Projects

### Overview: LlamaForge is an open-source control panel designed for managing and interacting with llama.cpp models. It provides real-time GPU compatibility checks, allows you to load models directly from a browser tab, and supports multiple models simultaneously. The project is built on Python and is compatible with Windows, Linux, and macOS. ### Key Features: - **Real-time GPU Compatibility Check**: Ensures that models fit your GPU before loading. - **One-Click Updates**: Automatically updates llama.cpp to the latest version. - **Multiple Models at Once**: Supports loading multiple models simultaneously. - **Embers (Local Agents)**: Checks quotes from models for accuracy. ### Tech Stack: - Python ### Why It Matters: LlamaForge simplifies the process of managing and interacting with llama.cpp models, ensuring compatibility and real-time updates. It provides a user-friendly interface for developers and enthusiasts to explore and use the latest model architectures. ### Repository: Repository: `dadwritestech/LlamaForge` GitHub Stars: 117 GitHub Description: Open-source control panel for the real llama.cpp ### Installation: - **Windows**: Run `irm https://raw.githubusercontent.com/dadwritestech/LlamaForge/master/install.ps1 | iex` - **Linux / macOS**: Run `curl -fsSL https://raw.githubusercontent.com/dadwritestech/LlamaForge/master/install.sh | sh` ### Early Preview: - **Windows + NVIDIA**: Most tested path. - **Linux / macOS**: Passes CI but has minimal real-hardware use. ### Screenshots:

### Overview: This project provides a fine-tuning toolkit for Chatterbox TTS and Chatterbox Turbo models, supporting 23 languages with smart vocabulary extension. It includes offline preprocessing, automatic VAD trimming, and voice cloning capabilities. The toolkit supports both full fine-tuning and LoRA training, with a recommendation for LoRA for better performance on smaller datasets. ### Key features: - Offline preprocessing - Automatic VAD trimming - Voice cloning capabilities - Support for 23 languages - Smart vocabulary extension - Full fine-tuning - LoRA training - Stable results with LoRA ### Tech stack: Python ### Why it matters: This toolkit is essential for developers looking to fine-tune Chatterbox models for new languages and voices, ensuring high-quality speech synthesis with minimal resources and faster training times. ### Repository: gokhaneraslan/chatterbox-finetuning ### GitHub stars: 120 ### GitHub description: Fine-tuning toolkit for Chatterbox TTS & Chatterbox TURBO models. Supports 23 languages with smart vocabulary extension. Features offline preprocessing, automatic VAD trimming, and voice cloning capabilities. Train custom TTS models with your own dataset in LJSpeech and file-based format.

### Overview: This repository is a collection of open-source generative AI projects that developers can contribute to, aiming to build a solid profile and land interviews. All repositories are actively being developed and maintained. ### Key Features: - **LLM Infra and Optimization**: Frameworks, runtimes, and systems for training, inference, and scaling. - [huggingface/transformers](https://github.com/huggingface/transformers) - [pytorch/pytorch](https://github.com/pytorch/pytorch) - [Lightning-AI/pytorch-lightning](https://github.com/Lightning-AI/pytorch-lightning) - [ggerganov/llama.cpp](https://github.com/ggerganov/llama.cpp) - [lm-sys/FastChat](https://github.com/lm-sys/FastChat) - [unslothai/unsloth](https://github.com/unslothai/unsloth) - [huggingface/trl](https://github.com/huggingface/trl) - [microsoft/DeepSpeed](https://github.com/microsoft/DeepSpeed) - [mlflow/mlflow](https://github.com/mlflow/mlflow) - **Agents**: Multi-agent frameworks and components. - [camel-ai/camel](https://github.com/camel-ai/camel) - [mem0ai/mem0](https://github.com/mem0ai/mem0) - [langchain-ai/langchain](https

Overview: DART-Math is an official code for the NeurIPS 2024 conference, focusing on difficulty-aware rejection tuning for mathematical problem-solving. Key features: - Difficulty-aware rejection tuning for mathematical problem-solving. - Uses 590K training examples for both DART-Math (Uniform) and DART-Math (Prop2Diff). - Compares favorably to MetaMath, MMIQC, and vanilla rejection tuning on mathematical benchmarks. - Proposes two strategies to balance and bias towards difficult queries. Tech stack: Jupyter Notebook Why it matters: DART-Math demonstrates advanced difficulty-aware tuning methods for mathematical problem-solving, showcasing the effectiveness of rejection tuning in improving accuracy across various benchmarks.

### Overview: SEO is losing relevance as people increasingly rely on generative AI engines for research. This list of tools helps companies optimize their visibility in AI responses to stay ahead of the curve. ### Key Features: - **[AiCarma](https://aicarma.com)**: Daily Visibility Score and weekly email reports. - **[AI Monitor](https://getaimonitor.com/)**: Tracks brand mentions in AI answers. - **[AI Rank Tracker](https://airank.dejan.ai)**: Shows how LLMs connect to your brand. - **[Am I on AI?](https://amionai.com)**: Simple checker for ChatGPT recommendations. - **[AppearOnAI](https://appearonai.com)**: Actionable audit and playbook for boosting site visibility. - **[AthenaHQ](https://athenahq.ai)**: Free visibility report and GEO playbook. - **[Attensira](https://attensira.com)**: AI search visibility tracking and optimization platform. - **[AutoGEO](https://github.com/cxcscmu/AutoGEO)**: Automated framework for improving generative search visibility. - **[Avenue Z — AIO](https://avenuez.com/services/ai-optimization/)**: Consultancy service for AI visibility. - **[Bluefish AI](https://bluefishai.com)**: Unified visibility and brand-safety console. - **[Botify](https://www.botify.com/)**: AI search optimization platform. - **[BrandBeacon

### Overview: Photo Abstract Editorial is an Agent Skill that transforms photographs into editorial artworks with adaptive layouts, controlled abstraction, and a strict fidelity composition path. This skill ensures the source photograph remains factual while creating a restrained and controlled abstract representation. ### Key Features: - Adaptive Layouts: The skill dynamically adjusts the layout of the photograph to fit the abstracted version. - Controlled Abstraction: The artwork maintains the essence of the original photograph through controlled abstraction. - Strict Fidelity Composition Path: The skill adheres to a strict fidelity composition path, ensuring the final output is as close as possible to the original source. ### Tech Stack: Python ### Why It Matters: Photo Abstract Editorial is essential for creating editorial artworks that maintain the factual integrity of the original photograph while transforming it into a visually compelling and abstract representation. This skill is particularly useful in fields such as journalism, design, and fine art where the focus is on creating impactful, yet accurate, visual narratives.

### Overview: GeoSynth is a PyTorch implementation of "GeoSynth: Contextually-Aware High-Resolution Satellite Image Synthesis," designed to generate satellite images with global style and image-driven layout control. The project aims to provide a comprehensive suite of models for synthesizing satellite images with high accuracy and flexibility. ### Key Features: - **Contextual-Aware Synthesis**: GeoSynth models can synthesize satellite images based on the global context of a region. - **Image-Driven Layout Control**: Users can control the layout of the synthesized images using image features. - **Global Style**: The models can apply global style to the synthesized images, ensuring consistency across the entire region. ### Tech Stack: - **Python**: The primary programming language used for the implementation. - **PyTorch**: The deep learning framework used for the model training and inference. ### Why It Matters: GeoSynth offers a powerful tool for researchers and developers looking to create high-resolution satellite images with detailed context and style. The ability to synthesize images based on global context and image-driven layout control makes it particularly useful in applications such as environmental monitoring, urban planning, and more. ### Repository: - **Repository**: `mvrl/GeoSynth` - **GitHub Stars**: 120 - **GitHub Description**: An official implementation of GeoSynth, a suite of models for synthesizing satellite images with global style and image-driven layout control. The repository includes models available on Hugging Face Spaces and a Gradio demo for quick inference. ### Model Zoo: - **ControlNet Models

Overview: DSC is a PyTorch-compatible tensor library and inference framework for machine learning models, featuring a C-compatible low-level API, built-in neural networks support, multiple backends, minimal external dependencies, and no runtime allocations. Key features: - Intuitive API - Built-in neural networks support - Multiple backends (CPU, CUDA, other backends being worked on) - Minimal external dependencies - No runtime allocations Tech stack: C++ Why it matters: DSC simplifies machine learning development by providing a high-performance, easy-to-use tensor library that integrates seamlessly with PyTorch, making it an excellent choice for developers looking to accelerate their machine learning projects.

Overview: This project is an implementation of a training-free approach to compositional text-to-image synthesis using structured diffusion guidance, inspired by the paper "Training-Free Structured Diffusion Guidance for Compositional Text-to-Image Synthesis" presented at ICLR 2023. Key features: - Incorporates language structured for compositional text-to-image synthesis - Efficiently incorporates language structures into the cross-attention layers of a state-of-the-art model like Stable Diffusion - Addresses three challenging phenomena: attribute leakage, interchanged attributes, and missing objects - Achieves better compositional skills in both qualitative and quantitative results Tech stack: Python Why it matters: This project provides an efficient and effective solution for compositional text-to-image synthesis, addressing common issues in the generation process and improving the quality of generated images.

Overview: PrimDiffusion is a method that generates 3D human models by denoising a set of volumetric primitives, enabling explicit pose, view, and shape control with real-time rendering in high resolution. Key features: - Generates 3D human models - Uses volumetric primitives for denoising - Supports real-time rendering with high resolution - Enables explicit control over pose, view, and shape Tech stack: Python Why it matters: PrimDiffusion is significant because it offers a novel approach to 3D human generation that combines volumetric primitives with real-time rendering, making it a valuable tool for applications requiring detailed and interactive 3D human models. Repository: FrozenBurning/PrimDiffusion GitHub stars: 119 GitHub description: [NeurIPS 2023] PrimDiffusion: Volumetric Primitives Diffusion for 3D Human Generation

Overview: This repository contains code for the experiments of the NeurIPS 2024 paper "Optimal Flow Matching: Learning Straight Trajectories in Just One Step" by Nikita Kornilov, Petr Mokrov, Alexander Gasnikov, and Alexander Korotin. Key features: - Novel Optimal Flow Matching (OFM) method for recovering straight trajectories between source and target distributions in just one Flow Matching step. - Solves quadratic Optimal Transport problem between source and target distributions. Tech stack: Jupyter Notebook Why it matters: This method offers a straightforward approach to learning trajectories that are straight and optimal, simplifying the process of recovering trajectories from source and target distributions.

### Overview: GVGEN is a novel diffusion-based framework designed to generate 3D Gaussian representations from text input. It introduces two innovative techniques: Structured Volumetric Representation and a Coarse-to-Fine Generation Pipeline. The project was accepted by ECCV 2024 and has been recognized for its efficiency and detail in text-to-3D generation. ### Key Features: - **Structured Volumetric Representation**: Arranges disorganized 3D Gaussian points into a structured form, capturing intricate texture details. - **Coarse-to-Fine Generation Pipeline**: Simplifies the generation of GaussianVolume and empowers the model to generate detailed 3D geometry. ### Tech Stack: Python ### Why It Matters: GVGEN offers a novel approach to text-to-3D generation, providing efficient and detailed 3D representations from textual inputs. Its structured representation and coarse-to-fine pipeline make it particularly effective for applications requiring high precision and detail. ### GitHub Stars: 119 ### Repository: SOTAMak1r/GVGEN ### Primary Language: Python ### GitHub Description: [ECCV 2024] GVGEN: Text-to-3D Generation with Volumetric Representation ### Project Page: [](https://sotamak1r.github.io/gvgen/) ### Hugging Face Spaces: [![Hugging Face Spaces](https://img.shields.io/badge/%F0%9F%A4%

Overview: This is a command line virtual assistant designed specifically for competitive programming, offering features such as running Python or C++ files, parsing problem sets, testing solutions, and more. Key features: - Runs Python or C++ files - Parses problem sets from online judges - Generates files with templates - Tests solutions against test cases - Adds test cases - Runs brute force tests - Automatically generates test case generator - Allows login and submit to online judges - Submits code - Parses contests - CF tool mode Tech stack: Python Why it matters: This virtual assistant helps users improve their programming skills and perform well in programming contests. Repository: Saurav-Paul/AI-virtual-assistant-python Primary language: Python GitHub stars: 119 GitHub description: Command line virtual assistant for competitive programming README excerpt: The repository provides a terminal-based virtual assistant for competitive programming, offering various features such as running Python or C++ files, parsing problem sets, testing solutions, and more. It supports login and submission to online judges, parses contests, and more. The assistant also includes features like speaking capability, voice command recognition, AI answering, and more. The README includes examples and screenshots of the assistant's functionality.

Overview: SmooSense is a web-based application designed for interactive exploration and analysis of large-scale multi-modal tabular data. Key features: - Natively visualizes multimodal data (images, videos, JSON, BBOX, image masks, 3D assets, etc) - Automatically looks at distribution, allowing for statistical analysis and random sample exploration - Offers graphical and interactive slicing and dicing of datasets - Supports large-scale data handling with up to 100 million rows on a laptop - Integrates seamlessly with various data formats, including Parquet, CSV, and JSONL - Provides a free and open-source solution for use on a laptop, optimizing performance when deployed Tech stack: TypeScript Why it matters: SmooSense simplifies the process of exploring and analyzing complex tabular data, making it accessible and efficient for both small and large datasets alike.

Overview: This repository is dedicated to sharing open-source implementations of fundamental generative models in artificial general intelligence (AGI), including Variational Autoencoder (VAE), Generative Adversarial Network (GAN), Autoregressive models, Normalizing Flows, Boltzmann Machines, Hopfield Networks, and Diffusion Models. Key features: - Variational Autoencoder (VAE) - Generative Adversarial Network (GAN) - Autoregressive models - Normalizing Flows - Boltzmann Machines - Hopfield Networks - Diffusion Model Tech stack: Python Why it matters: This repository serves as a comprehensive resource for researchers and practitioners interested in exploring and experimenting with generative models in artificial intelligence, providing open-source implementations of fundamental models in the field of generative models.

### Overview: Inaudible Voice Commands (DolphinAttack) is a novel attack that modulates voice commands on ultrasonic carriers to achieve complete inaudibility. By leveraging the nonlinearity of microphone circuits, the modulated low frequency audio commands can be successfully demodulated, recovered, and interpreted by speech recognition systems. This attack has been validated on popular speech recognition systems, including Siri, Google Now, Samsung S Voice, Huawei, HiVoice, Cortana, and Alexa. ### Key Features: - **Completely Inaudible**: The attack modulates voice commands on ultrasonic carriers to achieve complete inaudibility. - **Hardware and Software Defense Solutions**: Proposed hardware and software defense solutions to detect and mitigate the attack. - **Proof-of-Concept Attacks**: Successfully activated Siri to initiate a FaceTime call on iPhone, Google Now to switch the phone to the airplane mode, and even manipulated the navigation system in an Audi automobile. ### Tech Stack: - **Hardware**: Nonlinear microphone circuits. - **Software**: Support Vector Machine (SVM) for audio classification. ### Why It Matters: Inaudible voice commands can significantly enhance the security and privacy of voice-controlled systems, making them more resilient to attacks. This technology opens up new possibilities for future voice-controlled devices and systems, ensuring they remain secure even when voice commands are not audible.

### Overview: Alpaca-GlassOff is an image-acceptable Chat AI model that can be run on a laptop. It is based on Stanford Alpaca and Alpaca-Lora, and can be used for various image recognition and generation tasks. ### Key features: - Image recognition: Recognizes objects and scenes in images. - Image generation: Generates images based on textual prompts. - Chat functionality: Engages in text-based conversations. ### Tech stack: Jupyter Notebook ### Why it matters: Alpaca-GlassOff provides a simple and accessible way to integrate image recognition and image generation capabilities into your projects, making it useful for applications like augmented reality, content creation, and more.

Overview: SenseFlow is a text-to-image distillation model that addresses scalability issues in large-scale flow-based models like SD 3.5 and FLUX. It introduces implicit distribution alignment (IDA) and intra-segment guidance (ISG) to improve convergence and performance. Key features: - Implicit Distribution Alignment (IDA) - Intra-segment Guidance (ISG) - Improved convergence on large-scale models - Superior performance in distillation for diffusion-based and flow-matching models Tech stack: Python Why it matters: SenseFlow enhances the scalability and performance of text-to-image distillation models, making them more practical and efficient for large-scale applications. Repository: XingtongGe/SenseFlow Primary language: Python GitHub stars: 119 GitHub description: 🚀 [ICLR 2026] SenseFlow: Scaling Distribution Matching for Flow-based Text-to-Image Distillation

Overview: Listen to any audio stream on your machine and print out the transcribed or translated audio, based on OpenAI's Whisper project. Key features: - Transcribe or translate audio from a livestream or URL - Requires installation of ffmpeg and CUDA - Supports multiple devices and channels Tech stack: Python Why it matters: This project allows developers to easily transcribe or translate audio from any live stream or URL, making it useful for live streaming and transcription needs.

Overview: SinDDM is an official PyTorch implementation of the paper "SinDDM: A Single Image Denoising Diffusion Model," which presents a novel approach to generating random samples from a single image using diffusion models. Key features: - Trainable from a single natural image - Generates random samples from the given image - Can be used for image manipulation guided by text Tech stack: PyTorch Why it matters: SinDDM offers a unique method for generating random samples from a single image, which can be particularly useful for applications requiring image manipulation guided by text. The ability to train from a single image and generate samples from it makes it a valuable tool in the field of image processing and generative modeling. Repository: fallenshock/SinDDM GitHub stars: 119 GitHub description: Official pytorch implementation of the paper: "SinDDM: A Single Image Denoising Diffusion Model" README excerpt: [](https://github.com/topics/single-image-generation) [](https://www.python.org/downloads/release/python-38/) ![PyTorch](https://img.shields.io/badge/torch-1.13-red?PyTorch