Coding & Development

one-pixel-attack-keras is an open-source project offering a Keras implementation of the "One pixel attack for fooling deep neural networks." This tool demonstrates how minimal perturbations, specifically changing just one pixel's color, can cause deep neural networks to misclassify images. It leverages differential evolution on datasets like Cifar10 and ImageNet to iteratively generate adversarial images and minimize the network's classification confidence. The project includes tutorial notebooks, various CNN models (LeNet, ResNet, DenseNet, CapsNet), and scripts for training and attacking models. It's particularly useful for understanding and researching the robustness and vulnerabilities of deep learning models to adversarial attacks.

oneAPI Deep Neural Network Library (oneDNN) is an open-source, cross-platform performance library designed to provide basic building blocks for deep learning applications. As part of the UXL Foundation, oneDNN implements the oneAPI specification for its component, offering optimized functions for neural network operations. The library is highly optimized for Intel 64/AMD64 architecture-based processors, Arm(R) 64-bit Architecture (AArch64)-based processors, and Intel Graphics, with experimental support for NVIDIA* GPU, AMD* GPU, OpenPOWER* Power ISA, IBMz*, and RISC-V. It is intended for deep learning application and framework developers looking to enhance performance on CPUs and GPUs, and is integrated into popular frameworks like PyTorch and TensorFlow.

Penzai is an open-source JAX library developed by Google DeepMind, designed for building, editing, and visualizing neural networks. It enables users to represent models as legible, functional pytree data structures, making it particularly useful for research involving reverse-engineering, ablating model components, inspecting internal activations, and debugging architectures. The toolkit includes Treescope for interactive pretty-printing and array visualization, `penzai.core.selectors` for advanced pytree manipulation, and `penzai.core.named_axes` for flexible named axis programming. Its declarative combinator-based neural network library, `penzai.nn`, offers an alternative to other frameworks by exposing the full model structure, supporting mutable state and parameter sharing. Penzai also provides a modular implementation of Transformer architectures, including pre-trained weights for Gemma, Llama, Mistral, and GPT-NeoX/Pythia, simplifying complex model-manipulation workflows.

agent-sandbox is a Kubernetes-native project developing a Sandbox Custom Resource Definition (CRD) and controller designed for easy management of isolated, stateful, singleton workloads. It's particularly well-suited for use cases like AI agent runtimes, development environments, and persistent single-container sessions for tools like Jupyter Notebooks. The core Sandbox CRD offers a declarative API for managing a single, stateful pod with stable identity and persistent storage, addressing limitations of standard Kubernetes Deployments and StatefulSets for these specific needs. Key features include stable identity, persistent storage, and comprehensive lifecycle management. Extensions like SandboxTemplate, SandboxClaim, and SandboxWarmPool further enhance its capabilities by providing reusable templates, user-friendly provisioning, and pre-warmed pools for rapid allocation.

anago is a Python library designed for sequence labeling tasks, including Named Entity Recognition (NER) and Part-of-Speech (PoS) Tagging. Built with Keras, it leverages advanced models like Bidirectional LSTM-CRF and ELMo to achieve high performance. A key differentiator is its independence from language-dependent features, making it easily adaptable for various languages. The library offers essential methods for model training, evaluation, and text tagging, along with support for custom models, pre-trained model downloads, and GPU acceleration. It's particularly useful for researchers and developers working on natural language processing applications.

Lyrprompt is an AI prompt and knowledge base builder designed to streamline the AI application development process. It enables users to transform project context into optimized, platform-specific prompts, ensuring consistency and accuracy in AI outputs. The tool features a prompt editor and offers proven templates from sources like Lovable.dev and Bolt.new. Lyrprompt also helps in analyzing prompt structure and provides an optimization score. Users can sign up to unlock additional generations per day, making it a valuable asset for developers looking to build robust AI applications with well-structured prompts and knowledge bases.

CodeWP, now known as Telex, is an AI-assisted authoring environment specifically designed for WordPress development. It transforms ideas into fully functional WordPress projects by providing AI-driven code generation capabilities. Developers can leverage Telex to create WordPress plugins and code snippets efficiently, significantly speeding up their workflow. The platform aims to simplify complex development tasks, allowing users to focus on innovation rather than repetitive coding. With its focus on WordPress, Telex offers specialized assistance tailored to the ecosystem, making it a valuable tool for anyone involved in building or maintaining WordPress websites.

Chat with DeepSeek Coder 33B is an AI code assistant designed to streamline the coding process for developers. This tool allows users to input questions or requests related to programming, such as asking for specific code snippets, explanations of complex concepts, or general coding assistance. It leverages advanced language models trained on extensive datasets of code and natural language to generate relevant text or code as a reply. Users can further customize their interaction by adding a system prompt and adjusting various settings, making it a flexible solution for a wide range of coding needs. The application aims to enhance coding efficiency and productivity by providing on-demand support and generation capabilities.

Synthetik Applied Technologies is an innovative research and development company specializing in advanced computational modeling, physics-informed AI, and synthetic data solutions. They address complex real-world challenges across various sectors including defense, security, energy, environmental resilience, transportation security, and insurance. Their offerings include specialized solvers like blastFoam and blastCFD for explosive events, SynthShock for hypersonic flow simulations, and Segmentastic for efficient 3D data annotation. Synthetik also provides AI-driven analytics for multi-modal sensor data, anomaly detection, and advanced catastrophe modeling solutions for the insurance sector, leveraging tools like srccQuantum and CityScape. Their technology supports critical infrastructure projects and enhances operational readiness through precise simulations and data insights.

Cycorp specializes in Machine Reasoning AI, utilizing a vast knowledge base of 25 million assertions and higher-order logic to enable human-like cognitive processing. Unlike traditional machine learning, Cyc uses codified human common sense and knowledge to perform autonomous data gathering and logical decision-making, mitigating human error and biases. The platform offers real-time, continuous, and semantically enriched representations for various enterprise operations, ensuring trust and auditability. Cyc's applications include autonomous leveling/charge capture, denial management, post-acute care forecasting, and staffing, primarily serving the healthcare sector. Its unique approach allows for flexible knowledge representation, contextual understanding through microtheories, and efficient knowledge addition through ontological engineering, structured data integration, and specialized tools.

Guardrails Arena is an open-source platform designed to help users jailbreak Large Language Models (LLMs) and test their privacy guardrails. Developed by Lighthouz AI, this tool facilitates the stress testing of LLMs to identify vulnerabilities and evaluate data privacy within AI systems. It promotes community-driven AI testing, allowing users to collaborate in uncovering weaknesses in AI chatbot security. The platform is hosted on Hugging Face, making it accessible for developers and researchers interested in AI safety and security. While the current status shows a build error, its core purpose is to provide a sandbox for ethical hacking and security assessment of AI models.

Gradio Canvas 🤗 is a web application designed to assist users in generating and refining Python code. Inspired by ChatGPT's Canvas, this tool aims to provide a seamless coding experience by offering feedback and suggestions based on user input. Users can enter their coding requests, and the application will generate the corresponding code. While the current live website indicates a runtime error and workload eviction, the underlying concept focuses on leveraging AI for structured code generation, potentially utilizing models like Llama 3.1 405B via Fireworks AI and Instructor for its capabilities.

The tensorflow_cookbook is a comprehensive GitHub repository that serves as a practical guide for implementing machine learning algorithms with TensorFlow. It accompanies the Tensorflow Machine Learning Cookbook by Nick McClure, offering code examples across a wide range of topics. Users can explore chapters dedicated to linear regression, support vector machines, nearest neighbor methods, neural networks, natural language processing, and convolutional neural networks. The repository details how to set up TensorFlow, work with tensors, variables, and operations, implement activation functions, and handle various data sources. It also covers advanced topics like computational graphs, loss functions, backpropagation, and taking TensorFlow models to production, making it an invaluable resource for both learning and applying TensorFlow in real-world scenarios.

TinyZero offers a minimal reproduction of DeepSeek R1-Zero, focusing on reinforcement learning tasks. Built upon the veRL library, this tool allows 3B base Large Language Models (LLMs) to independently develop self-verification and search capabilities. The project provides scripts and instructions for data preparation and training, including configurations for single GPU and multi-GPU setups, and supports instruct ablation experiments. While the repository is no longer actively maintained, it serves as a valuable resource for understanding and replicating the core concepts of DeepSeek R1-Zero, particularly for researchers and developers exploring advanced RL techniques for LLMs.

TNN is a high-performance, lightweight neural network inference framework developed by Tencent Youtu Lab and Guangying Lab. It provides a uniform deep learning inference solution for mobile, desktop, and server environments. Key features include cross-platform compatibility, high performance, model compression, and code pruning. Building upon the foundations of ncnn and Rapidnet, TNN enhances support and optimizes performance specifically for mobile devices, while also incorporating the extensibility and high-performance characteristics of other open-source frameworks. It has been deployed in various Tencent applications like Mobile QQ, Weishi, and Pitu, and serves as a core acceleration framework for Tencent Cloud AI. TNN supports models from TensorFlow, PyTorch, MxNet, and Caffe via ONNX, and runs on Android, iOS, embedded Linux, Windows, and Linux, compatible with ARM CPU, X86 GPU, and NPU hardware.

tiny-llm provides a comprehensive course for system engineers focused on learning LLM inference serving, specifically tailored for Apple Silicon. The curriculum guides users through building a tiny vLLM using MLX and Qwen, with a codebase primarily utilizing MLX array/matrix APIs. This approach allows participants to construct model serving infrastructure from scratch, gaining deep insights into optimizations. The course covers essential components like attention, RoPE, KV cache, and continuous batching, with a roadmap extending to advanced topics such as Paged Attention and Speculative Decoding. It's designed for those who want to understand the underlying techniques for efficiently serving large language models.

WebLLM is a high-performance, in-browser LLM inference engine designed to bring language model inference directly onto web browsers with hardware acceleration. It operates entirely within the browser, eliminating the need for server support and leveraging WebGPU for enhanced performance. The engine is fully compatible with OpenAI API, allowing users to apply the same API functionalities, including streaming, JSON-mode, and function-calling, to open-source models locally. WebLLM supports a wide range of models like Llama 3, Phi 3, Gemma, and Mistral, and allows for custom model integration in MLC format. It offers structured JSON generation, real-time interactions, and supports Web Worker and Service Worker for optimized performance and offline capabilities.

WeightWatcher (WW) is an open-source, diagnostic tool designed to analyze Deep Neural Networks (DNNs) and predict their accuracy. It operates without requiring access to training or even test data, leveraging theoretical research into Heavy-Tailed Self-Regularization (HT-SR), Random Matrix Theory (RMT), Statistical Mechanics, and Strongly Correlated Systems. Users can analyze pre/trained pyTorch, Keras, and other DNN models (Conv2D and Dense layers), monitor model layers for over-training or over-parameterization, and predict test accuracies across different models. The tool also helps detect potential problems when compressing or fine-tuning pretrained models and provides layer warning labels like 'over-trained' or 'under-trained'. It offers various generalization metrics and advanced diagnostics like correlation trap analysis and experimental early stopping detection.

Agent Leaderboard is a Hugging Face Space designed to rank Large Language Models (LLMs) based on their performance in agentic tasks. This tool provides a dynamic platform for users to browse and filter performance leaderboards across various categories, methodologies, and metrics. Users can select specific criteria to instantly update the displayed tables and charts, offering a comprehensive overview of different models' capabilities. It's an essential resource for developers and data scientists looking to compare and evaluate LLMs for their agentic applications, ensuring they can make informed decisions based on up-to-date performance data.

hackingBuddyGPT is an open-source framework designed to empower ethical hackers and security researchers to leverage Large Language Models (LLMs) for discovering new attack vectors and performing security testing. It allows users to build AI hacking agents with minimal code, often in 50 lines or less. The tool supports both SSH connections to remote targets and local shell execution, providing flexibility for testing and development. It aims to become a leading framework for security professionals interested in using LLMs or LLM-based autonomous agents for tasks like privilege escalation, web penetration testing, and API testing. The project also offers reusable Linux privilege escalation benchmarks and publishes open-access reports to aid experimentation.

Hyperparameter-Optimization-of-Machine-Learning-Algorithms is an open-source GitHub repository offering practical implementations of hyperparameter optimization and tuning methods for machine learning and deep learning models. It aims to be easy and clear for users, providing sample code for both regression and classification problems using benchmark datasets like Boston-Housing and MNIST. The repository covers various HPO algorithms such as Grid Search, Random Search, Hyperband, Bayesian Optimization (with GP and TPE), Particle Swarm Optimization, and Genetic Algorithms. It also details the hyperparameter configuration spaces for common ML models like Random Forest, SVM, KNN, and ANN. This resource is ideal for industrial users, data analysts, and researchers looking to effectively tune their machine learning models.

fastn is an embedded integration and automation platform designed for SaaS companies and AI agent builders. It streamlines the process of integrating various tools and systems, eliminating the need for manual development and maintenance of integrations. The platform leverages AI agents to power these integrations, ensuring reliable and efficient workflows. Key features include adaptive context for agents, tool orchestration to reduce context bloat and token usage, and tool composition for lower latency and API costs. fastn also offers performance optimization through caching and schema adaptation, along with comprehensive tracking for auditability. It enforces enterprise-grade security, including RBAC, compliance, and prompt safety, and provides access to over 1000 tools with the ability to build infinite more. The platform is SOC 2 Type II, ISO-Certified, GDPR Compliant, and HIPAA- and PCI-Ready.

GRPO-Zero is an open-source project that provides a from-scratch implementation of DeepSeek R1's Group Relative Policy Optimization (GRPO) algorithm. This tool is specifically designed for training large language models using reinforcement learning, emphasizing minimal dependencies and efficient GPU memory usage. It supports training on GPUs with limited VRAM, such as a 24GB RTX 4090, by offloading the optimizer to the CPU, incurring only a small overhead. Key improvements over the original GRPO include token-level policy gradient loss, removal of KL Divergence for reduced GPU memory, and optional overlong episode filtering to stabilize training. The project demonstrates its capabilities by training Qwen2.5 models on a CountDown task, where the model learns to generate mathematical expressions and reasoning.

GenossGPT is an open-source initiative designed to offer a seamless alternative to OpenAI models like GPT 3.5 and 4, by leveraging open-source models such as GPT4ALL. It provides a single API endpoint that is compatible with the OpenAI ChatGPT API, supporting the same parameters and response format. This allows developers to easily integrate various LLMs, including private and public options like Anthropic, Llama V2, GPT 3.5/4, Vertex, GPT4ALL, and HuggingFace, into their applications. The project aims to enable local execution of LLM models, support HuggingFace inference APIs, and facilitate easy installation of LLMs. Future developments include broader cloud provider support and diverse model types beyond text-to-text.