Ruri-Large Open-source Japanese Text Model - Free Deployment and Precise Calculation of Text Similarity, Supports Long Texts

Ruri Large

Developed by cl-nagoya

Ruri-Large is a high-performance embedding model specialized in Japanese text similarity calculation, based on transformer architecture with support for long text processing (maximum length 8192).

Text Embedding

Safetensors

JapaneseOpen Source License:Apache-2.0 #Japanese Text Embedding #High-precision Semantic Similarity #Long Text Support

Downloads 6,784

Release Time : 8/28/2024

Model Overview

This model is primarily used for semantic similarity calculation and feature extraction of Japanese texts, demonstrating outstanding performance in the JMTEB benchmark. It is specially optimized for distinguishing between query and passage texts, requiring the addition of 'クエリ:' (query:) or '文章:' (passage:) prefixes before input.

Model Features

High-performance Japanese Processing

Achieved an average score of 73.31 in the JMTEB benchmark, outperforming similar Japanese embedding models

Long Text Support

Supports text processing up to 8192 tokens, suitable for long document analysis

Query/Passage Distinction

Optimizes retrieval effectiveness by distinguishing query texts from passage texts through prefix markers

Model Capabilities

Japanese Text Embedding

Semantic Similarity Calculation

Text Feature Extraction

Information Retrieval

Text Clustering

Use Cases

Information Retrieval

Q&A Systems

Enables precise question answering by calculating similarity between queries and knowledge base passages

Achieved 73.02 points in JMTEB retrieval tasks

Content Analysis

Text Clustering

Performs semantic clustering analysis on large volumes of Japanese texts

Achieved 51.82 points in JMTEB clustering tasks

🚀 Ruri: Japanese General Text Embeddings

Ruri is a model designed for Japanese general text embeddings, offering high - performance sentence similarity and feature extraction capabilities.

🚀 Quick Start

First, you need to install the Sentence Transformers library:

pip install -U sentence-transformers fugashi sentencepiece unidic-lite

Then, you can load this model and run inference.

import torch.nn.functional as F
from sentence_transformers import SentenceTransformer

# Download from the 🤗 Hub
model = SentenceTransformer("cl-nagoya/ruri-large")

# Don't forget to add the prefix "クエリ: " for query-side or "文章: " for passage-side texts.
sentences = [
    "クエリ: 瑠璃色はどんな色？",
    "文章: 瑠璃色（るりいろ）は、紫みを帯びた濃い青。名は、半貴石の瑠璃（ラピスラズリ、英: lapis lazuli）による。JIS慣用色名では「こい紫みの青」（略号 dp-pB）と定義している[1][2]。",
    "クエリ: ワシやタカのように、鋭いくちばしと爪を持った大型の鳥類を総称して「何類」というでしょう?",
    "文章: ワシ、タカ、ハゲワシ、ハヤブサ、コンドル、フクロウが代表的である。これらの猛禽類はリンネ前後の時代(17~18世紀)には鷲類・鷹類・隼類及び梟類に分類された。ちなみにリンネは狩りをする鳥を単一の目(もく)にまとめ、vultur(コンドル、ハゲワシ)、falco(ワシ、タカ、ハヤブサなど)、strix(フクロウ)、lanius(モズ)の4属を含めている。",
]

embeddings = model.encode(sentences, convert_to_tensor=True)
print(embeddings.size())
# [4, 1024]

similarities = F.cosine_similarity(embeddings.unsqueeze(0), embeddings.unsqueeze(1), dim=2)
print(similarities)
# [[1.0000, 0.9429, 0.6565, 0.6997],
#  [0.9429, 1.0000, 0.6579, 0.6768],
#  [0.6565, 0.6579, 1.0000, 0.8933],
#  [0.6997, 0.6768, 0.8933, 1.0000]]

✨ Features

New v3 Models Available: We recommend using the following v3 models going forward. |ID| #Param.|Max Len.|Avg. JMTEB| |-|-|-|-| |cl-nagoya/ruri-v3-30m|37M|8192|74.51| |cl-nagoya/ruri-v3-70m|70M|8192|75.48| |cl-nagoya/ruri-v3-130m|132M|8192|76.55| |cl-nagoya/ruri-v3-310m|315M|8192|77.24|

📚 Documentation

Benchmarks

JMTEB

Evaluated with JMTEB.

Model	#Param.	Avg.	Retrieval	STS	Classfification	Reranking	Clustering	PairClassification
cl-nagoya/sup-simcse-ja-base	111M	68.56	49.64	82.05	73.47	91.83	51.79	62.57
cl-nagoya/sup-simcse-ja-large	337M	66.51	37.62	83.18	73.73	91.48	50.56	62.51
cl-nagoya/unsup-simcse-ja-base	111M	65.07	40.23	78.72	73.07	91.16	44.77	62.44
cl-nagoya/unsup-simcse-ja-large	337M	66.27	40.53	80.56	74.66	90.95	48.41	62.49
pkshatech/GLuCoSE-base-ja	133M	70.44	59.02	78.71	76.82	91.90	49.78	66.39

sentence-transformers/LaBSE	472M	64.70	40.12	76.56	72.66	91.63	44.88	62.33
intfloat/multilingual-e5-small	118M	69.52	67.27	80.07	67.62	93.03	46.91	62.19
intfloat/multilingual-e5-base	278M	70.12	68.21	79.84	69.30	92.85	48.26	62.26
intfloat/multilingual-e5-large	560M	71.65	70.98	79.70	72.89	92.96	51.24	62.15

OpenAI/text-embedding-ada-002	-	69.48	64.38	79.02	69.75	93.04	48.30	62.40
OpenAI/text-embedding-3-small	-	70.86	66.39	79.46	73.06	92.92	51.06	62.27
OpenAI/text-embedding-3-large	-	73.97	74.48	82.52	77.58	93.58	53.32	62.35

Ruri-Small	68M	71.53	69.41	82.79	76.22	93.00	51.19	62.11
Ruri-Base	111M	71.91	69.82	82.87	75.58	92.91	54.16	62.38
Ruri-Large (this model)	337M	73.31	73.02	83.13	77.43	92.99	51.82	62.29

Model Details

Model Description

Property	Details
Model Type	Sentence Transformer
Base model	cl-nagoya/ruri-pt-large
Maximum Sequence Length	512 tokens
Output Dimensionality	1024
Similarity Function	Cosine Similarity
Language	Japanese
License	Apache 2.0
Paper	https://arxiv.org/abs/2409.07737

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BertModel 
  (1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
)

Framework Versions

Python: 3.10.13
Sentence Transformers: 3.0.0
Transformers: 4.41.2
PyTorch: 2.3.1+cu118
Accelerate: 0.30.1
Datasets: 2.19.1
Tokenizers: 0.19.1

🔧 Technical Details

Citation

@misc{
  Ruri,
  title={{Ruri: Japanese General Text Embeddings}}, 
  author={Hayato Tsukagoshi and Ryohei Sasano},
  year={2024},
  eprint={2409.07737},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2409.07737}, 
}

📄 License

This model is published under the Apache License, Version 2.0.

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