Coding & Development

Meshed-Memory Transformer (M²) is an open-source project that provides the reference code for the paper "Meshed-Memory Transformer for Image Captioning" presented at CVPR 2020. This tool is designed for researchers and developers working in computer vision and natural language processing. It allows users to set up a conda environment, download necessary data like COCO annotations and detection features, and then evaluate or train their own image captioning models. The repository includes scripts for both testing and training, with configurable arguments for batch size, number of memory vectors, and learning rate scheduling. It requires Python 3.6 and specific data preparation steps to function correctly.

Mocha is an AI-powered no-code app builder designed to help entrepreneurs and non-technical users quickly launch their ideas as live websites. It streamlines the development process, allowing users to build web applications without writing a single line of code. Mocha focuses on speed and ease of use, enabling rapid prototyping and deployment of online businesses. The platform leverages AI to assist in the creation process, making full-stack development accessible to a broader audience. It's an ideal solution for anyone looking to bring their app ideas to reality efficiently and without the need for traditional coding expertise.

DeepExplain offers a comprehensive framework for understanding the behavior of deep neural networks through various attribution methods. It enables researchers and practitioners to interpret existing models and benchmark new attribution techniques. The tool supports both gradient-based methods like Saliency maps, Gradient * Input, Integrated Gradients, DeepLIFT, and epsilon-LRP, as well as perturbation-based methods such as Occlusion and Shapley Value sampling. DeepExplain is compatible with TensorFlow (V1) and Keras with a TensorFlow backend, providing flexibility for different development environments. Its capabilities help in identifying which input features contribute most to a network's output, aiding in debugging and model transparency.

PLAI Accelerator is an Italian accelerator dedicated to fostering the growth of innovative startups and cultivating a community of talents within the Generative AI sector. The program provides comprehensive support, including equity funding up to 300 K€ and collaboration opportunities with Mondadori Group up to 100 K€. Startups benefit from tailored mentorship from experienced entrepreneurs and industry leaders, strategic business support, and access to a vast network of experts and resources. PLAI aims to empower ventures to achieve market leadership, enhance operational efficiency, and maximize company value through strategic exits. The accelerator focuses on industries such as education, retail, publishing, and digital media, leveraging Mondadori Group's extensive network and market presence to amplify startup reach and impact.

CerebrumEdge offers AI-powered solutions to enhance workplace safety and prevent injuries through its ErgoEdge and SafetyEdge platforms. ErgoEdge utilizes AI and computer vision to assess musculoskeletal disorder (MSD) risk from standard smartphone video, eliminating the need for sensors or specialized hardware. It delivers automated RULA, REBA, NIOSH, ART, and MAC scores with 3D posture simulation, providing EHS teams with objective, data-backed insights rapidly. SafetyEdge transforms existing CCTV cameras into AI-powered safety monitoring devices, detecting PPE non-compliance, restricted area breaches, vehicle hazards, and fire or smoke in real time, and sending instant alerts. The platform aims to reduce liability, decrease absenteeism, improve employee performance, and boost motivation by creating safer, more productive workplaces.

Deepdrive is an open-source simulator designed to facilitate experimentation and advancement in self-driving AI. It enables anyone with a PC to develop and test state-of-the-art autonomous driving systems within a realistic simulated environment. The simulator supports various AI agent types, including forward-agents, remote agents, and baseline agents like Mnet2 and C++ FSM/PID. Users can record training data for imitation learning, convert data to TFRecords, and train models using provided datasets or their own. Deepdrive offers detailed observation data, including vehicle dynamics, camera feeds (image, depth), and environmental information, all adhering to Unreal Engine conventions for units and rotations. It requires Linux, Python 3.6+, 10GB disk space, and 8GB RAM, with optional GPU requirements for baseline agents.

Learn_Prompting is a leading resource for individuals and businesses looking to master generative AI and prompt engineering. The platform offers a wide array of free resources, including a comprehensive Prompt Engineering Guide cited by OpenAI and Google, alongside 15 specialized courses designed to develop cutting-edge AI skills. Beyond self-paced learning, Learn_Prompting provides on-demand workshops and training for both individuals and businesses, with a track record of hosting sessions at major tech companies like OpenAI, Microsoft, and Deloitte. The initiative also organizes HackAPrompt, one of the largest AI red-teaming competitions, and conducts research on prompting techniques and LLM vulnerabilities, making it a valuable hub for both learning and advancing the field of AI.

Pulid AI is a cutting-edge AI-powered platform designed to generate professional headshots quickly and easily. Users can upload just one photo of themselves and then customize various aspects, including professional backgrounds, outfits, accessories, and styles. The tool leverages advanced AI technology to produce lifelike and high-quality headshots with exceptional detail and realism. It also offers diverse pose options to best represent individual personalities. Pulid AI provides a free service, making it accessible for students, job seekers, and professionals looking to enhance their online presence without the need for expensive photoshoots.

micro_diffusion is an open-source repository from Sony Research that provides a minimalistic implementation for training large-scale diffusion models from scratch with an extremely low budget. Utilizing only 37 million publicly available real and synthetic images, it can train a 1.16 billion parameter sparse transformer for approximately $1,890, achieving a strong FID score on the COCO dataset. The repository includes training code, dataset code, and pre-trained model checkpoints for off-the-shelf generation. It supports progressive training from low to high resolution and incorporates patch masking for performance optimization and reduced training time.

GitHub is a comprehensive platform designed for software development, offering a wide array of tools and services for individuals, teams, and enterprises. It facilitates code creation with AI assistance like GitHub Copilot, streamlines developer workflows through features such as Actions and Codespaces, and enhances application security with Advanced Security. The platform supports various use cases, from open-source projects to large-scale enterprise solutions, providing robust version control, collaboration tools, and project management capabilities. GitHub offers different pricing tiers, including a free plan with unlimited public/private repositories and basic CI/CD minutes, a Team plan with advanced collaboration features, and an Enterprise plan focused on security, compliance, and flexible deployment options. It also provides add-ons like GitHub Models for integrating AI into workflows and Premium Support for enhanced assistance.

how-to-optim-algorithm-in-cuda is a comprehensive open-source repository dedicated to optimizing algorithms using CUDA. It offers a wealth of resources including code implementations for fundamental CUDA operators like reduce, softmax, and elementwise operations, as well as detailed learning notes and blog translations related to GPU and large language models. The project covers advanced topics such as CUTLASS, CuTe DSL, Triton, and PTX ISA, making it an invaluable learning tool for developers aiming to enhance the performance of their CUDA code. It also includes notes on large language model inference/training optimization and GPU/AI system papers.

Digital Transit Limited specializes in advancing infrastructure technology systems through expertise in cybersecurity, AI, condition monitoring, and safety-critical software. The company provides expert consultancy, specialized training, and an AI-driven verification platform to help organizations navigate complex industry standards. Their product suite includes CyRail, an AI-powered tool for assessing compliance to industry standards, and AI For Net-Zero, which utilizes emerging AI technologies for optimizing micro-grids. Digital Transit aims to enable secure, digitally driven systems, support industry innovation, improve quality of life through infrastructure resilience, and guide organizations through digital transformation with confidence.

Chrono Rift The Last Day is a unique roguelite turn-based RPG that leverages AI in its development process. The game plunges players into a post-apocalyptic world where all life shares a single 'life time' resource. Players embark on dangerous dungeon explorations to recover 'time fragments' to extend their own life and protect their daughters. Combat features a D20 dice system, introducing an element of chance to attacks and evasions. Players can also develop powerful skills across three trees—Chrono Blade, Loop Guardian, and Fate Weaver—to customize their combat style. The game's development journey, detailed on the website, highlights the use of AI for tasks like planning, combat calculations, and code refactoring, showcasing a new paradigm for game creation.

Flyde is an open-source visual programming tool designed for backend logic, seamlessly integrating with existing codebases. It provides a visual extension of TypeScript, allowing both technical and non-technical team members to collaborate on the same visual flows. Flyde is particularly useful for prototyping, integrating, evaluating, and iterating on AI-heavy backend logic, such as AI agents, prompt chains, and API orchestration. It runs directly in your codebase, offering a VSCode extension and full compatibility with existing TypeScript/JavaScript code. This approach helps lower collaboration barriers and provides visual clarity for complex backend AI workflows, making it easier to build, debug, and maintain systems.

stablediffusion-infinity is an open-source tool designed for outpainting using Stable Diffusion on an infinite canvas. This innovative tool empowers users to seamlessly expand images beyond their original boundaries, offering a flexible and creative environment for digital art. It is particularly well-suited for generating expansive and detailed digital artworks, allowing for continuous image generation and exploration. The tool is accessible on platforms like Google Colab and Hugging Face Spaces, making it readily available for a wide range of users interested in advanced image manipulation and generation techniques.

Comfy Deploy is a platform designed to streamline the deployment and management of ComfyUI workflows for teams. It allows users to share ComfyUI workflows via links, run them instantly in the cloud, or create simplified user interfaces. The platform facilitates one-click deployment of production APIs, ensuring effortless scaling without requiring extensive engineering knowledge. Key features include the ability to install custom nodes and models, share multiple environments, and leverage powerful auto-scaling GPUs like H100s and A100s for parallel cloud generation. Comfy Deploy also offers full version history for workflows, making collaboration and iteration easy, and provides a playground for testing and refining workflows with auto-generated UIs.

LLMs-Zero-to-Hero is an open-source educational resource designed to guide individuals from basic understanding to advanced proficiency in Large Language Models (LLMs). The project emphasizes a hands-on approach, providing fully handwritten code examples and detailed explanations for each concept. It covers a wide range of topics, including the training process of dense and MOE models, pre-training, fine-tuning (SFT, DPO, RLHF), and deployment strategies like inference optimization and quantization. The resource also includes配套视频讲解 (accompanying video explanations) on Bilibili and offers GPU mirror images for model training, with a minimum requirement of 3090/4090 GPUs. It aims to provide a systematic learning path for aspiring LLM developers.

DeepSeek-V3 is a powerful Mixture-of-Experts (MoE) language model featuring 671B total parameters, with 37B activated for each token, ensuring efficient inference and cost-effective training. Building on the DeepSeek-V2 architecture, it introduces an innovative auxiliary-loss-free strategy for load balancing and a multi-token prediction training objective for enhanced performance. The model was pre-trained on 14.8 trillion diverse tokens and further refined through Supervised Fine-Tuning and Reinforcement Learning. DeepSeek-V3 demonstrates superior performance against other open-source models and rivals top closed-source alternatives, particularly excelling in math and code tasks. It supports local deployment on various hardware and open-source community software, including SGLang, LMDeploy, and TensorRT-LLM, with options for FP8 and BF16 inference.

AIforWork.co provides free access to an extensive library of over 2000 advanced ChatGPT prompts, specifically curated for professionals. This platform aims to boost productivity and efficiency for individuals working in diverse fields such as marketing, sales, business, law, human resources, finance, and education. Users can select their department to find relevant prompts and resources, including custom GPTs. The tool serves as a comprehensive resource for leveraging AI in professional tasks, offering a structured approach to utilizing ChatGPT for various projects and daily operations. It's designed to be a go-to source for professionals seeking to integrate AI into their workflows effectively.

Panels specializes in providing high-quality audio datasets for training and evaluating speech and audio models. The platform works closely with frontier voice labs and early-stage startups to curate data that matches specific team needs. Key offerings include proprietary, large-scale multilingual datasets with speaker-separated audio across diverse topic domains, single speaker scripted audio covering various recording environments, and multilingual datasets for evaluating human-agent turn-taking models. Panels also offers a custom data design service, allowing users to specify their unique data requirements. The process involves in-depth research to define use cases and data requirements, in-house collection with rigorous QA and transcription, and iterative expansion to grow coverage and performance over time.

Zeta is a modular PyTorch framework designed to simplify the development of high-performance AI models. It provides a comprehensive library of reusable building blocks, including attention mechanisms (multi-query, sigmoid, flash), Mixture of Experts (MoE), neural network modules (feedforward, activation, normalization), and quantization techniques like BitLinear. The framework also offers complete model implementations such as Transformers, encoders, decoders, vision transformers, and multi-modal architectures like PalmE and U-Net. Zeta emphasizes modularity, high-performance with optimized implementations, and production-readiness, making it suitable for quickly assembling state-of-the-art models without reinventing the wheel. It also includes optimization utilities like dynamic quantization and fused operations for improved performance.

MLAlgorithms provides a comprehensive collection of machine learning algorithm implementations, including deep learning models like MLP, CNN, RNN, and LSTM, as well as classical algorithms such as linear regression, logistic regression, Random Forests, SVM, K-Means, and Naive Bayes. All algorithms are implemented in Python, leveraging libraries like NumPy, SciPy, and Autograd, making the code easy to follow and experiment with. This project is ideal for those who wish to understand the core mechanics of these algorithms without the complexity of highly optimized libraries, offering a simplified approach to learning and building ML models from the ground up.

TensorFlow.NET (TF.NET) offers .NET Standard bindings for Google's TensorFlow, allowing C# and F# developers to build, train, and deploy Machine Learning models within the cross-platform .NET Standard framework. It aims to implement the complete TensorFlow API in C# and includes a built-in Keras high-level interface, released as an independent package TensorFlow.Keras. This project facilitates the migration of Python-based machine learning code to .NET, providing access to a vast ecosystem of TensorFlow resources. Unlike other projects that only expose low-level C++ APIs, TensorFlow.NET enables the entire training and inference pipeline to be constructed using pure C# and F#. It also serves as a backend for ML.NET, offering better integration within the .NET environment.

DeepMoji is a state-of-the-art deep learning model designed for analyzing sentiment, emotion, and sarcasm in textual data. The model was trained on an extensive dataset of 1.2 billion tweets, leveraging emojis to understand how language expresses various emotions. It offers transfer learning capabilities, allowing it to achieve state-of-the-art performance on numerous emotion-related text modeling tasks. Developers can use DeepMoji to extract emoji predictions, convert text into 2304-dimensional emotional feature vectors, or fine-tune the model for new datasets. The project is open-source and based on Keras, supporting either Theano or TensorFlow as a backend. A PyTorch implementation, torchMoji, is also available.