5 dépôts
Inference techniques designed to run massive models within limited system or video memory.
Distinct from Large Language Model Optimization: Focuses on memory-constrained execution for massive models, distinct from general speed-focused optimization.
Explore 5 awesome GitHub repositories matching artificial intelligence & ml · Memory-Constrained Inference. Refine with filters or upvote what's useful.
DiffSynth-Studio is a comprehensive platform for the lifecycle management of generative diffusion models, providing a unified environment for inference, fine-tuning, and training. It utilizes a modular pipeline architecture and a standardized abstraction layer to support consistent workflows across diverse model configurations for image and video generation. The platform distinguishes itself through a memory-optimized inference engine that dynamically manages resources to facilitate high-resolution generation on constrained hardware. It also integrates specialized training capabilities, inclu
Features a memory-optimized inference engine that dynamically manages resources to enable high-resolution generation on constrained hardware.
Airllm is a framework designed to execute and fine-tune large language models on consumer-grade hardware. By employing layer-wise model decomposition and memory-efficient loading techniques, the engine enables the operation of massive models that would otherwise exceed available system or video memory. The project distinguishes itself through a suite of optimization strategies that balance memory footprint with performance. It utilizes block-wise weight quantization and asynchronous layer prefetching to reduce resource consumption and hide data transfer latency. Additionally, the framework su
Runs massive language models on consumer hardware by optimizing memory usage to fit within the constraints of available graphics or system memory.
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
Limits the total memory used during inference to enable the execution of large models on low-RAM devices.
StoryDiffusion is a generative AI system designed for consistent character image and video generation. It utilizes a pluggable cross-attention module to inject shared character representations into pretrained diffusion models, allowing for visual identity stability across multiple images and scenes without retraining the base model. The project features a video generation pipeline that produces temporally coherent sequences from text prompts or condition images. It employs a latent space motion interpolator to predict intermediate frames and semantic motion, enabling long-range video generati
Implements techniques to run large generative models within the memory constraints of consumer GPUs.
FedML est une bibliothèque d'apprentissage automatique distribué, un framework d'apprentissage fédéré et un orchestrateur de charges de travail GPU. Il fournit les composants système essentiels pour exécuter l'entraînement et le fine-tuning de modèles à grande échelle sur des clusters GPU multi-cloud, sur site et décentralisés, tout en offrant un moteur dédié pour le déploiement de modèles évolutifs et un gestionnaire de pipeline MLOps pour la gestion du cycle de vie de bout en bout. La plateforme se distingue en permettant un apprentissage fédéré préservant la vie privée sur des appareils périphériques décentralisés et des silos organisationnels, en conservant les données brutes sur le matériel local. Elle propose également une place de marché de calcul par mise en commun de ressources qui permet aux utilisateurs de contribuer leur capacité GPU inutilisée à un pool partagé pour l'exécution de tâches distribuées. Le système couvre un large éventail de capacités, notamment l'orchestration GPU multi-cloud, la gestion automatisée des pipelines d'apprentissage automatique et le déploiement d'IA en périphérie pour les appareils IoT et les smartphones. Il intègre en outre des outils pour le fine-tuning de modèles fondamentaux, le déploiement d'inférence à faible latence et le suivi des expériences d'entraînement avec profilage des performances matérielles. Les utilisateurs peuvent lancer et planifier des charges de travail à l'aide d'une interface en ligne de commande et de fichiers de configuration déclaratifs.
Implements specialized memory management and optimization to run large models on hardware with limited VRAM.