Coding & Development

Awesome-totally-open-chatgpt is a comprehensive list specifically designed to showcase open alternatives to the popular ChatGPT. This resource focuses on instruct finetuned language models that are suitable for chat applications. The list maintains strict criteria, excluding any projects that merely integrate with OpenAI's APIs or utilize language models without proper fine-tuning. Its purpose is to highlight truly open and independently developed solutions for conversational AI.

Transmute AI Lab operates as a research network with a primary focus on artificial intelligence. The lab's core mission is to develop and innovate cutting-edge AI technologies. It aims to achieve significant breakthroughs in the field of AI, with a particular emphasis on experiencing and understanding new realities that can be enabled or enhanced by artificial intelligence. This initiative brings together researchers and innovators to push the boundaries of current AI capabilities.

Chat-UniVi is a tool designed to empower large language models (LLMs) by integrating image and video understanding capabilities. It achieves this through the use of a unified visual representation, allowing LLMs to process and interpret visual data more effectively. This tool is primarily aimed at AI research and development, providing a foundation for building advanced multimodal AI applications. Its availability on GitHub suggests a focus on open-source contributions and community-driven development.

Mystic Turbo Registry is a specialized Docker registry engineered for machine learning applications. Its primary function is to optimize the loading of machine learning models, leading to faster deployment times. By focusing on efficiency, the registry helps to significantly reduce cold starts for deployed models, which is crucial for maintaining responsive AI services. This optimization ultimately streamlines the entire model deployment process, making it more efficient for developers and MLOps teams.

BrushNet is an implementation of the ECCV 2024 paper, "BrushNet: A Plug-and-Play Image Inpainting Model with Decomposed Dual-Branch Diffusion." This tool is specifically designed for advanced image inpainting tasks, leveraging the power of diffusion models to fill in missing or corrupted parts of images seamlessly. It provides a robust framework for researchers and developers working in image generation and processing, enabling them to experiment with and integrate state-of-the-art inpainting capabilities into their projects. The dual-branch diffusion approach allows for more refined and context-aware image reconstruction.

box-convolutions offers a PyTorch implementation of the box convolution layer, as detailed in the research paper "Deep Neural Networks with Box Convolutions." This tool is designed to facilitate experimentation and replication of the research findings related to box convolutions within convolutional neural networks. It serves as a valuable resource for AI researchers and deep learning practitioners who are interested in exploring advanced convolution techniques.

chatgpt-artifacts is an open-source project designed to enhance ChatGPT by incorporating features akin to Claude's Artifacts. This tool allows users to expand the functionalities of their ChatGPT experience, offering new ways to interact and process information. To utilize chatgpt-artifacts, users need to provide an OpenAI API key and perform a local setup, indicating a more hands-on approach for integration. The project aims to bridge the gap between the capabilities of different AI models, offering users a richer and more versatile AI assistant experience.

checkpoint-engine is a middleware solution designed to streamline the process of updating model weights within Large Language Model (LLM) inference engines. Its core functionality focuses on enabling efficient, in-place weight updates, a feature particularly crucial for applications in reinforcement learning. The tool is engineered to support rapid updates of large models, even when distributed across multiple GPUs. It provides a lightweight implementation for effectively managing model checkpoints, ensuring that updates can be performed with minimal overhead.

Conditional_Diffusion_MNIST is an implementation of a conditional diffusion model designed to generate MNIST digits. This model leverages a U-Net architecture and incorporates principles from 'Classifier-Free Diffusion Guidance' to produce digits based on specified class labels. It serves as a minimal script, making it suitable for individuals looking to learn about or experiment with conditional diffusion models in a straightforward manner.

CoOp is a research project dedicated to advancing prompt learning within vision-language models. It provides a comprehensive codebase designed to facilitate the adaptation of existing models, such as CLIP, to various downstream datasets. The project specifically supports and implements techniques like conditional prompt learning and learning to prompt. CoOp is primarily intended for researchers and academics who are actively working on the adaptation and fine-tuning of vision-language models for specific applications or datasets.

ddpo-pytorch provides a PyTorch implementation of the Denoising Diffusion Policy Optimization (DDPO) technique. This tool is specifically designed for finetuning diffusion models, offering support for low-rank adaptation (LoRA) to enhance efficiency. A key feature is its ability to perform GPU-based finetuning of Stable Diffusion models while significantly reducing memory requirements. It is built for developers and researchers working with diffusion models and requires Python 3.10 or newer to operate.

DeepSentinel AI is a security solution designed to safeguard AI workflows. It offers affordable data privacy measures and real-time threat detection capabilities to protect AI models and their associated data. The tool aims to create a secure environment for the development and deployment of AI applications, helping users mitigate risks from potential threats.

deep_q_rl is an open-source project that offers a Theano-based implementation of the deep Q-learning algorithm. It serves as a foundational framework for researchers and developers interested in deep reinforcement learning. The implementation is directly inspired by and based on the original deep Q-learning research papers, ensuring a robust and academically sound approach. It leverages the Lasagne library for the construction and management of neural networks, facilitating the development and experimentation with complex AI models.

Dreyev offers an advanced in-vehicle system designed to improve driver safety by identifying and mitigating risky driving behaviors. Utilizing computer vision and machine learning, the device continuously monitors driver conditions such as head pose and eyelid closure. This analysis allows the system to detect signs of distraction or drowsiness in real-time. Upon detection, Dreyev issues immediate alerts to the driver, aiming to prevent potential accidents and promote safer driving practices. The technology is focused on proactive intervention to enhance overall road safety.

CleverHans is a Python library specifically developed for evaluating the robustness of machine learning systems when faced with adversarial examples. The library provides functionalities for constructing various types of adversarial attacks, implementing defensive strategies, and benchmarking the effectiveness of both. It is an actively maintained project that encourages contributions of new attacks and defenses from the research community. CleverHans serves as a crucial tool for researchers and developers aiming to identify and reduce vulnerabilities in artificial intelligence models, thereby enhancing their security and reliability.

Feast is an open-source feature store specifically built for artificial intelligence and machine learning applications. Its primary function is to manage and serve machine learning features efficiently. The tool supports both real-time and offline feature retrieval, providing flexibility for various ML workflows. By centralizing feature management, Feast aims to streamline the entire process of building and deploying machine learning models, making it easier for teams to develop and maintain their AI solutions.

Nemotron is an open-source family of large language models created by NVIDIA. It serves as a robust foundation for developers and researchers, enabling them to build and deploy custom generative AI applications. The models are designed to facilitate innovation across various AI tasks, particularly in natural language processing and understanding. Nemotron aims to provide the necessary tools and resources for advancing generative AI capabilities.

Untether AI specializes in delivering AI inference acceleration solutions designed for deployment across various environments, from edge devices to cloud infrastructure. The company's core offering revolves around an energy-centric architecture that efficiently supports a wide range of neural network models. A key focus is on significantly reducing data movement and overall power consumption, which are critical factors in AI workload efficiency. Untether AI empowers enterprises to execute their AI workloads effectively in diverse locations, including those outside traditional data centers.

VisRAG is a parsing-free Retrieval-Augmented Generation (RAG) tool that leverages Vision Language Models (VLMs). Its primary function is to support evidence-guided multi-image reasoning within visual RAG tasks. The tool is designed to assist researchers and engineers in developing and exploring advanced VLM applications. It is particularly useful for tasks that necessitate reasoning across multiple images, providing a robust framework for complex visual data analysis and generation.

AI & Computer Vision Lab (iVision) is a research institution at the Institute of Space Technology dedicated to exploring and advancing the fields of computer vision, machine learning, and signal processing. The lab serves as a hub for researchers to collaborate on fundamental advancements in these areas. iVision also actively engages with both public and private sector industries, applying its expertise to socio-economic projects and fostering practical applications of its research.

LLM-Adapters provides a framework for parameter-efficient fine-tuning (PEFT) of Large Language Models (LLMs). It integrates various adapter techniques directly into LLMs, allowing users to apply adapter-based PEFT methods for a wide range of tasks. This tool is built as an extension of HuggingFace's popular PEFT library, leveraging its existing capabilities while introducing additional functionalities for adapter management and application.

LMQL is a programming language designed for efficient and constraint-guided programming of Large Language Models (LLMs). Built as a superset of Python, it allows developers to seamlessly integrate traditional programming logic with calls to LLMs. This approach aims to provide a more structured and controlled way to interact with and program LLMs, moving beyond simple templating. It's intended for developers looking for advanced methods to build AI applications with greater precision and efficiency.

ChatGLM-Tuning provides a fine-tuning solution specifically designed for the ChatGLM-6B model, leveraging the LoRA (Low-Rank Adaptation) technique. This approach aims to offer an affordable and accessible method for users to implement a model with capabilities similar to ChatGPT. The tool supports the Alpaca dataset for training and requires a computing environment equipped with a GPU that has at least 16GB of VRAM, along with a Python environment configured with CUDA.

MinkowskiEngine is a specialized neural network library built for handling high-dimensional sparse tensors. It provides auto-differentiable capabilities, making it suitable for advanced deep learning tasks, particularly within the realm of 3D data processing. The library supports a variety of operations tailored for sparse tensors, which allows for more efficient computation and memory usage when dealing with large, sparse datasets common in 3D deep learning applications. Its design focuses on optimizing performance for these specific data structures.