xenova/transformers.js

Features

• Browser-Based Model Runtimes - Provides a JavaScript runtime for executing pretrained transformer models directly in the browser using ONNX Runtime.
• Client-Side Inference - Runs pretrained transformer models directly in the web browser for local inference.
• ONNX Runtime Inference - Integrates the ONNX Runtime to execute pre-trained models directly in the browser.
• Multimodal AI Toolkits - Provides a unified toolkit for processing and generating text, image, and audio data using transformer architectures.
• Natural Language Processing - Performs text analysis, translation, and summarization directly on the user's device.
• Web-Based Machine Learning - Serves as a comprehensive framework for integrating and executing various machine learning models directly within web browsers.
• Multimodal Pre-Processing - Implements structured tokenizers and feature extractors to convert raw multimodal input into numerical tensors.
• Privacy-Preserving Machine Learning - Ensures sensitive data remains on the user's device by performing model inference locally.
• Audio Transcriptions - Converts spoken language from audio files into written text using transformer models.
• Computer Vision - Extracts labels, object locations, and depth maps using specialized vision models.
• Object Detection - Identifies and locates specific objects within a visual frame using bounding boxes.
• Image Segmentation - Maps individual pixels to specific objects to create boundaries for semantic analysis.
• Text Embedding Extraction - Converts raw text into numerical vector representations to preserve semantic information for downstream processing.
• GPU Acceleration - Uses Web-GPU to offload tensor computations from the CPU to the graphics processor.
• Image Classification - Assigns descriptive labels to images, including zero-shot classification for unknown categories.
• Audio Model Execution - Transcribes speech and synthesizes voice using ML models running locally in the browser.
• Web-Based Computer Vision - Enables computer vision tasks like object detection and image segmentation directly within web browsers.
• Hardware Acceleration - Offloads tensor computations to the GPU via Web-GPU to increase model execution speed.
• Text Classification - Assigns predefined categories to text using sentiment analysis or zero-shot classification.
• Machine Translation - Converts text from one natural language to another using pretrained transformer models.
• Model Execution Engines - Combines model logic with input and output transformations to execute pretrained transformer models.
• Model Quantization - Reduces memory footprint and bandwidth by utilizing quantized data types for models in resource-constrained environments.
• Multimodal Model Runners - Loads and runs models that can process and classify data across text, image, and audio modalities.
• Text Summarization - Generates concise summaries of provided text or documents using language models.
• Weight Quantization - Supports loading models with lower precision data types to reduce memory and bandwidth usage.
• Question Answering - Implements semantic analysis to extract answers and determine similarity between pieces of text.
• Text Sequence Generation - Produces new text by predicting the next word in a sequence or converting one sequence into another.
• Speech Synthesis - Converts written text into natural sounding spoken audio using AI synthesis.
• AI Task Wrappers - Provides specialized pipeline classes that wrap model logic for specific AI tasks.
• Image Feature Extraction - Converts raw pixel data into numerical representations for downstream analytical tasks.
• Large Language Models - Machine learning models running in the browser.
• Transformer Implementations - Library for running transformer models directly in the browser.
• More to explore - Runs machine learning models in the browser.

Historial de estrellas