Awesome GitHub RepositoriesShared Memory Transports

Apollo est une pile logicielle complète conçue pour le développement de véhicules autonomes, fournissant les composants nécessaires à la perception, à la planification et au contrôle. Il fonctionne comme un middleware robotique haute performance, utilisant un bus de données de type publication-abonnement pour faciliter la communication à faible latence entre les modules distribués et les capteurs matériels. La plateforme intègre les données des caméras, du lidar et du radar via un framework de fusion de capteurs pour générer un modèle environnemental en temps réel pour la navigation. Le système dispose d'un framework d'exécution basé sur des composants qui gère la planification des tâches et l'allocation des ressources, soutenu par une couche d'abstraction matérielle qui découple la logique de conduite des configurations spécifiques des véhicules. Pour garantir un comportement cohérent lors des tests, il inclut un moteur de relecture déterministe pour les flux de données des capteurs et prend en charge la simulation hardware-in-the-loop. La plateforme utilise également une planification par graphe acyclique dirigé et un transport en mémoire partagée sans copie pour optimiser le flux de données et l'efficacité computationnelle à travers des systèmes robotiques complexes. Le logiciel fournit une interface de contrôle de véhicule standardisée pour traduire les décisions de navigation en commandes mécaniques. Une documentation étendue est disponible, incluant des instructions d'installation, des guides d'intégration matérielle et une série de manuels de démarrage rapide pour diverses versions de la plateforme.

Arrow is a cross-language development platform for in-memory data. It provides a standardized, language-independent columnar memory format designed to accelerate analytical operations and improve memory efficiency on modern computing hardware. By utilizing a schema-driven approach, the framework enables the efficient organization of both flat and nested data structures. The project functions as an analytical data processing engine that facilitates high-performance computation directly on memory-resident datasets. It distinguishes itself through a zero-copy architecture, which allows multiple

This project is an NGINX module that embeds the Lua scripting language directly into the server environment. It functions as a request processor and response filter, enabling the execution of scripts to handle HTTP requests, generate dynamic content, and manage server behavior without external application calls. The module provides a shared memory dictionary and cache manager, allowing data to be stored and retrieved across all active worker processes. This capability supports the collection of high-performance server metrics and the synchronization of information across concurrent processes.

MacType is a system-level utility that replaces the default Windows font rasterization engine. It functions as a background service that intercepts and modifies font rendering calls to provide custom anti-aliasing, weight, and contrast adjustments for desktop applications. The software operates by injecting custom libraries into running processes to override standard text layout and graphics routines. It utilizes a shared memory space to apply configuration updates across multiple processes instantly, allowing for granular control over visual parameters such as gamma, hinting, and font substi

Crossplane is a Kubernetes-based control plane framework that functions as a cloud resource orchestrator and infrastructure-as-code platform. It enables the management of heterogeneous infrastructure by extending the Kubernetes API to provision and maintain external cloud services through declarative configuration. By utilizing custom resource controllers, it continuously reconciles the state of external infrastructure with defined desired states, ensuring consistent deployment and lifecycle management across multiple cloud providers. The platform distinguishes itself through its composition-

CuPy est une bibliothèque de calcul de tableaux CUDA qui implémente une interface compatible avec NumPy pour exécuter des opérations sur tableaux et du calcul numérique sur des GPU NVIDIA. Elle sert de bibliothèque numérique accélérée par GPU et d'implémentation SciPy basée sur CUDA, déchargeant les calculs lourds sur le matériel graphique pour augmenter la vitesse de traitement pour les charges de travail scientifiques et d'ingénierie. La bibliothèque permet l'échange de tenseurs multi-framework, permettant aux tampons de données d'être partagés entre différents frameworks d'apprentissage profond en utilisant des mises en page mémoire standardisées pour éviter les copies mémoire. Elle prend également en charge l'intégration de noyaux GPU personnalisés, permettant aux données de tableaux d'être connectées à des API de bas niveau pour un contrôle précis sur l'exécution matérielle. Globalement, le projet couvre le traitement de tableaux haute performance et les flux de travail de calcul scientifique. Ses capacités incluent l'accélération des calculs de tableaux et la fourniture d'outils pour les calculs numériques à grande échelle.

OpenVINO is an AI inference engine and model serving platform designed to execute optimized deep learning models across CPUs, GPUs, and NPUs through a unified API. It includes a model optimization toolkit for converting, quantizing, and compressing models from various frameworks, alongside a specialized generative AI runtime for large language models. The project distinguishes itself through a plugin-based hardware acceleration layer that maps neural network operations to vendor-specific drivers. It features advanced execution mechanisms such as continuous batching, speculative decoding, and

Napajs is an embeddable JavaScript engine and multi-threaded runtime designed to be integrated directly into other software applications as a component. It serves as a parallel computation framework that allows JavaScript code to execute across multiple threads, bypassing the standard single-threaded event loop limitation to handle CPU-intensive tasks. The runtime is distinguished by its ability to load and execute modules from the NPM ecosystem and its pluggable execution environment. This architecture allows for custom implementations of memory allocation, system logging, and performance me

jetson-inference is a set of libraries and tools for executing optimized deep learning models on embedded GPU hardware. Its primary purpose is to enable real-time computer vision and AI inference at the edge with low latency and high throughput. The project distinguishes itself through high-performance streaming analytics and the ability to execute concurrent AI pipelines on auto-grade silicon. It provides specialized support for multi-sensor stream processing, utilizing zero-copy data transport to load camera frames directly into GPU memory. The codebase covers a broad surface of capabiliti

Metalsmith is a Node.js static site generator and static content processor that transforms source files into websites, eBooks, or technical documentation. It functions as a file-to-object transformer, converting directory trees into plain JavaScript objects that can be programmatically manipulated in memory. The project is built around a pluggable build pipeline where files are passed through a sequence of custom functions to transform content and metadata incrementally. This architecture allows users to extend functionality by writing their own plugins or using third-party modules to define

LMCache is a distributed key-value cache manager and tiering system designed to accelerate large language model inference. It functions as a tiered storage layer that offloads tensors from GPU memory to CPU RAM, local disks, or remote object stores, enabling the reuse of cached prefixes across different inference sessions and serving engines. The system differentiates itself through a disaggregated prefill-decode model, which separates prompt processing from token generation by transferring caches between distributed compute nodes. It utilizes peer-to-peer orchestration to share and retrieve

Feast is an open-source feature store for machine learning that provides a central platform for defining, storing, and serving features across both training and inference workflows. It operates as a declarative system where feature definitions are written as code in Python files, synchronized to a central registry, and made available for low-latency online retrieval or point-in-time correct historical joins for training datasets. The project abstracts storage behind a pluggable architecture, allowing offline and online backends to be swapped without changing retrieval logic, and coordinates ma

AFL++ is a coverage-guided fuzzing framework that discovers crashes and hangs in software by mutating inputs while tracking which code paths are exercised. It functions as both a fuzzing engine and a campaign manager, supporting targets with or without source code through compile-time instrumentation, dynamic binary instrumentation, and emulation. The framework includes tools for crash triage and analysis, test case minimization, and campaign deployment across local or distributed environments. The framework distinguishes itself through its breadth of instrumentation backends, allowing users

This project provides a comprehensive technical guide and framework for engineering large-scale machine learning systems. It covers the full lifecycle of model development, focusing on the infrastructure and computational principles required to build, train, and serve generative AI models across distributed GPU clusters. The repository distinguishes itself by offering deep-dive tutorials and implementation strategies for complex system challenges. It emphasizes high-performance architectural primitives, such as collective communication orchestration, distributed tensor sharding, and static gr

Ce projet est une ressource éducative fournissant un tutoriel de développement complet pour écrire et charger des programmes eBPF en utilisant C, Go et Rust au sein du noyau Linux. Il sert de guide technique pour développer une logique personnalisée à exécuter directement dans le noyau. Les matériaux couvrent des domaines spécialisés, notamment l'observabilité et le traçage du noyau, l'implémentation de la sécurité pour la détection d'intrusion et l'ingénierie réseau haute performance pour le filtrage de paquets et l'équilibrage de charge. Il inclut également des manuels dédiés pour le traçage du noyau Linux et l'utilisation de kprobes, uprobes et tracepoints. Le projet englobe un large éventail de domaines de capacités, tels que l'instrumentation du noyau, la surveillance et l'observabilité du système, l'analyse réseau et l'application de la sécurité. Il s'étend en outre au débogage au niveau matériel pour les GPU et les pilotes, ainsi qu'à la manipulation système de bas niveau et à la gestion des ressources.

Ce projet est un pilote de caméra système macOS et un plugin logiciel qui expose les flux vidéo logiciels en tant qu'entrées caméra reconnues par le matériel. Il fonctionne comme un plugin de caméra virtuelle OBS, permettant d'utiliser la sortie en direct d'OBS comme un périphérique webcam dans d'autres applications. L'outil permet le routage de la vidéo composée depuis une suite de production vers des applications de visioconférence telles que Zoom ou Google Meet. Cela permet de diffuser des scènes traitées au lieu d'un flux webcam brut. Le système s'intègre à macOS en utilisant un pilote de périphérique au niveau du noyau et des transferts de mémoire partagée pour déplacer les trames vidéo du processus de l'application vers le système d'exploitation. Il utilise le framework CoreMedia pour gérer la synchronisation des flux vidéo et les métadonnées.

liburing is a C library that provides a low-level wrapper for the Linux kernel io_uring interface. It serves as a programming interface for executing non-blocking disk and network operations and abstracts the system calls required to communicate with the Linux kernel. The library focuses on reducing system call overhead and context switching for high-throughput data processing. It implements mechanisms for shared ring buffers, zero-copy buffer registration, and fixed-file descriptor mapping to minimize internal lookup and reference counting overhead. The project covers asynchronous input and

pyslam is a framework for Simultaneous Localization and Mapping that combines Python flexibility with C++ performance. It is a sparse SLAM implementation designed to map environment geometry and track device location by processing image frames into 3D points. The project features a bridge for exposing high-performance C++ classes to Python scripts using zero-copy memory sharing. This integration allows for switching between a scripting interface for rapid prototyping and a compiled core for execution speed. The system includes a spatial map optimizer to refine 3D point and camera pose estima

Dora is a robotics dataflow framework and distributed orchestrator used to build and manage processing pipelines. It enables the deployment of robotics workloads across clusters with remote node execution and provides a real-time data pipeline for predictable performance. The system is distinguished by its support for multi-language nodes written in Rust, Python, C, or C++ that interoperate within a single dataflow. It utilizes a zero-copy shared-memory transport and columnar formats to minimize latency for large payloads, and it includes bidirectional bridges to integrate with external ecosy