Coding & Development

The MATLAB-Deep-Learning-Model-Hub is an open-source repository on GitHub offering a comprehensive collection of pretrained deep learning models specifically designed for use within the MATLAB environment. It covers a wide array of applications including computer vision tasks such as image classification, object detection, semantic segmentation, instance segmentation, image translation, pose estimation, 3D reconstruction, and video classification. Beyond vision, it also includes models for natural language processing (Transformers), audio analysis (embeddings, sound classification, pitch estimation, speech-to-text), and Lidar point cloud processing. This hub is ideal for researchers and developers looking to accelerate their deep learning projects by utilizing pre-trained models and applying transfer learning techniques.

ASC27 s.r.l. specializes in developing advanced artificial intelligence and cybersecurity solutions. The company focuses on creating software that leverages its proprietary AI engines to deliver value across various domains, including cognitive intelligence, video, audio, music, and media. Their AI innovation process leads in cognitive intelligence and analytic fields, aiming to bridge the gap between human and machine learning. Additionally, ASC27 develops robust cybersecurity solutions designed to provide peace of mind by protecting organizations from threats. They emphasize leading the CyberSec innovation process in strong and top security fields, offering a portfolio of products and solutions built on their expertise.

DeepTeam is a simple-to-use, open-source framework designed for red teaming large language models (LLMs) and LLM systems. It functions like penetration testing for AI, simulating various attacks such as jailbreaking, prompt injection, and multi-turn exploitation. The framework helps uncover vulnerabilities like bias, PII leakage, and SQL injection in AI agents, RAG pipelines, and chatbots. DeepTeam also provides production-ready guardrails to prevent these issues in real-time. It runs locally on your machine and is built on DeepEval, an open-source LLM evaluation framework. Users can define custom vulnerabilities and attacks, run red teaming from the CLI or programmatically in Python, and access risk assessments.

Deep_learning_cookbook is an open-source repository featuring 35 Python notebooks that illustrate fundamental machine learning techniques using the Keras framework. These notebooks are designed to accompany the book "Deep Learning Cookbook" but are fully functional as standalone educational resources. The collection covers a wide range of topics, from using pre-trained word embeddings and building recommender systems to generating text, classifying sentiments, and working with image recognition networks. It also includes examples for productionizing embeddings and preparing Keras models for deployment on platforms like TensorFlow Serving and iOS. While a GPU is not strictly required, its use is recommended for faster processing.

DeepQA is an open-source project that provides a TensorFlow implementation of "A neural conversational model," also known as the Google chatbot. This tool enables developers and researchers to build and experiment with deep learning-based conversational agents using a recurrent neural network (RNN) seq2seq model for sentence predictions. It supports various dialogue corpora, including Cornell Movie Dialogs, OpenSubtitles, Supreme Court Conversation Data, and Ubuntu Dialogue Corpus, with options to integrate custom datasets. DeepQA offers functionalities for training models, testing predictions, and visualizing computational graphs with TensorBoard. It also includes a web interface for user interaction, making it suitable for both development and demonstration purposes.

ml-diffucoder is an open-source research project accompanying the paper "DiffuCoder: Understanding and Improving Masked Diffusion Models for Code Generation." It explores how diffusion LLMs (dLLMs) differ from autoregressive models in code generation, investigates data modality differences (code vs. math), and examines diversity and post-training strategies for dLLMs. The project introduces Coupled-GRPO, a novel post-training method designed to enhance DiffuCoder's performance by improving the efficiency and accuracy of log-probability computations during training. The repository provides the implementation of Coupled-GRPO, built upon open-r1, along with scripts for training, data preparation, and inference. It also offers pre-trained DiffuCoder models (Base, Instruct, and cpGRPO) on HuggingFace, complete with usage examples for both base code completion and chat-based instruction following.

ML-University is a comprehensive, open-source platform designed for free learning in the field of machine learning. Curated by an ML enthusiast for the global community, it serves as a continuously updated repository of educational resources. The platform covers a wide array of topics, including Mathematics for ML, Machine Learning, Deep Learning, Natural Language Processing, Reinforcement Learning, Large Language Models, ML in Production, Quantum ML, and provides access to datasets, useful websites, and research papers. Users can contribute to its growth by suggesting improvements or sharing valuable resources through pull requests, fostering a collaborative learning environment for ML practitioners at all levels.

DL4NLP is a comprehensive GitHub repository dedicated to Deep Learning for Natural Language Processing (NLP). It serves as a valuable resource hub, offering state-of-the-art materials for various NLP sequence modeling tasks such as machine translation, image captioning, and dialog systems. The repository includes detailed notes on fundamental concepts like neural networks, RNNs, and LSTMs. It also curates links to prominent academic courses, including Stanford CS 224D and Oxford Deep Learning for NLP, complete with syllabi, slides, and lecture videos. Additionally, it provides access to seminal papers, code, and tutorials on key NLP topics like word vectors, sentiment analysis, neural machine translation, and conversation modeling, making it an essential reference for anyone studying or working in the field.

EasyRAG is a simple, lightweight, and efficient open-source framework for retrieval-augmented generation (RAG) specifically designed for automated network operations. It features an accurate question-answering scheme based on a specific data processing workflow, dual-route sparse retrieval for coarse ranking, an LLM Reranker, and LLM answer generation and optimization. The framework is easy to deploy, primarily consisting of BM25 retrieval and BGE-reranker reranking, requiring no model fine-tuning and occupying minimal VRAM. It also boasts efficient inference acceleration for the entire RAG process, significantly reducing latency while maintaining accuracy. EasyRAG provides a flexible code library with various search and generation strategies, facilitating custom process implementation.

Multimodal-GPT is an open-source project designed for training advanced multimodal chatbots capable of understanding and responding to both visual and language instructions. Built upon the OpenFlamingo model, it facilitates the creation of diverse visual instruction data by integrating open datasets from sources like VQA, Image Captioning, Visual Reasoning, Text OCR, and Visual Dialogue. The tool also enhances its language model component through training with language-only instruction data. This joint training approach significantly boosts the model's overall performance. Key features include support for various vision and language instruction data, parameter-efficient fine-tuning with LoRA, and the ability to tune vision and language simultaneously for complementary improvements. It's ideal for researchers and developers looking to build sophisticated conversational AI systems.

nextjs-ollama-llm-ui offers a comprehensive web interface for Ollama Large Language Models, designed for quick and easy local and offline deployment. Inspired by ChatGPT, it provides an intuitive user experience with features like fully responsive design, chat history, and light/dark mode. Users can easily download, pull, and delete models directly from the interface, and switch between them with a click. The tool also includes code syntax highlighting and one-click codeblock copying, making it ideal for developers and researchers working with LLMs. Its straightforward setup, requiring only Node.js and Ollama, makes it accessible for rapid experimentation and development.

ELITE (Encoding Visual Concepts into Textual Embeddings for Customized Text-to-Image Generation) is a method presented at ICCV 2023 that allows users to encode visual concepts from images into textual embeddings. These embeddings can then be flexibly composed into new scenes using text-to-image generation models like Stable Diffusion. The tool features a two-module architecture: a global mapping network for encoding concept images into multiple textual word embeddings, and a local mapping network that projects foreground objects into the textual feature space for detailed local control. ELITE is built on the diffusers version of Stable Diffusion and provides scripts for environment setup, customized generation, and training, including a Gradio demo for interactive testing.

Open Paws is a non-profit organization dedicated to ensuring the future of AI benefits all sentient beings. They achieve this by creating anti-speciesist artificial intelligence through open-source tools, hackathons, and research, specifically designed to power the animal advocacy movement. The organization also provides free technical training to animal rights activist groups, empowering them to leverage AI effectively. Furthermore, Open Paws assists AI companies in implementing ethical guidelines, promoting truly safe and trustworthy AI development that considers all life forms. They offer programs like the Code for Compassion Campus, which teaches participants to build AI tools for animal advocacy, climate action, and AI safety with ethics and impact as core principles.

FollowYourPose is an open-source implementation of the "Follow-Your-Pose: Pose-Guided Text-to-Video Generation using Pose-Free Videos" research paper from AAAI 2024. This tool allows users to generate character videos by combining pose information with text descriptions, leveraging pre-trained text-to-image models like Stable Diffusion. It features a two-stage training scheme that uses image-pose pairs and pose-free videos to achieve continuously pose-controllable character videos while retaining the editing and concept composition abilities of the underlying text-to-image model. The project provides code, configurations, and checkpoints, along with a local Gradio demo for easy experimentation, requiring an A100/3090 GPU.

Ovis (Open VISion) is an innovative Multimodal Large Language Model (MLLM) architecture available as an open-source project on GitHub. It is specifically designed to structurally align visual and textual embeddings, enabling advanced multimodal understanding and generation. Key features include native-resolution visual perception, enhanced reflective reasoning (thinking mode), and leading performance across STEM, chart analysis, grounding, and video understanding. Ovis supports various model sizes, from 2B to 34B parameters, and offers quantized versions for optimized deployment. It provides comprehensive installation and inference instructions, including examples for both transformers and vLLM, and supports fine-tuning with in-repo code or ms-swift.

FunClip is a fully open-source, locally deployable video clipping tool that leverages Alibaba TONGYI speech lab's FunASR Paraformer series models for highly accurate speech recognition. It allows users to select text segments or speakers from recognition results to generate corresponding video clips. A key differentiator is its integration of LLM-based AI for smart clipping, enabling users to utilize large language models like Qwen or GPT series with customizable prompts to extract specific video segments. FunClip also supports hotword customization for enhanced ASR accuracy, speaker diarization, and multi-segment free clipping. The tool provides a user-friendly Gradio interface for easy installation and server deployment, making it accessible for various video editing needs.

Free-Auto-GPT is an open-source repository providing a simplified version of autonomous AI agents like Auto GPT and BabyAGI. Unlike many other implementations, this tool is designed to function without reliance on paid OpenAI APIs, making it accessible and cost-effective for users. It leverages reverse-engineered ChatGPT, HuggingChat, Bing Chat, and Google Bard to provide free access to large language models. The project aims to democratize AI by offering a plug-and-play solution with LangChain, allowing users to create custom agents with internet access, Python code execution, and Wikipedia knowledge. It supports local usage and offers a quick start via Colab notebooks.

Weights was an AI platform that provided a comprehensive suite of free tools for various creative applications. Users could leverage the platform for voice cover generation, transforming text into speech, creating AI-generated images, and engaging in AI character chats. The platform aimed to empower creators, artists, and visionaries by offering accessible AI tools that could be integrated into their projects. However, as of April 1st, 2026, Weights has officially ceased operations, and all associated services and content are no longer available. The platform's journey was marked by a vibrant community that actively utilized and shaped its offerings, pushing the boundaries of what was possible with AI-powered creative tools.

ResShift is an efficient open-source diffusion model designed for image super-resolution, developed by Zongsheng Yue and others. It addresses the common limitation of slow inference speeds in diffusion-based SR methods by introducing a novel residual shifting technique, which drastically reduces the required sampling steps to as few as 15, or even 4 in its journal version, without compromising output quality. This approach constructs a Markov chain that efficiently transfers between high-resolution and low-resolution images. Beyond super-resolution, ResShift also supports applications like image deblurring, natural and face image inpainting, and blind face restoration. The project has been recognized at NeurIPS 2023 (Spotlight) and published in TPAMI@2024, highlighting its advanced capabilities and efficiency in image enhancement.

Embedl provides a comprehensive platform for developing and deploying efficient Edge AI. It offers both on-premise and cloud solutions tailored for Edge AI developers, focusing on optimizing performance and reducing costs. The platform includes Embedl Hub, a secure MLOps solution for compliant edge AI workflows, and Embedl Models, which provides popular models optimized for specific edge hardware. Embedl Deploy facilitates Edge AI conversion, compilation, and quantization to get models running on hardware easily. It supports a wide range of hardware platforms including Xilinx FPGAs, Nvidia GPUs, Texas Instruments DSPs, ARM CPUs, NXP NPUs, and Intel CPUs, GPUs, and FPGAs, and is compatible with any inference engine. The Embedl Model Optimization SDK helps developers prune, quantize, and compress models, significantly reducing model size and speeding up inference times.

Langtrace is an open-source observability and evaluations platform designed for AI agents, enabling the transformation of AI prototypes into enterprise-grade products. It offers a simple, non-intrusive setup with SDKs for Python and TypeScript, supporting frameworks like CrewAI, DSPy, LlamaIndex, and Langchain. Users can track vital metrics such as token usage, cost, and latency, and explore API requests with automatically traced GenAI stacks. The platform also provides evaluation capabilities to measure baseline performance, curate datasets, and offers prompt version control for managing and comparing different prompt versions. Langtrace emphasizes enterprise-grade security, is SOC2 Type II certified, and is proudly open source, allowing for customization and community contributions.

spaCy is a powerful, open-source library for advanced Natural Language Processing (NLP) in Python and Cython. Designed for production use, it incorporates the latest research and provides pre-trained pipelines for over 70 languages, enabling tokenization and training. Key features include state-of-the-art speed, neural network models for tasks like tagging, parsing, named entity recognition, and text classification, as well as multi-task learning with transformers like BERT. It boasts a robust training system, easy model packaging, deployment, and workflow management, making it suitable for industrial-strength applications. spaCy is released under the MIT license, offering a comprehensive solution for developers and researchers working with NLP.

ktrain is a lightweight Python library designed to simplify deep learning and AI application, acting as a wrapper for TensorFlow Keras and other machine learning libraries. It aims to make AI more accessible for both newcomers and experienced practitioners by providing a "Swiss Army knife" of functionalities. Users can quickly build, train, and deploy neural networks and other machine learning models with minimal code. The library offers pre-canned models for text, vision, graph, and tabular data, covering tasks like text classification, image recognition, named entity recognition, and generative AI. It also includes features like learning rate optimization, data preprocessing, and a simple prediction API for deploying models.

OpenBuddy is a robust multilingual AI chatbot model designed for conversational AI, offering seamless bilingual capabilities primarily in English and Chinese, with support for other languages. Built upon the Falcon model from Tii and the LLaMA model from Facebook, OpenBuddy delivers enhanced performance for handling complex conversational tasks. The platform is continuously evolving, with future development plans including support for more languages, multimodal capabilities, and optimization of model quality. It provides an accessible solution for developers and researchers looking to integrate advanced, customizable AI into their projects without proprietary restrictions, focusing on open-source principles.