jinaai/jina-embeddings-v4 · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Python Library
Summary
A Python library for inverting text embeddings back into readable text sequences, based on research showing that deep representations like sentence embeddings can be decoded to reveal the original text.
Highlights
- Inverts text embeddings (e.g., sentence embeddings) back into reconstructed text sequences
- Based on papers 'Text Embeddings Reveal (Almost) As Much As Text' and 'Language Model Inversion'
- Uses HuggingFace transformers architecture with pretrained models available via load_pretrained_corrector
- Installable via pip and includes a Colab demo for quick experimentation
- Supports both single-guess and multi-step inversion for improved reconstruction quality
auto-generated
Context
Audience
NLP researchers, ML engineers, and security researchers interested in the privacy implications of text embeddings
DomainNatural Language Processing
Formatopen source library
Accessopen source, free on PyPI
Source Overview
utilities for decoding deep representations (like sentence embeddings) back to text - vec2text/vec2text
Topics
Related
HuggingFace TransformersSentence EmbeddingsEmbedding InversionText Embedding SecurityLanguage Model Inversion
