay that enables speech-to-text functionality for any application, allowing you to transcribe your spoken words in real-time. This feature is accessible via a global push-to-talk hotkey, making it easy to use without interrupting your workflow. Key features: - Real-time live transcriptions using CUDA and faster-whisper - Server-client architecture for easy model hosting - Native Waybar integration for status display - Supports Wayland composers like sway and hyprland - Utilizes `layer-shell` and `virtual-keyboard-v1` for seamless integration Tech stack: Rust Why it matters: This tool simplifies the process of integrating speech recognition into applications, making it accessible for developers and users alike. The server-client architecture allows for flexibility in model deployment, and the real-time and high-fidelity transcription capabilities make it particularly useful for applications that require immediate feedback. GitHub stars: 91 Repository: oddlama/whisper-overlay

Overview: This repository provides a comprehensive guide for beginners and experts alike in the field of generative AI, offering curated resources and a roadmap for understanding and applying generative AI principles. Key features: - A curated list of articles and projects on generative AI. - Background knowledge on neural networks, including backpropagation and transformer architectures. - Information on large language models (LLMs), diffusion models, and large multimodal models (LMMs). - Discussion on implicit structured parameters and new model architectures. Tech stack: Python, Markdown, and GitHub. Why it matters: This guide is essential for anyone interested in the rapidly evolving field of generative AI, providing a structured approach to learning and staying updated with the latest research and applications. Repository: pittisl/Generative-AI-Tutorial GitHub stars: 92 GitHub description: A subjective learning guide for generative AI research.

Overview: This project provides a comprehensive set of prompts and examples for the Flux 2 API, including 8 tested recipes with outputs, a detailed model id and routes table, pricing information, and a quickstart guide for integrating with the APIMart OpenAI-compatible gateway. Key features: - 8 tested recipes with outputs - Model id and routes table - Pricing information - Quickstart guide for integrating with APIMart Tech stack: Python Why it matters: This project offers developers a reliable and efficient way to integrate with the APIMart OpenAI-compatible gateway, providing a cost-effective solution for generating images based on text or references.

Overview: WavEncoder is a Python library designed for encoding audio signals, applying audio augmentation transforms, and training audio classification models using PyTorch. Key features: - Pretrained models for wav2vec, wav2vec2, SincNet, and RawNet - Baseline models like 1DCNN, LSTM Classifier, and LSTM Attention Classifier - Audio augmentation transforms such as noise, speed change, padcrop, clip, reverberation, time shift, time mask, and frequency mask Tech stack: Python Why it matters: WavEncoder simplifies the process of audio signal processing and classification by providing a comprehensive library with pre-trained models and flexible transforms, making it easier for developers to integrate audio processing into their projects.

Overview: This repository contains an implementation of a framework for uncertainty prediction in Reynolds-averaged Navier-Stokes flows around airfoils using Denoising Diffusion Probabilistic Models (DDPMs). It features datasets for evaluating simulation uncertainty and training code for diffusion models, as well as baselines using Bayesian neural networks and heteroscedastic uncertainty models. Key features: - Generate datasets for evaluating simulation uncertainty - Train diffusion models for uncertainty prediction - Baselines using Bayesian neural networks and heteroscedastic uncertainty models Tech stack: Jupyter Notebook Why it matters: This framework offers a novel approach to uncertainty quantification in turbulence simulations, leveraging DDPMs to accurately estimate the uncertainty of simulations and provide detailed samples from the distribution of solutions.

### Overview: This repository is a comprehensive survey and outlook on the Multiple Modalities for Time Series Analysis (MM4TSA) field, focusing on the benefits of reusing foundation models, multimodal extensions, and cross-modality interactions in time series analysis. The survey aims to provide a detailed review and outlook for researchers and practitioners interested in this emerging research area. ### Key Features: - **Reusing Foundation Models**: Discusses how existing models from other modalities can be used efficiently in TSA. - **Multimodal Extensions**: Explores how additional modalities can enhance TSA performance. - **Cross-Modality Interaction**: Examines how different modalities can interact to improve time series analysis. - **Domain-Specific Applications**: Covers applications in spatial-temporal data, medical time series, and financial time series. - **Gaps and Outlooks**: Identifies the challenges and opportunities in the field, including selecting appropriate modalities, combining heterogeneous data, and generalizing to unseen tasks. ### Tech Stack: The repository is built using Markdown and GitHub Pages. ### Why It Matters: This survey provides a valuable resource for researchers and practitioners looking to advance their work in time series analysis by leveraging multiple modalities. It offers a comprehensive overview of the field, including key papers and resources, making it an essential tool for staying updated on the latest developments. ### GitHub Stars: The repository has been starred by 91 users, indicating its popularity and relevance in the community. ### Contributing: Contributions are welcome, and the repository is maintained by the AdityaLab team. If you

Overview: A framework for Context Engineering using Google Gemini, designed to enhance software development by structuring context, templates, rules, and commands within the Antigravity CLI for more reliable, consistent, and complex code generation. Key features: - Structured context management - Native, non-interactive action blocks - Unified project-specific rulesets - Visual dependency and architecture maps Tech stack: Antigravity CLI Why it matters: By providing a structured approach to context and code generation, this framework aims to eliminate AI code bloat and automate software lifecycle tasks, ensuring more reliable and consistent software development.

Overview: This repository contains a collection of end-to-end large language model projects, showcasing the versatility and capabilities of Generative AI models like Meta LLaMA3, GPT4, and more. Key features: - Multi-PDFs Chat App AI Agent - Image to Speech Generator - End-to-End Generative AI Projects - Meta LLaMA3 GenAI UseCases End-to-End Implementation Guides - Advanced Retrieval-Augmented Generation (RAG) - GPT4o-API-Implementation-GPT4-RAG - LLM Security 2024 - Phi-3 LLM by Microsoft - AI Agents Tech stack: Jupyter Notebook Why it matters: These projects demonstrate the practical applications and advanced capabilities of Generative AI models, showcasing their potential in various industries and use cases. GitHub stars: 91 Repository: GURPREETKAURJETHRA/Generative-AI-LLM-Projects

Overview: This project is a production-grade multimodal RAG system designed for financial document analysis and recall. It integrates chart understanding, hybrid retrieval, numeric guardrails, multi-tenancy, and full observability. Key features: - Chart understanding - Hybrid retrieval - Numeric guardrails - Multi-tenancy - Full observability Tech stack: Python Why it matters: This system addresses the challenges of financial document analysis by providing accurate chart interpretation, robust retrieval, precise number verification, secure PII handling, and resilient fallbacks in case of vendor failures. Repository: Mattral/RAG-Multimodal-Financial-Doc-Analysis-and-Recall Primary language: Python GitHub stars: 91 GitHub description: Production-grade multimodal RAG for financial document intelligence. Chart understanding · hybrid retrieval · numeric guardrails · multi-tenancy · full observability. README excerpt: # RAG Financial Multimodal — Enterprise v2.0 > **Production-grade multimodal RAG for financial document intelligence.** > Chart understanding · hybrid retrieval · numeric guardrails · multi-tenancy · full observability. [](https://github.com/Mattral/RAG-Multimodal-Financial-Doc-Analysis-and-Recall/actions) [![Python 3.11+](https://img.shields.io/badge

### Overview: Texture a whole game level from a gray blockout with image models: plan cameras, paint, project, bake. Real shadows, glTF out. ### Key features: - Plan cameras where a player can stand - Texture level with image models painting it - Real shadows baked into the texture - Realistic lighting with real shadows - Image models run on Fal (https://fal.ai) - Sample levels: Village, Courtyard, Hall - Quick start guide with npm scripts ### Tech stack: TypeScript ### Why it matters: Realistic texture mapping and lighting effects for game levels, enhancing immersion and realism.

Overview: This project introduces Noise-Consistent Siamese-Diffusion for medical image synthesis and segmentation, specifically focusing on medical polyp images. The method is designed to enhance the quality and consistency of medical images through diffusion models. Key features: - Noise-Consistent Siamese-Diffusion - Medical Polyp Images - ISIC2016 and ISIC2018 Images Tech stack: Python Why it matters: This approach aims to improve the accuracy and consistency of medical image synthesis and segmentation, particularly in the context of polyp detection and treatment planning in radiology. Repository: Qiukunpeng/Siamese-Diffusion Primary language: Python GitHub stars: 91 GitHub description: [CVPR 2025] Noise-Consistent Siamese-Diffusion for Medical Image Synthesis and Segmentation

Overview: This project provides a stable diffusion model and its companion models running on a GPU-enabled Kubernetes cluster using Helm. It includes a web interface for easy access and automatic model fetching, making it a quick and efficient solution for developers. Key features: - Automatic model fetching - Works with `gpu-operator`, bundling CUDA libraries - Interactive UI with many features, and more on the way! - GFPGAN for face reconstruction, RealESRGAN for super-sampling - Textual Inversion support - Many more features! Tech stack: Helm charts using `nvidia/cuda` as a base. Why it matters: This project simplifies the process of deploying stable diffusion models on Kubernetes, making it accessible for developers without requiring extensive setup or knowledge of Kubernetes.

### Overview: This repository contains a sample designed to accelerate the AWS Well-Architected Framework Review (WAFR) process by leveraging generative AI. The sample includes features for uploading technical content, creating architecture assessments, and generating recommendations. It also allows for optional configurations such as Amazon Bedrock Guardrails and disabling redundancy for Amazon OpenSearch Serverless. ### Key Features: - Ability to upload technical content (PDF format) - Creation of architecture assessment with solution summary, assessment, best practices, recommendations, and risk analysis - Integration with Amazon Bedrock Guardrails for Responsible AI - Option to disable redundancy for Amazon OpenSearch Serverless collections - Chat functionality for document review ### Tech Stack: Python ### Why It Matters: This sample streamlines the AWS Well-Architected Framework Review process, making it more efficient and accessible to organizations. By leveraging generative AI, it accelerates the review process and provides valuable insights to improve AWS architecture. ### GitHub Stars: 91 ### Implementation Guide: - Ensure access to Amazon Bedrock models (Titan Text Embeddings V2, Claude 3-5 Sonnet) - Install AWS CDK prerequisites and AWS Cloud Development Kit (AWS CDK) v2 - Docker CE is required for the sample ### Pre-requisites: - AWS Well-Architected Framework documents - Amazon Bedrock knowledge base ### Cloning Repository: ``` git clone https://github.com/aws-samples/sample-well-architected-acceleration-with-generative-ai.git ``` Alternatively, download the code as

Overview: This repository provides code samples for quickly building diverse AI applications using Amazon Bedrock's foundation models. Key features: - Image Generation - Text Examples - Chatbot - RAG Example Tech stack: Python Why it matters: This repository offers a practical guide for developers to accelerate project development in image and text generation and beyond, leveraging Amazon Bedrock's foundation models.

Overview: This repository refactors and rewrites TensorFlow Lite examples for Android, focusing on maintaining the same UI designs and inference approaches as the official TensorFlow website. The repository aims to simplify machine learning on Android by providing base facilities for quick understanding of inference and result representation. Key features: - Image Classification - Object Detection - Pose Estimation - Gesture Recognition - Smart Reply - Image Segmentation - Style Transfer - Digit Classifier - Video Classification - Super Resolution - Optical Character Recognition Tech stack: Kotlin Why it matters: This repository aims to make machine learning tasks on Android more accessible and easier to understand, providing a simpler way to run inference and represent results on the user interface. Repository: dailystudio/tensorflow-lite-examples-android Primary language: Kotlin GitHub stars: 91 GitHub description: Examples of Tensorflow Lite on Android [](https://www.youtube.com/watch?v=ctn-t1pg9pA&feature=youtu.be)

Overview: ProGPT is an open-source Python client designed for interacting with ChatGPT web conversations, offering both generative and conversational modes. Key features: - Supports both Generative and Conversation modes. - Provides a simple interface for ChatGPT interactions. - Compatible with the ChatGPT web service, including authentication via OAuth. Tech stack: Python Why it matters: ProGPT simplifies the integration of ChatGPT into various applications, making it easier for developers to leverage the conversational capabilities of the AI model. Repository: diezo/ProGPT Primary language: Python GitHub stars: 90 GitHub description: 🧠 Reverse-Engineered Python Client for ChatGPT Web Conversations