Projects

### Overview: This Streamlit app is designed to generate high-quality Q&A training datasets from text and PDF files, leveraging various AI models for fine-tuning. It supports multiple models like Gemini, Claude, and OpenAI, allowing for the creation of question-answer pairs in various formats compatible with different models. ### Key features: - **Multiple AI Models**: Choose from Gemini, Claude, or OpenAI for dataset generation. - **File Upload**: Support for both text files (.txt) and PDF files (.pdf). - **Smart Chunking**: Split content by word count instead of manual delimiters. - **Customizable Generation**: Control the number of questions per chunk and conversation exchanges. - **Multiple Model Formats**: Support for Gemma, Llama, ChatML, Alpaca, ShareGPT, and Generic formats. - **Custom Prompts**: Write your own prompts for dataset generation. - **Real-time Progress**: Live progress tracking during generation. - **Instant Download**: Download generated datasets in JSONL format. ### Tech stack: Python ### Why it matters: This tool streamlines the process of creating high-quality training datasets for language models, making it easier for developers to fine-tune their models effectively and efficiently.

Overview: This repository is a comprehensive guide for developers building AI-driven applications and products. It provides practical entry points across agents, speech, vision, and model APIs, focusing on modern generative AI technologies. Key features: - AI Agents - STT Models - TTS Models - Text-to-Image - Talking Head - Model APIs Tech stack: Python Why it matters: This list is designed for developers who want to build real-world applications using modern generative AI, offering practical, actionable information rather than academic or generic lists. GitHub stars: 41 Repository: Mrkomiljon/awesome-generative-ai

Overview: An easy-to-use reverse-engineered Python API wrapper for Lexica.art, providing access to image generation capabilities. Key features: - Reverse-engineered Python API for Lexica.art - Supports querying images based on a given query - Generates images using the Lexica API - Stores and retrieves cookies for authenticated access - Offers additional configurations such as negative prompt, guidance scale, and portrait settings Tech stack: Python Why it matters: This API simplifies the process of generating images from text descriptions, making it accessible and user-friendly for developers and artists alike. Repository: NotCookey/Lexica.art Primary language: Python GitHub stars: 41

Overview: HOSCY is a comprehensive tool designed to simplify communication and OSC usage, offering features such as OSC, media control, translation, and speech recognition. Key features: - **OSC Support**: Configurable routing, sending, and receiving of OSC data. - **Media Control**: Voice-based "Now Playing" display. - **Translation**: Utilizes external APIs for text-to-speech and speech recognition. - **Speech Recognition**: Supports Windows Speech Recognition, local AI, and Azure Cognitive Services. Tech stack: C# Why it matters: HOSCY provides a user-friendly interface for OSC communication, making it easier to interact with devices and services. Repository: PaciStardust/HOSCY GitHub stars: 41 GitHub description: Companion for OSC and Communication README excerpt: # REWORK IN PROGRESS For the following reasons this project is currently being rewritten from the ground up: - The original is reliant on WPF and other Windows-Only components. - The project started development with 1 year of experience, leading to an unmaintainable mess. - As a personal challenge to try out things. **If you wish to view the code of the current release version, refer to any tag lower than v1** ## Current Progress ### Initial Release V1 - ❌ **Initial Startup** - ✅ Dependency injection - ✅ Logging - ✅ Config Loading - ✅ Displaying startup errors (Currently windows only) - ✅ Loading splash screen - ❌ Version checking & updating -

### Overview: PaintMind is a fast and controllable text-to-image model designed for developers. It leverages PyTorch and supports both training and inference using xformers and accelerate acceleration. ### Key Features: - **xformers** support for both training and inference. - **accelerate** support for mixed precision and multi-gpu training. ### Tech Stack: - Python ### Why It Matters: PaintMind is designed to be fast and controllable, making it suitable for developers looking to quickly generate high-quality images from text descriptions. ### Installation: To install the latest version, use: ```bash pip install git+https://github.com/Qiyuan-Ge/PaintMind.git ``` Alternatively, for a more stable installation, ensure you have PyTorch 2.0.0 on Windows or PyTorch 1.12.1, 1.13.1, or 2.0.0 on Linux. ### To Do: - Test end-to-end, including stage1 vqtokenizer and stage2 text-to-image. - Final test and release the stable version V1. ### Repository: - **Repository**: Qiyuan-Ge/PaintMind - **Stars**: 41 - **Description**: Fast and controllable text-to-image model. ### Features: - **xformers** support for acceleration. - **accelerate** support for mixed precision and multi-gpu training. ### To Do: - Test end-to-end, including stage1 vqtokenizer and stage2 text-to-image. - Final

Overview: Sinapsis is a modular and universal AI platform that allows developers to build, experiment, and deploy AI workflows seamlessly. Key features: - Modular and scalable Agents based on composite and reusable Templates - Supports various AI tools including computer vision, NLP, GenAI, speech processing, and time-series analysis - Empowers developers to create end-to-end workflows and unlock new possibilities in AI-driven innovation Tech stack: Python Why it matters: Sinapsis simplifies the integration and management of AI tools, enabling developers to focus on their core AI projects without the complexity of managing multiple libraries and frameworks. Repository: Sinapsis-AI/sinapsis GitHub stars: 41 GitHub description: Modular and Universal AI

### Overview: FLUX.1 Kontext Pro API (`flux-kontext-pro`) is a comprehensive API designed for image editing and text-to-image tasks, offering a single endpoint that supports OpenAI compatibility without any monthly subscription. This API allows developers to quickly integrate image editing capabilities into their projects at a cost of $0.032 per image. ### Key Features: - **Single Endpoint**: A unified API endpoint that supports multiple image, video, and language models. - **OpenAI Compatibility**: Compatible with OpenAI's API, making it suitable for applications requiring text-to-image functionality. - **Pricing**: Offers a flat rate of $0.032 per image, with no need for a monthly subscription. - **Parameter Flexibility**: Supports various parameters such as `model`, `prompt`, `size`, `resolution`, `n`, and `image_urls`. ### Tech Stack: - **Python**: The primary programming language used to develop and integrate the API. ### Why It Matters: - **Cost Efficiency**: Offers a cost-effective solution for developers looking to integrate image editing capabilities without the need for a monthly subscription. - **Ease of Integration**: The API is designed to be simple and straightforward to use, making it easy for developers to integrate into their projects. - **Versatility**: Supports multiple image, video, and language models, providing a versatile solution for various applications. ### Repository: - **Repository**: `a41fx2qbyb41bb/flux-kontext-pro-api-pt` - **GitHub Stars**: 33 - **

### Overview: YoloGen is a powerful tool that simplifies the training process for YOLO and Vision Language Models (VLMs) by automating the generation of VLM training data from YOLO labels. This eliminates the need for manual labeling, making the process faster and more efficient. ### Key Features: - **YOLO + VLM Training**: Train YOLO and fine-tune a VLM using a single command. - **Auto-generated VLM Training Data**: Generate VLM training data from YOLO labels, including hard negatives mined directly from existing images. - **Flexible QA Modes**: Supports two modes: Descriptive and Verification, catering to different use cases. - **Hard Negative Mining**: Automatically generates hard negatives from GT bboxes without requiring a YOLO detector. - **Multi-layer Safeguards**: Ensures robustness and accuracy in VLM training by multiple checks. ### Tech Stack: YoloGen leverages Python for its implementation, utilizing libraries such as PyTorch and Transformers for training and fine-tuning. ### Why It Matters: YoloGen streamlines the training process for YOLO and VLMs, making it faster and more efficient. By automating the generation of VLM training data, it eliminates the need for manual labeling, reducing time and effort. The inclusion of hard negative mining ensures that the VLM is trained to correctly identify true positives and reject false positives, improving overall model accuracy and robustness. ### GitHub Stars: 41 ### Repository: ahmetkumass/yolo-gen

Overview: IDEA-Bench is an official repository designed to benchmark image generation models, focusing on professional-grade image design tasks. It includes 100 tasks and 275 cases, covering Text to Image, Image to Image, Multi-image to Image, Text to Multi-image, and Multi-image to Multi-image tasks. Key features: - Comprehensive coverage of professional image generation tasks - Robust evaluation framework with 1,650 binary scoring items - MLLM-assisted assessment for enhanced image understanding - Representative subset of 18 tasks with enhanced criteria Tech stack: Python Why it matters: IDEA-Bench provides a standardized benchmark for evaluating image generation models, addressing the gap between current models and professional design requirements, thereby advancing the field towards more realistic and contextually relevant results. GitHub stars: 41 Repository: ali-vilab/IDEA-Bench

### Overview: This repository contains papers on adversarial examples and defense techniques in the audio domain. It serves as a resource for researchers and developers interested in understanding and implementing adversarial attacks and defenses in speech recognition systems. ### Key features: - **Adversarial Attacks:** - **Audio Adversarial Examples: Targeted Attacks on Speech-to-Text** (2018) - Carlini & Wagner - **Robust Audio Adversarial Example for a Physical Attack** (2019) - Yakura & Sakuma - **Adversarial Black-Box Attacks on Automatic Speech Recognition Systems Using Multi-Objective Evolutionary Optimization** (2019) - Khare, Aralikatte, Mani - **Imperceptible, Robust, and Targeted Adversarial Examples for Automatic Speech Recognition** (2019) - Qin, Carlini, Cottrell, Goodfellow, Raffel - **Universal Adversarial Perturbations for Speech Recognition Systems** (2019) - Neekhara, Hussain, Pandey, Dubnov, McAuley, Koushanfar - **Enabling Fast and Universal Audio Adversarial Attack Using Generative Model** (2020) - Xie, Li, Shi, Liu, Chen, Yuanl - **Defending Your Voice: Adversarial Attack on Voice Conversion** (2020) - Huang, Lin, Lee, Lee -

Overview: Anirudhmkdev/Txt2Img is an open-source AI image generator featuring multiple AI models, an AI prompt enhancer, local history saving, and premium models without API keys. Key features: - Multiple AI Models: Generate art using DALL·E 3, Gemini 2.5 Flash, Grok 2 Image, and GPT Image variants. - AI Prompt Enhancer: Stuck on ideas? Click "Enhance" to let AI rewrite your basic ideas into highly detailed, optimal prompts. - Privacy First: All generated images and your gallery history are saved locally right in your browser. - Negative Prompts: Fine-tune your results by specifying exactly what you *don't* want in your image. - Responsive & Beautiful UI: Fully responsive design that works on mobile and desktop, featuring a seamless Dark/Light mode toggle. - Export Options: Download your creations directly or copy them to your clipboard with a single click. Tech stack: Vanilla HTML5, CSS3, JavaScript, and client-side code using Puter.js. Why it matters: This project is an open-source AI image generator that provides stunning visuals using state-of-the-art models, all running completely free without API keys, making it perfect for developers looking to create their own AI-driven image generation tools.

Overview: This is a full-stack LLM application built using OpenAI, Flask, React, and Pinecone. Key features: - URL Q&A feature - Retrieval Augmented Generation (RAG) with OpenAI - Vector database indexing with Pinecone Tech stack: Python Why it matters: This application demonstrates the integration of various technologies in a full-stack development context, showcasing how different components can be used together to build a powerful language model application.

### Overview: BioTrove is a large curated image dataset designed for AI applications in biodiversity research. It provides comprehensive multimodal data for training and validating models. ### Key Features: - **Large Dataset:** Contains over 1 million images across various taxa. - **Multimodal Representation:** Includes both images and text labels for enhanced understanding. - **Easy Access:** Available via Jupyter Notebook and Hugging Face datasets. - **Model Training:** Supports training of three models using BioTrove-40M dataset. ### Tech Stack: Jupyter Notebook ### Why it Matters: BioTrove offers a rich, diverse dataset that can significantly advance research in biodiversity and artificial intelligence. Its multimodal nature allows for more accurate and comprehensive analysis, making it invaluable for various applications in the field.

### Overview: Scene Graph Generation is a crucial task in computer vision that involves representing images as graphs where nodes represent objects and edges represent relationships between them. This project introduces **Hierarchical Relationships and Commonsense Validation (HIERCOM)**, which leverages commonsense knowledge from large language models (LLMs) to enhance the accuracy of scene graph generation. By replacing the flat relation classification head with a Bayesian model, HIERCOM significantly improves performance, making it a valuable tool for researchers and practitioners in the field. ### Key Features: - **Hierarchical Relationships**: The project introduces a hierarchical relationship model that improves the accuracy of scene graph generation by incorporating commonsense knowledge. - **Commonsense Validation**: Commonsense validation ensures that the generated scene graphs are consistent with common sense, enhancing the reliability of the generated graphs. - **Model-Agnostic**: HIERCOM can be easily integrated into existing state-of-the-art scene graph generation models, allowing researchers to achieve better performance without significant changes to their existing codebases. ### Tech Stack: - **Python**: The project is implemented using Python, leveraging the PyTorch framework for deep learning applications. ### Why It Matters: HIERCOM offers a significant improvement in scene graph generation by incorporating commonsense knowledge, making it a powerful tool for researchers and practitioners in the field of computer vision. By leveraging large language models, HIERCOM ensures that the generated scene graphs are more accurate and consistent with common sense, leading to better performance in various applications.

Overview: This project provides a set of tools and libraries for optimizing and profiling network performance in DGX Spark inference recipes, focusing on RoCE (RDMA over Connext) and NCCL (NCCL2) collective communication. Key features: - One-shot RDMA all-reduce and all-gather for small collectives. - NCCL profile with balanced channel policy for bulk collectives. - Explicit collective policy for backend selection. - Node map validation and discovery tooling. - Fabric probe for fabric qualification before model loading. Tech stack: Python Why it matters: This project offers a comprehensive solution for optimizing network performance in DGX Spark clusters, ensuring efficient communication between GPUs and tensor-parallel ranks without the need for a switch, thereby improving inference recipes' execution times and resource utilization.

Overview: Create a Python Reflex App using OpenAI, Neon Postgres, and a Vector Database, and deploy it to a Virtual Machine. Key features: - Reflex App Development - Integration with Neon Postgres and Vector Database - Deployment to a Virtual Machine Tech stack: Python Why it matters: This project demonstrates the integration of modern AI and database technologies, providing a practical example for developers looking to build scalable and efficient applications.

### Overview: Tree Prompting is an efficient method for task adaptation that leverages tree-based models to improve classification accuracy without the need for fine-tuning. This approach uses pre-trained models to generate a tree of prompts, which are then used to make predictions, resulting in better performance compared to ensemble methods. ### Key Features: - **Tree Prompting Interface**: A scikit-learn-compatible interface for improved prompting. - **Efficient Adaptation**: Learn a tree of prompts to adapt tasks without fine-tuning. - **High Accuracy**: Significantly higher accuracy compared to baseline ensembles. ### Tech Stack: - Python 3.6+ - TreePrompt library ### Why It Matters: Tree Prompting offers a more efficient and effective way to adapt tasks by leveraging pre-trained models to generate a tree of prompts, leading to improved accuracy and reduced computational resources. ### Repository: csinva/tree-prompt - GitHub Stars: 41 - GitHub Description: Tree prompting: easy-to-use scikit-learn interface for improved prompting. - Primary Language: Jupyter Notebook ### Quickstart: ```python from treeprompt.treeprompt import TreePromptClassifier import datasets import numpy as np from sklearn.tree import plot_tree import matplotlib.pyplot as plt # Set up data rng = np.random.default_rng(seed=42) dset_train = datasets.load_dataset('rotten_tomatoes')['train'] dset_train = dset_train.select(rng.choice( len(dset_train), size=100, replace=False)) dset_val

### Overview: Style transfer allows you to copy the style of one image to another, enabling creative and artistic transformations without the need for manual labor. ### Key features: - **DreamshaperXL_lightningDPMSDE.safetensors**: A high-quality image generation model. - **CLIP-ViT-H-14-laion2B-s32B-b79K.safetensors**: A powerful image encoder model. - **ip-adapter-plus_sdxl_vit-h.safetensors**: A controlnet model for depth enhancement. ### Tech stack: Python ### Why it matters: Style transfer is a powerful tool for artists, designers, and anyone looking to add a unique touch to their work by replicating the style of a favorite image.

### Overview: This collection features 252 curated FLUX 2 Max prompts across 9 categories, providing preview images to help you choose the best option before running the prompt. ### Key Features: - **Portrait & Avatar**: 28 prompts - **Product & E-commerce**: 28 prompts - **Poster & Typography**: 28 prompts - **Infographic & Diagram**: 28 prompts - **Anime & Illustration**: 28 prompts - **3D & Material**: 28 prompts - **Food & Lifestyle**: 28 prompts - **Architecture & Interior**: 28 prompts - **Nature & Landscape**: 28 prompts ### Tech Stack: None specified ### Why It Matters: This collection offers a curated selection of FLUX 2 Max prompts, grouped by use case, with preview images to help you choose the best option before running the prompt.

### Overview: This repository contains 252 curated Nano Banana Pro prompts across 9 categories, grouped by use case, with preview images for easy reference. ### Key Features: - **Portrait & Avatar** - No. 1: Editorial Studio Portrait - No. 2: Candid Street Portrait - **Product & E-commerce** - No. 3: Product Detail - No. 4: Product Comparison - **Poster & Typography** - No. 5: Poster Design - No. 6: Typography Prompt - **Infographic & Diagram** - No. 7: Data Visualization - No. 8: Flowchart Prompt - **Anime & Illustration** - No. 9: Anime Character - No. 10: Illustration Prompt - **3D & Material** - No. 11: 3D Model - No. 12: Material Design - **Food & Lifestyle** - No. 13: Food Recipe - No. 14: Lifestyle Prompt - **Architecture & Interior** - No. 15: Interior Design - No. 16: Architecture Prompt - **Nature & Landscape** - No. 17: Landscape Photography - No. 18: Nature Prompt - **All Prompts** ### Tech Stack: The repository uses Markdown for documentation and images for previews. ### Why It Matters: This collection provides a curated