Awesome GitHub RepositoriesInference Batching

Jina is a cloud-native framework for building and deploying multimodal AI applications that process text, images, and audio across distributed microservices. It functions as an inference orchestrator and a distributed model gateway, providing a containerized stack to organize AI executors into operational pipelines. The system manages large language model workloads through token-streamed response delivery and dynamic batching to increase hardware throughput. It utilizes a protocol-agnostic communication layer to route data across different machine learning frameworks. The framework covers hi

Faster-Whisper is a high-performance implementation of the Whisper speech-to-text model designed for efficient audio transcription. It provides an end-to-end processing pipeline that converts spoken audio into written text while maintaining lower memory consumption and faster execution speeds than standard implementations. The project achieves its performance through a specialized inference engine that utilizes optimized kernels and weight quantization to reduce computational complexity. It supports large-scale operations by grouping audio segments into dynamic batches and filtering out non-s

FunASR is an automatic speech recognition toolkit and multilingual speech-to-text engine designed to convert spoken audio into written text across more than fifty languages. It provides a framework for speaker diarization, an OpenAI-compatible transcription API for local server hosting, and speech models compatible with the ONNX format. The project distinguishes itself by supporting high-performance inference on edge hardware via self-contained binaries and portable model exports. It incorporates specialized capabilities for natural speech generation with adjustable timbre and emotional expre

Clip-as-service is a deployable framework for generating multi-modal embeddings and executing neural searches. It provides a vector embedding server and a CLIP embedding API to convert images and text into shared vector representations via network interfaces. The system functions as a multi-modal ranking system and neural search engine, enabling the retrieval of images through text queries or the identification of matching text descriptions for images. It also includes a visual reasoning service used to analyze images and verify object presence, counts, and colors by comparing visual data aga

Triton Inference Server is a high-performance AI model inference server and multi-framework model runtime designed for deploying machine learning models across cloud, data center, and embedded edge infrastructure. It serves as an execution engine that allows for the concurrent running of models from various frameworks to optimize hardware utilization. The project features a dynamic batching inference engine that groups individual requests into larger batches to increase total processing throughput. It also provides a model ensemble pipeline, which enables the chaining of multiple models toget

Triton Inference Server is a high-performance server designed to deploy machine learning models from multiple frameworks across GPUs and CPUs. It functions as a hardware-accelerated inference engine and a gRPC inference gateway, providing a standardized communication layer for transmitting binary tensor data with low latency. The system acts as a multi-framework model orchestrator, allowing users to link multiple AI models into ensembles and scripts to create complex inference pipelines. It also serves as a model lifecycle manager, providing controls to load, unload, and monitor the performan

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

Moondream is a small-scale vision language model designed to reason across images to generate captions and answer natural language questions. It functions as an edge-optimized system capable of performing visual question answering, image captioning, and object detection. The project distinguishes itself through a lightweight architecture designed for local inference on embedded devices, workstations, and air-gapped hardware. It supports the execution of models on local GPUs and Apple Silicon to ensure data privacy and low latency. The system's capabilities include identifying precise object

This project is a platform for the deployment of open source large language and multimodal models. It provides a unified interface to serve text, image, and speech models across local or cloud hardware. The system enables distributed AI inference by orchestrating model workloads across multiple nodes and devices. It includes a unified API adapter layer to standardize inputs and outputs, as well as tools for multimodal chat and structural image generation. The platform covers a broad capability surface including request batching for throughput optimization, dynamic model loading, and integrat

PowerInfer is a high-performance local large language model inference engine and sparse inference framework. It provides a runtime for executing models on consumer-grade hardware, utilizing a GPU acceleration backend to optimize tensor operations for graphics processors. The system distinguishes itself through a sparse inference framework that increases generation speed by skipping computations based on activation sparsity in model weights. It includes a GGUF model converter for transforming weights and metadata into a unified binary format, as well as an OpenAI API compatible server for inte

BentoML is a machine learning model serving framework and GPU-accelerated inference server designed to package, deploy, and scale AI models as production-ready REST APIs. It functions as an AI model lifecycle manager and an inference graph orchestrator, enabling the chaining of multiple models and custom logic into complex pipelines for advanced task sequences. The framework distinguishes itself through a dynamic batching engine that optimizes GPU throughput and an artifact-based packaging system that bundles model weights and dependencies into immutable archives for consistent deployment. It

Sana is a framework for high-resolution image and video synthesis based on a linear diffusion transformer. It provides a toolkit for the training, fine-tuning, and execution of text-to-image and text-to-video models, as well as a video generative world model capable of simulating physical environments with precise spatial control. The project is distinguished by its use of linear complexity layers to handle high resolutions and its support for long-form, minute-length video generation in real time. It implements a two-stage inference paradigm that separates structural generation from visual t

Anomalib is a PyTorch-based library for visual anomaly detection, offering a modular framework, a comprehensive model zoo, and a benchmarking suite designed for industrial defect detection. It provides a wide range of algorithms—including generative, discriminative, teacher-student, and vision-language approaches—that support unsupervised, few-shot, and zero-shot settings. The library enables deployment through model export to ONNX and OpenVINO for edge devices, and includes a no-code web application for training and inference. It also features a command-line interface for orchestrating multi

RF-DETR is a Python library for training and deploying object detection, instance segmentation, and keypoint detection models built on a vision transformer architecture. It provides a unified command-line interface and Python API for the full workflow, from fine-tuning pretrained checkpoints on custom datasets to running inference on images, video files, and live camera streams. The project supports training on datasets in COCO or YOLO format, with automatic format detection and configurable augmentation pipelines. Models can be exported to ONNX, TFLite, or TensorRT for deployment across edge

SAHI est un framework d'inférence par découpage (sliced inference) et un pipeline de vision par ordinateur conçu pour détecter de petits objets dans des images haute résolution. Il fournit un système pour diviser les grandes images en patchs chevauchants afin d'éviter la perte de détails qui se produit généralement lors de la réduction d'échelle standard des modèles, aux côtés d'un utilitaire de tuilage d'image et d'une boîte à outils de jeu de données COCO. Le projet se distingue en offrant un wrapper de prédiction agnostique au modèle qui standardise différents frameworks d'apprentissage automatique dans une interface unifiée. Cela lui permet d'implémenter l'inférence par découpage et la détection d'objets à travers divers backends de modèles tout en maintenant un format de sortie cohérent. Au-delà de l'inférence, le framework couvre la gestion de jeux de données pour les formats COCO et YOLO, incluant des outils pour le découpage d'images annotées, le remapping de catégories et la fusion de jeux de données. Il inclut également une suite pour l'évaluation et la surveillance des performances des modèles, présentant le calcul de métriques pour la précision et le rappel, l'analyse des erreurs de détection et la visualisation des résultats. La boîte à outils est accessible via une interface en ligne de commande pour automatiser les workflows d'inférence à travers les répertoires d'images et les flux vidéo.

Daft is a distributed dataframe library and multimodal data processor designed to handle large-scale structured and unstructured data. It functions as a vectorized execution engine that processes tables alongside images, audio, and video, utilizing a unified schema to manage diverse data types. The project distinguishes itself by combining distributed data engineering with large-scale AI inference. It provides an AI data pipeline for batch-optimizing model prompts and generating high-dimensional text embeddings, while utilizing zero-copy memory sharing to execute custom Python functions witho

Qwen3-Omni is an omni-modal large language model designed to process and generate text, audio, images, and video within a single unified neural architecture. It functions as a real-time voice assistant and multimodal AI agent capable of reasoning across different media types and executing external tool-calling functions via APIs. The system supports low-latency conversational AI through autoregressive token streaming and natural turn-taking. It enables multilingual speech translation and generation across dozens of languages, featuring customizable speaker profiles and tones. The model's cap

VLMEvalKit is a vision-language model evaluation framework and inference engine designed to run standardized benchmarks and measure model accuracy across diverse visual datasets. It serves as a multimodal model benchmark and performance toolkit for calculating metrics and comparing model responses. The toolkit includes a specialized visual reasoning evaluator that uses adversarial samples to distinguish actual image understanding from reliance on language patterns. It also provides capabilities for image generation evaluation, testing a model's ability to create or modify visuals based on tex