GIST-large-Embedding-v0 Open-source Text Embedding Model - Achieve Retrieval Query Encoding without Instructions

GIST Large Embedding V0

Developed by avsolatorio

A text embedding model fine-tuned based on BAAI/bge-large-en-v1.5, trained using the MEDI dataset and mined triplets from the MTEB classification task training set, capable of encoding retrieval queries directly without instructions.

Text Embedding

Safetensors

EnglishOpen Source License:MIT #Instruction-Free Embedding #Contrastive Learning Optimization #Cross-Task Generalization

Downloads 110.09k

Release Time : 2/14/2024

Model Overview

This model is primarily used for text embedding tasks, converting text into high-dimensional vector representations, suitable for scenarios such as information retrieval and semantic similarity computation.

Model Features

Instruction-Free

Directly encodes retrieval queries without the need for constructing prompt templates.

High Performance

Significant improvement in performance for most retrieval tasks.

Based on GISTEmbed Technology

Utilizes Guided In-batch Selection for Training Negatives technology to optimize embedding effectiveness.

Model Capabilities

Text Embedding

Semantic Similarity Computation

Information Retrieval

Use Cases

Information Retrieval

Document Retrieval

Used to retrieve documents semantically similar to the query.

Significant improvement in performance for most retrieval tasks.

Semantic Similarity Computation

Text Similarity Comparison

Computes the semantic similarity between two pieces of text.

🚀 GIST-large-Embedding-v0

GIST-large-Embedding-v0 is a model in the sentence-transformers library, which can be used for various natural language processing tasks such as sentence similarity, feature extraction, etc.

📚 Documentation

Model Information

Property	Details
Library Name	sentence-transformers
Pipeline Tag	sentence-similarity
Tags	feature-extraction, mteb, sentence-similarity, sentence-transformers

Model Results

The model named GIST-large-Embedding-v0 has been tested on multiple datasets with different tasks. Here are the detailed results:

Classification Tasks

MTEB AmazonCounterfactualClassification (en):
- Accuracy: 75.5820895522388
- AP: 38.32190121241783
- F1: 69.44777155231054
MTEB AmazonPolarityClassification:
- Accuracy: 93.40514999999998
- AP: 90.2011565132406
- F1: 93.39486246843605
MTEB AmazonReviewsClassification (en):
- Accuracy: 49.05999999999999
- F1: 48.58702718571088
MTEB Banking77Classification:
- Accuracy: 88.10064935064933
- F1: 88.08460758973867

Retrieval Tasks

MTEB ArguAna: Multiple metrics are provided, including map_at_1, map_at_10, etc. For example, map_at_1 is 38.407000000000004.
MTEB CQADupstackAndroidRetrieval: Various retrieval - related metrics are available, such as map_at_1 with a value of 33.440999999999995.
MTEB CQADupstackEnglishRetrieval: Metrics like map_at_1 (32.126) and others are reported.
MTEB CQADupstackGamingRetrieval: For instance, map_at_1 is 41.312.
MTEB CQADupstackGisRetrieval: map_at_1 has a value of 27.609.
MTEB CQADupstackMathematicaRetrieval: map_at_1 is 18.071.

Clustering Tasks

MTEB ArxivClusteringP2P: The v_measure is 48.49766333679089.
MTEB ArxivClusteringS2S: The v_measure is 42.57731111438094.
MTEB BiorxivClusteringP2P: The v_measure is 39.338228337929976.
MTEB BiorxivClusteringS2S: The v_measure is 36.179156232378226.

Reranking Task

MTEB AskUbuntuDupQuestions:
- MAP: 64.70120072857361
- MRR: 77.86714593501297

STS Task

MTEB BIOSSES: Multiple similarity - related metrics are reported, such as cos_sim_pearson with a value of 90.73821860690765.

📄 License

This project is licensed under the MIT 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