INSTRUCTOR-XL Open-source General Embedding Model - Instruction-Controlled Generation of Multi-type Text Embedding Vectors

Instructor Xl

Developed by hku-nlp

INSTRUCTOR-XL is a general-purpose embedding model capable of generating domain-specific and task-aware text embeddings through instruction control, suitable for various text types (e.g., titles, sentences, documents).

Text Embedding

Transformers

EnglishOpen Source License:Apache-2.0 #Instruction-Controlled Embedding #Domain-Specific Vectors #Zero-Shot Task Adaptation

Downloads 36

Release Time : 12/17/2022

Model Overview

This model can map arbitrary texts to fixed-length vectors during the testing phase without additional training. Through instruction control, the generated embeddings are domain-specific (e.g., science, finance) and task-aware (e.g., classification, information retrieval).

Model Features

Instruction-Controlled Embedding

Generates domain-specific and task-aware embeddings through instruction control, adapting to different scenarios without additional training.

General-Purpose Text Processing

Supports processing various text types, including titles, sentences, and documents, to generate fixed-length vector representations.

Domain Adaptability

Capable of generating optimized embeddings for different domains (e.g., science, finance).

Model Capabilities

Text embedding generation

Sentence similarity computation

Domain-specific feature extraction

Use Cases

Academic Research

Scientific Literature Title Embedding

Generates domain-specific embeddings for scientific literature titles, used for literature retrieval or classification.

Improves retrieval accuracy for scientific literature relevance

Financial Analysis

Financial News Analysis

Generates embeddings for financial news, used for market sentiment analysis or event detection.

Enhances semantic understanding of financial texts

🚀 hku-nlp/instructor-xl

A general embedding model that maps any text to a fixed - length vector without further training at test time. It generates domain - specific and task - aware embeddings with instructions.

🚀 Quick Start

This is a general embedding model. It can map any piece of text (such as a title, a sentence, a document, etc.) to a fixed - length vector during testing without further training. By using instructions, the generated embeddings are domain - specific (e.g., specialized for science, finance, etc.) and task - aware (e.g., customized for classification, information retrieval, etc.). The model is easy to use with the sentence - transformer library.

📦 Installation

git clone https://github.com/HKUNLP/instructor-embedding
cd sentence-transformers
pip install -e .

💻 Usage Examples

Basic Usage

Then you can use the model like this to calculate domain - specific and task - aware embeddings:

from sentence_transformers import SentenceTransformer
sentence = "3D ActionSLAM: wearable person tracking in multi-floor environments"
instruction = "Represent the Science title; Input:"
model = SentenceTransformer('hku-nlp/instructor-xl')
embeddings = model.encode([[instruction,sentence,0]])
print(embeddings)

Advanced Usage

You can further use the model to compute similarities between two groups of sentences, with customized embeddings.

from sklearn.metrics.pairwise import cosine_similarity
sentences_a = [['Represent the Science sentence; Input: ','Parton energy loss in QCD matter',0], 
               ['Represent the Financial statement; Input: ','The Federal Reserve on Wednesday raised its benchmark interest rate.',0]]
sentences_b = [['Represent the Science sentence; Input: ','The Chiral Phase Transition in Dissipative Dynamics', 0],
               ['Represent the Financial statement; Input: ','The funds rose less than 0.5 per cent on Friday',0]]
embeddings_a = model.encode(sentences_a)
embeddings_b = model.encode(sentences_b)
similarities = cosine_similarity(embeddings_a,embeddings_b)
print(similarities)

📄 License

This project is licensed under the Apache - 2.0 license.

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご