multilingual e5 large instruct

提供商intfloat
分类feature-extraction
许可证mit
下载量3.1K
星标7

简介

Multilingual E5 Large Instruct 是一款高性能的多语言文本向量化模型，由 intfloat 团队开发。与传统的 Embedding 模型不同，它引入了指令微调机制，用户可以通过在输入前添加特定的任务指令（如“检索相关文档”），引导模型在不同场景下生成更精准的向量表示。它在跨语言检索和语义匹配上表现强劲，能够将中文、英文等多种语言映射到同一向量空间，非常适合用于构建 RAG（检索增强生成）知识库，是目前替代开源 BGE 或 OpenAI embedding 的优质选择，上手门槛低，兼容主流向量数据库。

核心亮点

  • 支持指令微调，可根据具体任务灵活调整检索效果
  • 强大的多语言对齐能力，支持高效的跨语言语义搜索
  • RAG 架构的核心组件，大幅提升知识库检索准确率
  • 采用 MIT 协议开源，企业级部署无压力且成本低

使用方法

安装依赖
# 安装 Hugging Face transformers
pip install transformers torch
SDK 使用
# 使用 transformers 加载模型
from transformers import AutoModel, AutoTokenizer

model = AutoModel.from_pretrained("intfloat/multilingual-e5-large-instruct")
tokenizer = AutoTokenizer.from_pretrained("intfloat/multilingual-e5-large-instruct")

Hugging Face 下载

我们推荐使用命令行或者 Hugging Face Hub SDK 来进行模型的下载。

操作指引：在下载前，请先通过如下命令安装 huggingface_hub：

操作指引 
pip install -U huggingface_hub

命令行下载

下载完整模型库

下载完整模型库 
huggingface-cli download intfloat/multilingual-e5-large-instruct

下载单个文件到指定本地文件夹（以下载 config.json 到当前路径下 ./dir 目录为例）

下载单个文件到指定本地文件夹（以下载 config.json 到当前路径下 ./dir 目录为例） 
huggingface-cli download intfloat/multilingual-e5-large-instruct config.json --local-dir ./dir

更多命令行下载选项，可参见官方文档

SDK 下载

SDK 下载 
# 模型下载
from huggingface_hub import snapshot_download
model_dir = snapshot_download('intfloat/multilingual-e5-large-instruct')

Git 下载

请确保 lfs 已经被正确安装

Git 下载 
git lfs install
git clone https://huggingface.co/intfloat/multilingual-e5-large-instruct

如果您希望跳过 lfs 大文件下载，可以使用如下命令

跳过 LFS 
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/intfloat/multilingual-e5-large-instruct

模型文件托管在 Hugging Face Hub，使用 HF CLI / SDK / Git 直接下载，不经过本站。

PyTorch / Transformers 使用

安装 Transformers

安装 Transformers 
pip install -U transformers torch

模型加载和推理

模型加载和推理 
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained('intfloat/multilingual-e5-large-instruct')
tokenizer = AutoTokenizer.from_pretrained('intfloat/multilingual-e5-large-instruct')

模型下载

我们推荐使用命令行或者 ModelScope SDK 来进行模型的下载。

操作指引：在下载前，请先通过如下命令安装 ModelScope：

操作指引 
pip install modelscope

命令行下载

下载完整模型库

下载完整模型库 
modelscope download --model intfloat/multilingual-e5-large-instruct

下载单个文件到指定本地文件夹（以下载 README.md 到当前路径下 dir 目录为例）

下载单个文件到指定本地文件夹（以下载 README.md 到当前路径下 dir 目录为例） 
modelscope download --model intfloat/multilingual-e5-large-instruct README.md --local_dir ./dir

更多更丰富的命令行下载选项，可参见具体文档

SDK 下载

SDK 下载 
# 模型下载
from modelscope import snapshot_download
model_dir = snapshot_download('intfloat/multilingual-e5-large-instruct')

Git 下载

请确保 lfs 已经被正确安装

Git 下载 
git lfs install
git clone https://www.modelscope.cn/intfloat/multilingual-e5-large-instruct.git

如果您希望跳过 lfs 大文件下载，可以使用如下命令

跳过 LFS 
GIT_LFS_SKIP_SMUDGE=1 git clone https://www.modelscope.cn/intfloat/multilingual-e5-large-instruct.git

ModelScope 模型页直接下载模型文件；无需将模型文件放在本站服务器。

Notebook 快速开发

下载并安装 ModelScope library

下载并安装 ModelScope library 
pip install "modelscope[audio,cv,nlp,multi-modal,science]" -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html

模型加载和推理

模型加载和推理 
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks

p = pipeline('text-generation', 'intfloat/multilingual-e5-large-instruct')

完整文档

来源: HuggingFace

---
tags:

  • mteb

  • sentence-transformers

  • transformers

model-index:
  • name: multilingual-e5-large-instruct

results:
- task:
type: Classification
dataset:
type: mteb/amazon_counterfactual
name: MTEB AmazonCounterfactualClassification (en)
config: en
split: test
revision: e8379541af4e31359cca9fbcf4b00f2671dba205
metrics:
- type: accuracy
value: 76.23880597014924
- type: ap
value: 39.07351965022687
- type: f1
value: 70.04836733862683
- task:
type: Classification
dataset:
type: mteb/amazon_counterfactual
name: MTEB AmazonCounterfactualClassification (de)
config: de
split: test
revision: e8379541af4e31359cca9fbcf4b00f2671dba205
metrics:
- type: accuracy
value: 66.71306209850107
- type: ap
value: 79.01499914759529
- type: f1
value: 64.81951817560703
- task:
type: Classification
dataset:
type: mteb/amazon_counterfactual
name: MTEB AmazonCounterfactualClassification (en-ext)
config: en-ext
split: test
revision: e8379541af4e31359cca9fbcf4b00f2671dba205
metrics:
- type: accuracy
value: 73.85307346326837
- type: ap
value: 22.447519885878737
- type: f1
value: 61.0162730745633
- task:
type: Classification
dataset:
type: mteb/amazon_counterfactual
name: MTEB AmazonCounterfactualClassification (ja)
config: ja
split: test
revision: e8379541af4e31359cca9fbcf4b00f2671dba205
metrics:
- type: accuracy
value: 76.04925053533191
- type: ap
value: 23.44983217128922
- type: f1
value: 62.5723230907759
- task:
type: Classification
dataset:
type: mteb/amazon_polarity
name: MTEB AmazonPolarityClassification
config: default
split: test
revision: e2d317d38cd51312af73b3d32a06d1a08b442046
metrics:
- type: accuracy
value: 96.28742500000001
- type: ap
value: 94.8449918887462
- type: f1
value: 96.28680923610432
- task:
type: Classification
dataset:
type: mteb/amazon_reviews_multi
name: MTEB AmazonReviewsClassification (en)
config: en
split: test
revision: 1399c76144fd37290681b995c656ef9b2e06e26d
metrics:
- type: accuracy
value: 56.716
- type: f1
value: 55.76510398266401
- task:
type: Classification
dataset:
type: mteb/amazon_reviews_multi
name: MTEB AmazonReviewsClassification (de)
config: de
split: test
revision: 1399c76144fd37290681b995c656ef9b2e06e26d
metrics:
- type: accuracy
value: 52.99999999999999
- type: f1
value: 52.00829994765178
- task:
type: Classification
dataset:
type: mteb/amazon_reviews_multi
name: MTEB AmazonReviewsClassification (es)
config: es
split: test
revision: 1399c76144fd37290681b995c656ef9b2e06e26d
metrics:
- type: accuracy
value: 48.806000000000004
- type: f1
value: 48.082345914983634
- task:
type: Classification
dataset:
type: mteb/amazon_reviews_multi
name: MTEB AmazonReviewsClassification (fr)
config: fr
split: test
revision: 1399c76144fd37290681b995c656ef9b2e06e26d
metrics:
- type: accuracy
value: 48.507999999999996
- type: f1
value: 47.68752844642045
- task:
type: Classification
dataset:
type: mteb/amazon_reviews_multi
name: MTEB AmazonReviewsClassification (ja)
config: ja
split: test
revision: 1399c76144fd37290681b995c656ef9b2e06e26d
metrics:
- type: accuracy
value: 47.709999999999994
- type: f1
value: 47.05870376637181
- task:
type: Classification
dataset:
type: mteb/amazon_reviews_multi
name: MTEB AmazonReviewsClassification (zh)
config: zh
split: test
revision: 1399c76144fd37290681b995c656ef9b2e06e26d
metrics:
- type: accuracy
value: 44.662000000000006
- type: f1
value: 43.42371965372771
- task:
type: Retrieval
dataset:
type: arguana
name: MTEB ArguAna
config: default
split: test
revision: None
metrics:
- type: map_at_1
value: 31.721
- type: map_at_10
value: 49.221
- type: map_at_100
value: 49.884
- type: map_at_1000
value: 49.888
- type: map_at_3
value: 44.31
- type: map_at_5
value: 47.276
- type: mrr_at_1
value: 32.432
- type: mrr_at_10
value: 49.5
- type: mrr_at_100
value: 50.163000000000004
- type: mrr_at_1000
value: 50.166
- type: mrr_at_3
value: 44.618
- type: mrr_at_5
value: 47.541
- type: ndcg_at_1
value: 31.721
- type: ndcg_at_10
value: 58.384
- type: ndcg_at_100
value: 61.111000000000004
- type: ndcg_at_1000
value: 61.187999999999995
- type: ndcg_at_3
value: 48.386
- type: ndcg_at_5
value: 53.708999999999996
- type: precision_at_1
value: 31.721
- type: precision_at_10
value: 8.741
- type: precision_at_100
value: 0.991
- type: precision_at_1000
value: 0.1
- type: precision_at_3
value: 20.057
- type: precision_at_5
value: 14.609
- type: recall_at_1
value: 31.721
- type: recall_at_10
value: 87.411
- type: recall_at_100
value: 99.075
- type: recall_at_1000
value: 99.644
- type: recall_at_3
value: 60.171
- type: recall_at_5
value: 73.044
- task:
type: Clustering
dataset:
type: mteb/arxiv-clustering-p2p
name: MTEB ArxivClusteringP2P
config: default
split: test
revision: a122ad7f3f0291bf49cc6f4d32aa80929df69d5d
metrics:
- type: v_measure
value: 46.40419580759799
- task:
type: Clustering
dataset:
type: mteb/arxiv-clustering-s2s
name: MTEB ArxivClusteringS2S
config: default
split: test
revision: f910caf1a6075f7329cdf8c1a6135696f37dbd53
metrics:
- type: v_measure
value: 40.48593255007969
- task:
type: Reranking
dataset:
type: mteb/askubuntudupquestions-reranking
name: MTEB AskUbuntuDupQuestions
config: default
split: test
revision: 2000358ca161889fa9c082cb41daa8dcfb161a54
metrics:
- type: map
value: 63.889179122289995
- type: mrr
value: 77.61146286769556
- task:
type: STS
dataset:
type: mteb/biosses-sts
name: MTEB BIOSSES
config: default
split: test
revision: d3fb88f8f02e40887cd149695127462bbcf29b4a
metrics:
- type: cos_sim_pearson
value: 88.15075203727929
- type: cos_sim_spearman
value: 86.9622224570873
- type: euclidean_pearson
value: 86.70473853624121
- type: euclidean_spearman
value: 86.9622224570873
- type: manhattan_pearson
value: 86.21089380980065
- type: manhattan_spearman
value: 86.75318154937008
- task:
type: BitextMining
dataset:
type: mteb/bucc-bitext-mining
name: MTEB BUCC (de-en)
config: de-en
split: test
revision: d51519689f32196a32af33b075a01d0e7c51e252
metrics:
- type: accuracy
value: 99.65553235908142
- type: f1
value: 99.60681976339595
- type: precision
value: 99.58246346555325
- type: recall
value: 99.65553235908142
- task:
type: BitextMining
dataset:
type: mteb/bucc-bitext-mining
name: MTEB BUCC (fr-en)
config: fr-en
split: test
revision: d51519689f32196a32af33b075a01d0e7c51e252
metrics:
- type: accuracy
value: 99.26260180497468
- type: f1
value: 99.14520507740848
- type: precision
value: 99.08650671362535
- type: recall
value: 99.26260180497468
- task:
type: BitextMini

查看官方来源