Projects

Overview: Chatito is a tool designed to help developers generate datasets for training chatbot models, natural language processing tasks, and named entity recognition using a simple DSL. It provides an online IDE, DSL specification, AST parser, and a TypeScript npm package for easy integration. Key features: - Online chatito IDE - Chatito DSL specification - DSL AST parser in pegjs format - Generator implemented in TypeScript + npm package Tech stack: TypeScript Why it matters: Chatito simplifies the process of generating datasets for AI chatbots and NLP tasks, making it easier for developers to train and validate their models without needing to manually create complex datasets.

Overview: This repository is a curated list of AI memory projects, categorized by open-source projects, closed-source projects, and companies, as well as by storage types (graph, vector, both). Key features: - Browse examples in the Cognee QuickStart - Contact us at info@topoteretes.com or join us on Discord - Follow us on Twitter Tech stack: Python Why it matters: This list provides a comprehensive overview of AI memory tools, helping developers and AI practitioners choose the best tools for their projects. Repository: topoteretes/awesome-ai-memory GitHub stars: 889 GitHub description: A list of AI memory projects

### Overview VT Code is an open-source Rust terminal coding agent designed for efficient and auditable code management. It allows you to explore, plan, edit, and verify your code with every step auditable. ### Key Features - **Planning**: Plan read-only workflows. - **Safety**: Auditable command policy and sandboxing. - **Self-correct**: HOVER loop on tool mismatch. - **Long runs**: Headless execution with compaction and resumption. - **Integrations**: MCP, Skills, plugins, and editor bridges. - **Models**: Hosted or local providers per task. ### Tech Stack Rust 1.98.1+ (edition 2024) is used for the core functionality. ### Why It Matters VT Code ensures every step of your code changes is auditable, making it a valuable tool for teams looking to improve collaboration and maintainability. ### Repository Repository: [vinhnx/VTCode](https://github.com/vinhnx/VTCode) GitHub Stars: 887 GitHub Description: VT Code is an open-source Rust terminal coding agent.

Overview: AnyGPT is a unified multimodal language model that integrates text, speech, images, and music, offering a comprehensive solution for intermodal conversions and multimodal conversations. Key features: - Unified processing of text, speech, images, and music - Base model weights available for inference - Chat model capable of engaging in free multimodal conversations - AnyInstruct dataset for generative models Tech stack: Python Why it matters: AnyGPT leverages discrete sequence modeling to unify and process various modalities, enabling seamless intermodal conversions and enhancing the capabilities of large language models. Repository: OpenMOSS/AnyGPT GitHub stars: 888 GitHub description: A unified multimodal language model based on discrete sequence modeling

### Overview: OpenWorldLib is a unified codebase designed to facilitate advanced world modeling and simulation. It provides a comprehensive framework for understanding and predicting the complex world through multimodal interactions and long-term memory capabilities. ### Key Features: - **Unified Codebase**: A single, centralized platform for developing and managing world models. - **Multimodal Interaction**: Supports various types of input and output, including text, images, and sensor data. - **Long-Term Memory**: Stores and leverages historical data for informed decision-making. - **Simulation Capabilities**: Enables realistic simulations of various environments and scenarios. ### Tech Stack: - Python ### Why It Matters: OpenWorldLib serves as a foundational tool for researchers and developers looking to create sophisticated world models and simulations. Its unified approach simplifies the development process and enhances the accuracy and efficiency of predictive models, making it invaluable for fields such as artificial intelligence, robotics, and game development.

Overview: 这个项目是一个面向中文输入法的语言模型工程，包括语料、统计模型、神经模型以及它们的评测集。仓库中有两条并行的路线，一个是基于 n-gram 的统计模型，另一个是基于字级神经模型。两者都使用了相同的评测集进行评估，评测集本身比任何一个模型都更值得先看，因为它提供了"某个改动好不好"的数字衡量标准。 Key features: - n-gram 路线：语料清洗、KenLM 统计、Viterbi 解码 - 神经模型路线：语料获取、字级 Transformer 训练、评测 Tech stack: Rust Why it matters: 通过使用评测集，可以量化模型的好坏，而不仅仅是依赖于模型的性能排名。评测集本身比任何一个模型都更值得先看，因为它提供了"某个改动好不好"的数字衡量标准。

Overview: RingNet is an official repository for the paper "Learning to Regress 3D Face Shape and Expression from an Image without 3D Supervision," which introduces a method for generating a 3D mesh of a complete head from a single image. This project is part of a NoW Benchmark dataset and challenges, providing valuable insights into 3D face reconstruction. Key features: - Inference code for generating a 3D mesh from an input face image. - Pre-trained weights and models available for use. - A PyTorch implementation of the decoder FLAME with dynamic contour loading. - A detailed project page and preprint on the NoW Benchmark dataset and 3D face reconstruction challenge. - Ground truth scans for the validation set of the NoW Benchmark dataset. - A pre-trained model for FLAME 2019, which can be used to neutralize pose and expression while using the shape as a template. Tech stack: Python, TensorFlow, PyTorch, OpenDR. Why it matters: This project offers a practical solution for 3D face reconstruction from images, which is crucial in fields such as computer vision, virtual reality, and augmented reality. The ability to generate a complete 3D mesh from a single image without the need for 3D supervision is particularly valuable, as it simplifies the process of creating realistic 3D models from existing images.

### Overview: This repository contains a collection of prompts designed to be used with the ChatGPT model. The prompts are intended to help communicate effectively with the AI, ensuring it generates responses that are clear, informative, and relevant to the context. ### Key features: - **Bing AI Prompts**: Primarily used with Bing AI, these prompts are designed for balanced and creative responses. - **ChatGPT Prompts**: Used exclusively with ChatGPT, these prompts are more precise and detailed. - **No Search Prompts**: Used to prevent Bing from performing searches, which can be useful for certain types of queries. ### Tech stack: Python ### Why it matters: This repository serves as a valuable resource for developers and AI enthusiasts looking to improve their interactions with AI models, specifically ChatGPT and Bing AI. The prompts are designed to enhance the quality and relevance of the AI's responses, making them more useful and informative.

Overview: CascadePSP is a deep learning model designed for high-resolution segmentation refinement, achieving class-agnostic and very high-resolution segmentation. It includes a comprehensive implementation with both training and testing functionalities, and is accompanied by an annotated UHD dataset and pretrained models. Key features: - PyTorch implementation - Training and testing functionalities - Annotated UHD dataset (BIG) - Pretrained models Tech stack: Python Why it matters: CascadePSP offers a significant improvement in segmentation accuracy and efficiency, making it a valuable tool for various applications requiring high-resolution and class-agnostic segmentation. Repository: hkchengrex/CascadePSP Primary language: Python GitHub stars: 885 GitHub description: [CVPR 2020] CascadePSP: Toward Class-Agnostic and Very High-Resolution Segmentation via Global and Local Refinement

Overview: GigaAM is a foundational model for speech recognition tasks, pre-trained on diverse speech data across multiple languages. It serves as the backbone for the GigaAM family, including models for ASR, emotion recognition, and fine-tuning. Key features: - GigaAM-v1: Pretrained on Russian, ASR on 2,000 hours. - GigaAM-v2: Pretrained on HuBERT–CTC, ASR on 2,000 hours. - GigaAM-v3: Pretrained on HuBERT–CTC, ASR on 4,000 hours. - Multilingual models: Pretrained on 2,000,000 hours, ASR on 50,000 hours. Tech stack: Python, HuggingFace, ONNX export, Triton Inference Server. Why it matters: GigaAM offers state-of-the-art performance in speech recognition and emotion recognition, with a wide range of pre-trained models and fine-tuning capabilities.

### Overview: Xberg-io/html-to-markdown is a high-performance and CommonMark compliant HTML to Markdown converter. It is maintained by the Kreuzberg team, a fast, polyglot document intelligence engine with a Rust core, capable of extracting structured data from 98+ document formats using streaming parsers and built-in OCR. ### Key Features: - **High Performance**: Designed for fast processing. - **CommonMark Compliance**: Ensures output is CommonMark compliant. - **Cross-Platform Support**: Works with multiple programming languages and environments. ### Tech Stack: - **Primary Language**: HTML - **GitHub Stars**: 884 - **Maintained by**: Kreuzberg team ### Why It Matters: Xberg-io/html-to-markdown is essential for developers looking to convert HTML content into Markdown, ensuring consistent and readable documentation while maintaining the integrity of the original HTML structure.

Overview: SpeechPy is a comprehensive library designed for speech processing and recognition, offering essential features such as Mel Frequency Cepstral Coefficients (MFCCs) and Filterbank Energies. Key features: - Mel Frequency Cepstral Coefficients (MFCCs) - Filterbank Energies - Log Filterbank Energies Tech stack: Python Why it matters: SpeechPy simplifies speech processing and recognition tasks, making it an indispensable tool for developers working in the field of speech technology. Repository: astorfi/speechpy Primary language: Python GitHub stars: 884 GitHub description: :speech_balloon: SpeechPy - A Library for Speech Processing and Recognition: http://speechpy.readthedocs.io/en/latest/ README excerpt: .. image:: _images/speechpy_logo.gif :target: https://github.com/astorfi/speech_feature_extraction/blob/master/images/speechpy_logo.gif =============================================== `SpeechPy Official Project Documentation`_ =============================================== .. image:: https://pepy.tech/badge/speechpy :target: https://pepy.tech/project/speechpy .. image:: https://img.shields.io/badge/contributions-welcome-brightgreen.svg?style=flat :target: https://github.com/astorfi/speechpy/pulls .. image:: https://coveralls.io/repos/github/astorfi/speechpy/badge.svg?branch=master :target: https://coveralls.io/github/astorfi/speechpy

### Overview: TangoFlux is a cutting-edge text-to-audio generator that leverages the power of diffusion transformers to convert written prompts into high-fidelity audio in just a few seconds. This innovative tool is designed to streamline the process of generating music and sound effects, making it an invaluable asset for creators and developers alike. ### Key Features: - **High-Quality Audio Generation**: TangoFlux can produce 44.1kHz stereo audio up to 30 seconds in just 3 seconds on a single A40 GPU. - **Versatile Text Input**: Supports a wide range of text inputs, including lyrics, descriptions, and phrases. - **Pre-training, Fine-Tuning, and Preference Optimization**: The training pipeline includes pre-training, fine-tuning, and preference optimization using CRPO (Continuous Random Prompt Optimization) for improved quality and diversity. - **CRPO Dataset**: Provides a dataset for generating synthetic audio data, enhancing the realism of the generated audio. ### Tech Stack: - **Stability AI**: Powered by the Hugging Face Stable Diffusion model. - **Jupyter Notebook**: The primary development and demonstration environment. - **Hugging Face Spaces**: A platform for sharing and exploring TangoFlux-related models and datasets. - **Datasets**: The TangoFlux dataset is available for use in various applications. ### Why It Matters: TangoFlux is a game-changer in the field of text-to-audio generation, offering unparalleled speed and quality. Its ability to convert written prompts into high-fidelity audio in seconds makes it an indispensable tool for

### Overview: Tiny Engineer is a small desktop robot designed to bring an AI coding assistant to life. It physically reacts to your coding AI agent's actions, including reading files, writing code, thinking, and finishing tasks. The robot is built with a main control board, servos, OLED display, and a speaker, and it's open source, allowing for customization and community builds. ### Key Features: - **Physical Interaction**: The robot moves, changes its face, and rings a bell as your coding AI agent works. - **Integration with Tools**: Works with Claude Code, Cursor, Antigravity, and any script that can POST. - **Easy Assembly**: Requires a 3D printer and electronics, and can be assembled with a simple setup guide. ### Tech Stack: C++ and ESP32 with Arduino IDE. ### Why It Matters: Tiny Engineer is a practical and engaging way to visualize and interact with AI coding assistants, making the technology more accessible and fun to use.

Overview: PentestCode is an autonomous AI penetration testing agent designed for real-time engagement in offensive security operations. It leverages multi-agent architecture, persistent state tracking, and parallel autonomous operations to automate and enhance the process of ethical hacking. Key features: - Continuous engagement state tracking - Multi-agent architecture for coordinated operations - Support for 20+ language models for various tasks Tech stack: TypeScript Why it matters: PentestCode simplifies the complex process of penetration testing by automating the scanning, enumeration, and exploitation of targets. It provides a structured, memory-based approach that ensures thoroughness and efficiency, making it valuable for both practical engagements and competitive challenges like Capture the Flag (CTFs). Repository: s0ld13rr/pentestcode Primary language: TypeScript GitHub stars: 883 GitHub description: PentestCode - Multi-agent AI penetration testing system with persistent engagement state, strategic coordination, and parallel autonomous operations.

### Overview: REINVENT 4 is an AI molecular design tool that leverages Reinforcement Learning (RL) to generate optimized molecules based on user-defined property profiles. It supports de novo design, scaffold hopping, R-group replacement, linker design, and molecule optimization. The software is designed for modern AI-driven generative molecule design and is supported on Linux, Windows, and macOS. It uses Python 3 and PyTorch for its implementation, with GPU support for better performance and transfer learning capabilities. ### Key Features: - **Reinforcement Learning (RL) Algorithm**: Generates optimized molecules using RL techniques. - **Transfer Learning**: Pre-trains models to generate molecules closer to input molecules. - **GPU Support**: Recommended for better performance, especially for transfer learning and model training. - **Multi-Component Score**: User-defined property profiles for molecule optimization. - **Open Access Paper**: Published in the Journal of Cheminformatics. ### Tech Stack: - Python 3 (>= 3.11) - PyTorch - NVIDIA GPUs - AMD ROCm - Intel XPU - Apple GPUs ### Why It Matters: REINVENT 4 is a powerful tool for modern AI-driven molecular design, offering advanced capabilities for de novo design, scaffold hopping, and optimization. Its integration with reinforcement learning and transfer learning makes it highly versatile and effective for various molecular design tasks.

Overview: This repository is a curated collection of resources, research, and tools aimed at securing AI systems. It is managed by the AISecHub community and powered by InnovGuard. Key features: - Governance & Management Frameworks - Controls & Verification Standards - Implementation Guides & Patterns - Testing, Evaluation & Red Teaming - Agentic Systems - Standards, Governance & Patterns - Threat Modeling - Policy Templates - Toolkits & Self-Assessments - Critical Infrastructure - Tools - Models - Research Working Groups - Attack & Defense Matrices - Checklists - Newsletter - Datasets (Phishing, Cybersecurity Knowledge, Secure Coding & Vulnerability Detection, Deepfake, Jailbreak, Prompt Injection, System Prompts) - Courses & Certifications - Training - Reports and Research (Vendor Reports, Research Papers, Research Feed, Reports) - Books - Communities & Social Groups - Benchmarking - Incident Response (Incident Repositories, Trackers & Monitors, Publicly Disclosed Vulnerabilities, Guides & Playbooks) - Supply Chain Security - Videos & Playlists - Conferences - Foundations: Glossary, SoK/Surveys & Taxonomies - Podcasts - Market Landscape - Blogs (Industry Leaders) - Related Awesome Lists - Common Acronyms Tech stack: Unknown Why it matters: This repository provides a comprehensive set of resources and tools for securing AI systems, covering various aspects of AI security, from governance and management frameworks to threat modeling and incident response.

### Overview: JoyVASA is a diffusion-based method for generating facial dynamics and head motion in audio-driven facial animation. It decouples dynamic facial expressions from static 3D facial representations, enabling longer video generation and identity-independent motion sequences. The model is trained on a hybrid dataset of private Chinese and public English data, supporting multilingual support. ### Key Features: - **Decoupled Facial Representation**: Separates dynamic facial expressions from static 3D facial representations. - **Identity-Independent Motion Generation**: Generates motion sequences directly from audio cues, independent of character identity. - **Longer Video Generation**: Combines any static 3D facial representation with dynamic motion sequences to produce longer videos. - **High-Quality Animation**: Uses a generator trained in the first stage to render high-quality animations. ### Tech Stack: - Python ### Why It Matters: JoyVASA extends the capabilities of audio-driven portrait animation by seamlessly animating animal faces, making it versatile for various applications such as portrait animation and beyond. The model's ability to generate identity-independent motion sequences and support multilingual support makes it a valuable tool in the field of facial animation and portrait generation.

### Overview: Wanda is an efficient pruning approach designed for large language models, specifically LLMs like LLaMA-2 and LLaMA-7B. It introduces a novel method to prune weights based on both weight magnitudes and input activation norms, significantly reducing sparsity without compromising performance. ### Key Features: - **Pruning of LLaMA-2 and LLaMA-7B Models**: Supports pruning of LLaMA-2 and LLaMA-7B models. - **Structured and Unstructured Sparsity**: Offers both structured (2:4) and unstructured (unstructured) sparsity options. - **Ablation Study**: Provides ablation study results to understand the impact of different pruning methods. - **Zero-Shot Evaluation**: Enables zero-shot evaluation for LLaMA models. ### Tech Stack: - Python ### Why It Matters: Wanda offers a more effective and efficient way to prune LLMs compared to traditional magnitude pruning, maintaining higher performance while significantly reducing sparsity. This makes it particularly useful for large-scale models where reducing sparsity is crucial without sacrificing model quality.

### Overview: LLM.swift is a simple and readable library designed for interacting with large language models locally on macOS, iOS, watchOS, tvOS, and visionOS. ### Key features: - **Local Model Interaction**: Allows seamless interaction with large language models without needing to host them on a server. - **Swift Integration**: Built with Swift, ensuring compatibility and ease of use. - **Readability**: Simple and easy-to-understand code structure. - **Model Support**: Compatible with various models from different providers, including Hugging Face and local model bundling. ### Tech stack: Swift ### Why it matters: LLM.swift simplifies the process of interacting with large language models, making it accessible and efficient for developers working across multiple platforms.