Awesome GitHub RepositoriesGPU Acceleration Backends

ds4 is a local inference engine for DeepSeek models that includes a distributed runtime for splitting transformer layers across networked computers. It functions as a reasoning controller with a local weight streamer and an API server that streams chat completions via industry standard endpoints. The system employs a memory management model that loads model experts from disk on demand to execute models that exceed available system RAM. It provides controls for reasoning effort and model behavior steering, allowing the modification of response characteristics through activation directions. Th

Dask एक पैरेलल कंप्यूटिंग फ्रेमवर्क और डिस्ट्रीब्यूटेड टास्क शेड्यूलर है जिसे Python डेटा साइंस वर्कफ़्लो को सिंगल मशीनों से बड़े क्लस्टर्स तक स्केल करने के लिए डिज़ाइन किया गया है। यह एक क्लस्टर रिसोर्स मैनेजर के रूप में कार्य करता है जो कार्यों और उनकी डिपेंडेंसी को डायरेक्टेड एसाइक्लिक ग्राफ (DAGs) के रूप में प्रस्तुत करके कम्प्यूटेशनल लॉजिक को व्यवस्थित करता है। यह आर्किटेक्चर सिस्टम को जटिल निष्पादन आवश्यकताओं का प्रबंधन करते हुए उपलब्ध हार्डवेयर पर वर्कलोड के वितरण को स्वचालित करने की अनुमति देता है। यह प्रोजेक्ट एक लेज़ी इवैल्यूएशन इंजन के माध्यम से खुद को अलग करता है जो डेटा ऑपरेशन्स को तब तक स्थगित कर देता है जब तक कि उन्हें स्पष्ट रूप से अनुरोध न किया जाए, जिससे ग्लोबल ग्राफ ऑप्टिमाइज़ेशन और कुशल संसाधन आवंटन सक्षम होता है। इसमें उपलब्ध मेमोरी से अधिक डेटासेट को प्रोसेस करते समय सिस्टम क्रैश को रोकने के लिए मेमोरी-अवेयर डेटा स्पिलिंग शामिल है, और यह टास्क ग्राफ फ्यूजन का उपयोग ऑपरेशन्स के अनुक्रमों को एकल निष्पादन चरणों में संयोजित करने के लिए करता है, जिससे शेड्यूलिंग ओवरहेड और इंटर-नोड संचार कम हो जाता है। यह प्लेटफॉर्म बड़े पैमाने पर डेटा एनालिटिक्स के लिए एक व्यापक क्षमता सतह प्रदान करता है, जिसमें डिस्ट्रीब्यूटेड मशीन लर्निंग, उच्च-प्रदर्शन कंप्यूटिंग एकीकरण, और पैरेलल डेटा प्रोसेसिंग के लिए समर्थन शामिल है। यह क्लस्टर लाइफसाइकिल मैनेजमेंट, परफॉरमेंस प्रोफाइलिंग, और टास्क निष्पादन की रीयल-टाइम मॉनिटरिंग के लिए व्यापक उपकरण प्रदान करता है। उपयोगकर्ता इन वातावरणों को स्थानीय हार्डवेयर, क्लाउड प्रदाताओं, कंटेनरीकृत सिस्टम, और उच्च-प्रदर्शन कंप्यूटिंग क्लस्टर्स सहित विविध बुनियादी ढांचे पर तैनात कर सकते हैं।

cuDF is a GPU-accelerated dataframe library and data processing engine designed for manipulating and analyzing large tabular datasets. It provides a high-level API for executing filtering, joining, and aggregating operations directly on GPU hardware. The project integrates the Apache Arrow memory format to enable zero-copy data transfers and includes a just-in-time compiler for executing custom user-defined functions on the GPU. The library features specialized acceleration for existing workflows by redirecting standard Pandas dataframe calls and Polars query plans to a GPU backend. It also p

Chainer is an open-source deep learning framework built around define-by-run automatic differentiation, where computation graphs are constructed dynamically during forward execution. This imperative approach allows networks to be built using standard Python control flow, with gradients computed automatically through reverse-mode differentiation on the dynamically recorded graph. The framework supports GPU acceleration through a NumPy-compatible array backend with CUDA and cuDNN support, and provides a pluggable device abstraction that lets users switch between CPU and GPU computation without c

AidLearning-Framework is an integrated development platform for building and deploying AI applications on ARM-based devices. It combines Android and Linux operating systems running simultaneously on a single device, providing a unified runtime environment for cross-system AI development. The platform includes hardware acceleration across CPU, GPU, and NPU, with a unified API that automatically selects the optimal compute backend for inference. The framework distinguishes itself by enabling Python-based AI projects to be packaged directly into Android APK files for installation on mobile devic

NVIDIA DALI is a GPU-accelerated data loading and preprocessing library designed for deep learning workflows. It constructs high-performance data pipelines that offload decoding, augmentation, and normalization to the GPU, eliminating CPU bottlenecks in training and inference. The library reads data from multiple storage formats and streams it directly into GPU memory, with support for multi-GPU execution to scale throughput across large-scale workloads. DALI distinguishes itself by enabling data pipelines to be built once and executed across multiple deep learning frameworks without code cha

SAHI is a sliced inference framework and computer vision pipeline designed to detect small objects in high-resolution images. It provides a system for dividing large images into overlapping patches to prevent the detail loss that typically occurs during standard model downscaling, alongside an image tiling utility and a COCO dataset toolkit. The project distinguishes itself by offering a model-agnostic prediction wrapper that standardizes different machine learning frameworks into a unified interface. This allows it to implement sliced inference and object detection across various model backe

Thrust is a C++ parallel algorithms library that provides a suite of standard-library-inspired interfaces for execution on multi-core and accelerator hardware. It serves as a CUDA-accelerated data library and a generic parallel programming interface designed to enable high-performance data processing across GPUs and CPUs. The project implements a portable abstraction layer that allows for heterogeneous computing workflows, enabling the same core algorithm logic to run on different hardware accelerators. This is achieved through a generic programming policy design and a backend-agnostic execut

Deep Java Library is a Java deep learning framework and JVM model inference engine. It provides a high-level API for building and deploying deep learning models within the Java ecosystem, acting as a cross-platform runtime for executing models across CPUs, GPUs, and mobile devices. The library is engine-agnostic, allowing users to switch between different deep learning engines such as PyTorch, TensorFlow, and MXNet while maintaining a single unified API. This enables the deployment of the same model across different backends without changing the application code. The framework supports the f

POT is an optimal transport library providing a collection of solvers for computing Wasserstein, Gromov-Wasserstein, and Fused Gromov-Wasserstein distances between probability distributions. It functions as a differentiable tensor framework that integrates with various tensor libraries to enable automatic differentiation and GPU acceleration. The project is distinguished by its ability to align data distributions across different metric spaces by comparing internal relational structures rather than coordinates. It implements mathematical optimization algorithms as differentiable layers, allow

CUDA.jl provides a programming interface for executing custom kernels and performing parallel array computing directly on NVIDIA graphics hardware using the Julia language. It serves as a framework for compiling and scheduling user-defined functions across multiple processing cores, enabling high-performance data processing and task synchronization. The library distinguishes itself through a custom compiler backend that translates high-level language functions into hardware-specific machine code. It manages complex hardware interactions through asynchronous stream scheduling, unified memory m