21 dépôts
Tools for transforming trained model weights from legacy formats into native array structures.
Distinct from Model Parameter Management: Focuses on the conversion of weight files specifically, distinct from general parameter lifecycle management.
Explore 21 awesome GitHub repositories matching artificial intelligence & ml · Weight Conversion Utilities. Refine with filters or upvote what's useful.
This project is a PyTorch transformer model library and pre-trained model framework. It serves as a deep learning model hub and multimodal inference engine, providing a centralized system for loading, executing, and fine-tuning state-of-the-art model checkpoints. The library focuses on multimodal machine learning, enabling predictions across text, vision, and audio data. It provides specialized capabilities for model framework interoperability, allowing the conversion of weights and definitions between different deep learning libraries. The platform covers the full model lifecycle, including
Includes utilities to transform model weights between different deep learning framework formats for interoperability.
VoxCPM is a multilingual speech synthesis system and text-to-speech inference server. It functions as an AI voice cloning tool and a synthetic voice designer, capable of generating natural speech across global languages and regional dialects using a GPU-accelerated audio generator. The project features a speech model fine-tuning framework that supports both full parameter updates and low-rank adaptation for customizing voice characteristics. It enables high-fidelity voice cloning from reference audio, including cross-lingual voice transfer and acoustic environment mimicry, as well as the crea
Includes a utility to transform weights from standard hub formats into specialized runtime formats for optimized execution.
This project is a deep learning framework designed for constructing, training, and deploying neural networks across diverse hardware environments. It functions as a high-performance tensor computation library that provides both imperative and symbolic programming interfaces, allowing developers to balance flexible, step-by-step model building with the efficiency of compiled computation graphs. The framework distinguishes itself through a hybrid execution engine that integrates declarative graph compilation with imperative runtime logic. It supports scalable, distributed training across multip
Transforms trained model weights and data files from legacy formats into native array structures.
This project is a comprehensive toolkit for adapting large language models to the Chinese language, providing a specialized framework for fine-tuning, inference, and local deployment. It serves as a coordinated suite for language-specific adaptation, including tools for expanding tokenizers and implementing retrieval-augmented generation. The project distinguishes itself through a complete pipeline for model adaptation, featuring multilingual tokenizer expansion and a fine-tuning framework that supports instruction-based supervised training and adapter merging. It also includes a dedicated de
Transforms original model weights into formats compatible with standard machine learning libraries.
This project is a collection of implementation guides, recipes, and developer resources for building applications with Llama models. It serves as a comprehensive kit for developing autonomous agents, establishing retrieval-augmented generation systems, and executing model fine-tuning. The resource provides specific patterns for multimodal workflows that process text, images, and audio. It includes specialized guidance on adapting pre-trained model weights for targeted tasks and implementing tool-calling orchestration to connect models with external APIs and functions. The codebase covers a b
Includes utilities for transforming model checkpoints from proprietary formats into standardized open-source library formats.
Megatron-LM is a distributed transformer training library and large language model training framework designed to scale models across thousands of GPUs. It functions as a GPU-optimized deep learning toolkit and a scaling engine for mixture-of-experts architectures, enabling the training of models with hundreds of billions of parameters. The project implements multi-dimensional model parallelism, combining tensor, pipeline, data, expert, and context-based workload distribution. It specifically optimizes mixture-of-experts architectures through integrated memory and communication improvements t
Provides utilities to translate model weight files between different framework formats to ensure interoperability.
This project is a comprehensive toolkit designed for the full lifecycle management of large language and multimodal models. It functions as a unified orchestrator that handles the entire development process, ranging from dataset preparation and supervised fine-tuning to advanced reinforcement learning alignment and production-ready inference deployment. The platform distinguishes itself through a specialized reinforcement learning library that supports complex optimization algorithms, including group relative policy optimization and leave-one-out techniques, to improve model instruction-follo
The platform transfers model weights between different formats and prepares datasets for efficient training through pre-tokenized caching to streamline the development workflow.
CogVideo is a video generation framework and large language model architecture designed for synthesizing high-resolution video clips from natural language descriptions and images. It functions as a text-to-video and image-to-video generator, while also providing a model for video captioning to analyze visual content into descriptive text summaries. The system supports animating static images into motion sequences and transforming series of images into video based on prompts. It includes capabilities for extending the length of generated video clips to create longer sequences of motion. The f
Includes utilities for transforming model weights between different formats for environment compatibility.
OpenPi is a vision-language-action robot control framework designed to generate physical control actions for robotic systems. It functions as a distributed robot model trainer, a model format converter, and a robot action streaming server. The framework provides tools for transforming model checkpoints between different framework formats to ensure interoperability across various development environments. It also includes a server that uses websocket connections to stream model-generated control actions from remote inference servers to physical robot hardware in real-time. The system supports
Provides utilities for transforming model checkpoints between different tensor formats for cross-environment compatibility.
PowerInfer is an inference engine and serving framework designed to run large language models on local hardware. It combines a hybrid CPU-GPU offloader, a quantization tool, and a sparse model optimizer to enable the execution of high-parameter models on consumer-grade devices. The system distinguishes itself through neuron-activation-based offloading, using a predictor model to preload frequent neurons into VRAM while keeping rare neurons in system memory. This hybrid execution model balances workloads between the GPU and CPU based on input patterns to optimize memory access and increase tok
Transforms standard model weights into specialized sparse structures optimized for fast local retrieval and inference.
This project is a toolkit for fine-tuning and managing text-to-image diffusion models. It focuses on low-rank adaptation to create small, portable weight files that customize model styles and behaviors without modifying the entire base model. The project provides specialized utilities for model distillation using singular value decomposition to extract adapters from fully trained models, as well as tools for blending and merging multiple adapters through weight interpolation. It includes capabilities for subject inversion and pivotal tuning to increase the visual fidelity of specific identiti
Provides utilities for transforming trained model weights between different storage formats to ensure cross-engine compatibility.
This project is an object detection framework implementing the YOLOv3 architecture using Keras and TensorFlow. It functions as a deep learning vision model and computer vision toolset designed to locate and classify multiple entities within images and video streams using bounding boxes. The system includes a multi-GPU inference engine to distribute computational loads across several graphics processing units. It also provides a pipeline for creating custom object detectors by retraining pre-trained weights on annotated datasets to recognize user-defined object classes. The framework covers m
Transforms trained network parameters between frameworks for deployment across different TensorFlow environments.
Ce projet est un framework de transformer basé sur JAX et un entraîneur de modèles de langage large conçu pour construire et entraîner des modèles distribués sur des accélérateurs matériels TPU. Il fournit un système pour le pré-entraînement et le fine-tuning de modèles autorégressifs en divisant les poids et les calculs sur un maillage de périphériques pour réduire la surcharge mémoire et augmenter la vitesse de traitement. Le framework inclut un orchestrateur de calcul TPU pour provisionner les ressources et automatiser l'installation des dépendances sur des nœuds distribués distants. Il dispose également d'un convertisseur de poids de modèle capable de transformer et de re-sharder les checkpoints entre différentes configurations matérielles et précisions numériques. Le projet couvre des capacités plus larges, notamment la gestion de checkpoints shardés pour le stockage cloud, le chargement de données par flux avec restauration d'état et la génération de texte basée sur le noyau pour l'inférence. Il prend en charge l'accélération matérielle compilée XLA pour les clusters TPU et GPU et fournit des outils de benchmarking de performance sur des tâches linguistiques standardisées.
Implements utilities to transform model weights into native array structures for seamless loading across different libraries.
bert4keras est une réimplémentation légère de l'architecture transformer BERT pour le framework de deep learning Keras. Il sert de boîte à outils de traitement du langage naturel et de bibliothèque de modèles transformer utilisée pour la classification de texte, l'étiquetage de séquences et l'extraction d'embeddings sémantiques. Le framework inclut un système de modèle sequence-to-sequence pour la réponse aux questions et la génération de texte, ainsi qu'un serveur d'inférence de modèle pour déployer des transformers entraînés en tant qu'API web pour des prédictions en temps réel. Les capacités couvrent un large éventail de tâches de compréhension du langage naturel, incluant la compréhension de lecture, l'extraction de relations et le traitement de textes longs. La bibliothèque fournit des outils pour le pré-entraînement et le fine-tuning de modèles de langage, aux côtés de techniques d'optimisation telles que la réduction de paramètres, l'entraînement contradictoire pour la robustesse et la configuration du taux d'apprentissage par couche. Le projet inclut un chargeur de conversion de poids pour transformer les poids pré-entraînés depuis des formats externes en structures Keras compatibles.
Ships a weight-conversion loader to transform pre-trained weights from external formats into Keras structures.
stable-diffusion.cpp is a high-performance C++ inference engine designed for generating images and video from text prompts using Stable Diffusion models. It functions as a latent diffusion model runtime and a lightweight machine learning framework that enables local diffusion model execution on consumer hardware. The project distinguishes itself as a CPU-based image generator capable of running without a dedicated GPU. It employs a specialized C++ tensor backend and cross-backend hardware abstraction to dispatch compute tasks across different processor instruction sets and graphics APIs. The
Transforms machine learning weights between different storage formats for cross-runtime compatibility.
OpenVLA is a vision-language-action model and framework designed for general-purpose robotic manipulation. It provides a robotic policy training framework and a control inference engine that map visual and textual inputs to robotic control actions, enabling zero-shot instruction following on hardware. The project includes a robotics dataset pipeline for standardizing diverse trajectory data and managing dataset mixtures. It supports large-scale model training through distributed GPU compute and sharded data parallelism, alongside parameter-efficient adaptation for fine-tuning models to new ta
Transforms weights from custom formats into standardized structures for HuggingFace library loading.
whisper-jax est une implémentation haute performance du modèle de reconnaissance vocale automatique Whisper réécrite en utilisant le framework JAX. Il est conçu pour une inférence accélérée et utilise la compilation XLA pour optimiser l'exécution du modèle sur des accélérateurs matériels. Le projet se concentre sur la transcription optimisée pour TPU pour atteindre un débit et une vitesse élevés. Il inclut un pipeline de traduction de poids qui convertit les paramètres de modèles pré-entraînés de PyTorch en tableaux compatibles JAX. Le système prend en charge la transcription de l'audio en texte, la traduction de la parole dans plusieurs langues et la génération d'horodatages audio. Il permet le traitement audio par lots et met à l'échelle les performances grâce au traitement par lots parallèle aux données et au partitionnement de tenseurs parallèle au modèle. Le projet fournit une méthode pour déployer le modèle de transcription en tant qu'endpoint d'inférence distant avec une interface web.
Includes a pipeline for transforming pre-trained PyTorch weights into native JAX array structures.
Ce projet est une implémentation PyTorch d'un réseau antagoniste génératif (GAN) conçu pour la synthèse d'images haute résolution. Il fournit un modèle de synthèse d'images qui produit des images réalistes à partir de vecteurs latents et de conditions de classe apprises, soutenu par un outil de projection dans l'espace latent pour trouver des vecteurs numériques représentant des images cibles spécifiques. L'implémentation propose l'augmentation adaptative du discriminateur, une technique d'entraînement utilisée pour éviter le surapprentissage du discriminateur lors de l'entraînement sur des jeux de données limités. Elle inclut également une suite d'évaluation de modèles génératifs fournissant des métriques quantitatives pour mesurer la fidélité et la diversité des images synthétisées. La bibliothèque couvre des workflows génératifs plus larges, notamment le mélange de styles d'images, la projection image-vers-latent et l'entraînement de réseaux génératifs sur des jeux de données personnalisés. Elle fournit des utilitaires pour la préparation de jeux de données d'images et la conversion de poids de réseaux.
Includes utilities for converting neural network weights from legacy formats into compatible native structures.
Ce projet est une implémentation PyTorch de l'architecture Faster R-CNN pour la détection d'objets. Il fournit un framework pour identifier plusieurs classes d'objets et leurs boîtes englobantes (bounding boxes) correspondantes dans des images à l'aide d'un système d'apprentissage profond. L'implémentation inclut un pipeline d'entraînement pour optimiser les modèles sur des jeux de données personnalisés et un utilitaire pour convertir des poids pré-entraînés depuis des formats externes vers une structure compatible pour l'initialisation du modèle. Le système couvre un pipeline de détection en deux étapes comprenant un réseau de proposition de régions et une couche de pooling ROI. Il intègre des fonctions de perte multi-tâches et une régression de boîte englobante basée sur des ancres pour affiner les emplacements des objets. Le projet inclut des outils pour la visualisation en temps réel de la perte d'entraînement et de la précision des prédictions afin de surveiller les performances du modèle.
Ships utilities for converting pretrained weights from external formats into native array structures.
picoGPT is a lightweight, low-level runtime environment and inference engine designed to load pre-trained checkpoints and execute generative transformer model inference. It provides a minimal implementation of the generative pre-trained transformer architecture to facilitate local language model execution. The project includes a C++ machine learning library for converting model parameters and executing greedy token generation without heavy external dependencies. It handles remote asset synchronization by downloading pre-trained weights, hyperparameters, and vocabulary files from remote server
Transforms trained model weights from checkpoint formats into native numerical structures for inference.