IBM Granite Open-Source Multilingual Embedding Model - Supports 12 Languages for Information Retrieval and Multilingual Applications

Granite Embedding 278m Multilingual GGUF

Developed by bartowski

IBM Granite multilingual embedding model, supporting text embedding tasks in 12 languages, suitable for information retrieval and multilingual application scenarios.

Text Embedding Supports Multiple LanguagesOpen Source License:Apache-2.0 #Multilingual Embedding #Efficient Retrieval #Low-resource Optimization

Downloads 4,815

Release Time : 12/18/2024

Model Overview

This is a multilingual text embedding model capable of converting texts in different languages into high-dimensional vector representations, suitable for tasks such as cross-language retrieval and semantic similarity calculation.

Model Features

Multilingual Support

Supports text embedding in 12 languages, including major languages such as English, Chinese, and Arabic

Efficient Quantization

Provides multiple quantized versions, including recommended versions like Q4_K_M, balancing model quality and size

Hardware Optimization

Supports ARM/AVX hardware optimization, performing well on mobile devices

Excellent Retrieval Performance

Demonstrates outstanding retrieval performance metrics on datasets like Miracl

Model Capabilities

Text Embedding

Cross-language Retrieval

Semantic Similarity Calculation

Multilingual Text Processing

Use Cases

Information Retrieval

Cross-language Document Retrieval

Using embedding vectors for multilingual document similarity search

Achieved NDCG@10 of 0.49372 (English) on the Miracl dataset

Semantic Analysis

Multilingual Semantic Similarity Calculation

Calculating semantic similarity between texts in different languages

🚀 ibm-granite/granite-embedding-278m-multilingual

This is a multilingual text - generation model that supports multiple languages including English, Arabic, Czech, German, Spanish, French, Italian, Japanese, Korean, Dutch, Portuguese, and Chinese. It provides high - quality embedding capabilities for various retrieval tasks.

📚 Documentation

Model Information

Property	Details
Quantized By	bartowski
Pipeline Tag	text - generation
Languages Supported	en, ar, cs, de, es, fr, it, ja, ko, nl, pt, zh
Tags	language, granite, embeddings, multilingual
License	apache - 2.0
Base Model	ibm - granite/granite - embedding - 278m - multilingual

Model Performance Metrics

The model has been tested on the Miracl dataset for retrieval tasks across multiple languages. Here are the detailed metrics:

Miracl (en)

Metric Type	Value
ndcg_at_1	0.45557
ndcg_at_10	0.49372
ndcg_at_100	0.5728
ndcg_at_1000	0.59187
ndcg_at_20	0.52863
ndcg_at_3	0.43969
ndcg_at_5	0.45551
recall_at_1	0.21785
recall_at_10	0.59513
recall_at_100	0.85785
recall_at_1000	0.96041
recall_at_20	0.69357
recall_at_3	0.40403
recall_at_5	0.48499

Miracl (ar)

Metric Type	Value
ndcg_at_1	0.57459
ndcg_at_10	0.64238
ndcg_at_100	0.6867
ndcg_at_1000	0.6951
ndcg_at_20	0.66455
ndcg_at_3	0.58162
ndcg_at_5	0.60831
recall_at_1	0.38064
recall_at_10	0.75098
recall_at_100	0.91203
recall_at_1000	0.96706
recall_at_20	0.81978
recall_at_3	0.58618
recall_at_5	0.66353

Miracl (bn)

Metric Type	Value
ndcg_at_1	0.60341
ndcg_at_10	0.68055
ndcg_at_100	0.72008
ndcg_at_1000	0.72716
ndcg_at_20	0.69914
ndcg_at_3	0.60805
ndcg_at_5	0.64486
recall_at_1	0.37948
recall_at_10	0.80609
recall_at_100	0.94305
recall_at_1000	0.98625
recall_at_20	0.86141
recall_at_3	0.61095
recall_at_5	0.71316

Miracl (de)

Metric Type	Value
ndcg_at_1	0.45574
ndcg_at_10	0.48123
ndcg_at_100	0.56049
ndcg_at_1000	0.57979
ndcg_at_20	0.51785
ndcg_at_3	0.41243
ndcg_at_5	0.4386
recall_at_1	0.20401
recall_at_10	0.58779
recall_at_100	0.8584
recall_at_1000	0.97364
recall_at_20	0.69061
recall_at_3	0.36573
recall_at_5	0.47495

Miracl (es)

Metric Type	Value
ndcg_at_1	0.5571
ndcg_at_10	0.49688
ndcg_at_100	0.60493
ndcg_at_1000	0.62922
ndcg_at_20	0.54438
ndcg_at_3	0.47981
ndcg_at_5	0.46584
recall_at_1	0.1638
recall_at_10	0.54155
recall_at_100	0.85136
recall_at_1000	0.96951
recall_at_20	0.65329
recall_at_3	0.31503
recall_at_5	0.40356

Miracl (fa)

Metric Type	Value
ndcg_at_1	0.39873
ndcg_at_10	0.50226
ndcg_at_100	0.56517
ndcg_at_1000	0.57967
ndcg_at_20	0.5292
ndcg_at_3	0.42738
ndcg_at_5	0.45843
recall_at_1	0.25369
recall_at_10	0.63776
recall_at_100	0.87686
recall_at_1000	0.9671
recall_at_20	0.72099
recall_at_3	0.43808
recall_at_5	0.52378

Miracl (fi)

Metric Type	Value
ndcg_at_1	0.60818
ndcg_at_10	0.6746
ndcg_at_100	0.71516
ndcg_at_1000	0.7218
ndcg_at_20	0.69692
ndcg_at_3	0.6006
ndcg_at_5	0.63842
recall_at_1	0.39264
recall_at_10	0.78577
recall_at_100	0.93291
recall_at_1000	0.97493
recall_at_20	0.85435
recall_at_3	0.61055
recall_at_5	0.69774

Miracl (fr)

Metric Type	Value
ndcg_at_1	0.3965
ndcg_at_10	0.49891
ndcg_at_100	0.56492
ndcg_at_1000	0.57837
ndcg_at_20	0.53163
ndcg_at_3	0.39843
ndcg_at_5	0.44416
recall_at_1	0.22644
recall_at_10	0.65169
recall_at_100	0.89786
recall_at_1000	0.98081
recall_at_20	0.75338
recall_at_3	0.39798
recall_at_5	0.51001

Miracl (hi)

Metric Type	Value
ndcg_at_1	0.36857
ndcg_at_10	0.46141
ndcg_at_100	0.52565
ndcg_at_1000	0.54319
ndcg_at_20	0.49384
ndcg_at_3	0.39469
ndcg_at_5	0.4184
recall_at_1	0.20185
recall_at_10	0.59474
recall_at_100	0.83385
recall_at_1000	0.94813
recall_at_20	0.69437
recall_at_3	0.38993
recall_at_5	0.47881

Miracl (id)

Metric Type	Value
ndcg_at_1	0.46354
ndcg_at_10	0.47229
ndcg_at_100	0.5525
ndcg_at_1000	0.57648
ndcg_at_20	0.50606
ndcg_at_3	0.42538
ndcg_at_5	0.43717
recall_at_1	0.20787
recall_at_10	0.54771
recall_at_100	0.80689
recall_at_1000	0.94032
recall_at_20	0.63842
recall_at_3	0.36229
recall_at_5	0.44437

Miracl (ja)

Metric Type	Value
ndcg_at_1	0.56279
ndcg_at_10	0.6281
ndcg_at_100	0.67757
ndcg_at_1000	0.68667
ndcg_at_20	0.6521
ndcg_at_3	0.56226
ndcg_at_5	0.5866
recall_at_1	0.36648
recall_at_10	0.7496
recall_at_100	0.92461
recall_at_1000	0.97827
recall_at_20	0.82326
recall_at_3	0.55845
recall_at_5	0.63854

Miracl (ko)

Metric Type	Value
ndcg_at_1	0.52582
ndcg_at_10	0.59216
ndcg_at_100	0.65093
ndcg_at_1000	0.66204
ndcg_at_20	0.62427
ndcg_at_3	0.5373
ndcg_at_5	0.55886
recall_at_1	0.30521
recall_at_10	0.71159
recall_at_100	0.90203
recall_at_1000	0.96714
recall_at_20	0.80209
recall_at_3	0.515
recall_at_5	0.6071

Miracl (ru)

Metric Type	Value
ndcg_at_1	0.47524
ndcg_at_10	0.52349
ndcg_at_100	0.59725
ndcg_at_1000	0.61313
ndcg_at_20	0.55669
ndcg_at_3	0.46812
ndcg_at_5	0.48442
recall_at_1	0.24337
recall_at_10	0.62437
recall_at_100	0.86489
recall_at_1000	0.95266
recall_at_20	0.71411
recall_at_3	0.42927
recall_at_5	0.51258

Miracl (sw)

Metric Type	Value
ndcg_at_1	0.5166
ndcg_at_10	0.61271
ndcg_at_100	0.66099
ndcg_at_1000	0.66867
ndcg_at_20	0.63643
ndcg_at_3	0.54828
ndcg_at_5	0.57382
recall_at_1	0.35277
recall_at_10	0.74368
recall_at_100	0.92261
recall_at_1000	0.97109
recall_at_20	0.81888
recall_at_3	0.56739
recall_at_5	0.6421

Miracl (te)

Metric Type	Value
ndcg_at_1	0.63768
ndcg_at_10	0.79193
ndcg_at_100	0.80243
ndcg_at_1000	0.80438
ndcg_at_20	0.79549
ndcg_at_3	0.76031
ndcg_at_5	0.77915
recall_at_1	0.63084
recall_at_10	0.92411
recall_at_100	0.97363
recall_at_1000	0.98833
recall_at_20	0.9374
recall_at_3	0.84159
recall_at_5	0.88627

Miracl (th)

Metric Type	Value
ndcg_at_1	0.66712
ndcg_at_10	0.73324
ndcg_at_100	0.76633
ndcg_at_1000	0.77119
ndcg_at_20	0.75243
ndcg_at_3	0.67393
ndcg_at_5	0.70201
recall_at_1	0.47106
recall_at_10	0.84294
recall_at_100	0.95949
recall_at_1000	0.98874
recall_at_20	0.90085
recall_at_3	0.68456
recall_at_5	0.75915

Miracl (yo)

Metric Type	Value
ndcg_at_1	0.4958
ndcg_at_10	0.68705

📄 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