XYZ-embedding-zh Open-source Chinese Embedding Model - Free for Sentence and Paragraph Clustering and Semantic Search

XYZ Embedding Zh

Developed by fangxq

XYZ-embedding-zh is a Chinese embedding model based on sentence-transformers, which can map sentences and paragraphs to a 1792-dimensional dense vector space and is suitable for tasks such as clustering and semantic search.

Text Embedding #Chinese semantic retrieval #Medical Q&A re-ranking #High-precision vectorization

Downloads 22

Release Time : 4/25/2025

Model Overview

This model is specifically designed for Chinese text and can efficiently convert text into high-dimensional vector representations, supporting various natural language processing tasks such as information retrieval and re-ranking.

Model Features

High-dimensional vector representation

Map sentences and paragraphs to a 1792-dimensional dense vector space to capture rich semantic information.

Multi-task support

Support various tasks, including sentence similarity calculation, feature extraction, re-ranking, and information retrieval.

Chinese optimization

Optimized specifically for Chinese text to better handle Chinese semantics.

Model Capabilities

Sentence similarity calculation

Feature extraction

Text re-ranking

Information retrieval

Use Cases

Information retrieval

Medical Q&A retrieval

Conduct information retrieval in the medical Q&A dataset to help users quickly find relevant answers.

On the MTEB Cmedqa retrieval dataset, the map_at_10 reaches 41.228.

E-commerce product retrieval

Conduct product retrieval on the e-commerce platform to improve the user search experience.

On the MTEB e-commerce retrieval dataset, the ndcg_at_10 reaches 69.719.

Text re-ranking

Medical Q&A re-ranking

Re-rank the medical Q&A results to improve the relevance of the answers.

On the MTEB CMedQAv1 dataset, the map reaches 89.618.

General text re-ranking

Re-rank the general text retrieval results to optimize the search results.

On the MTEB T2 re-ranking dataset, the map reaches 69.066.

🚀 XYZ-embedding-zh

This is a sentence-transformers model that maps sentences and paragraphs to a 1792-dimensional dense vector space. It can be applied to tasks such as clustering or semantic search, offering efficient and accurate solutions for text analysis.

🚀 Quick Start

Using this model becomes easy when you have sentence-transformers installed. First, install the sentence-transformers library:

pip install -U sentence-transformers

Then you can use the model like this:

from sentence_transformers import SentenceTransformer
sentences = ["This is an example sentence", "Each sentence is converted"]

model = SentenceTransformer('fangxq/XYZ-embedding-zh')
embeddings = model.encode(sentences)
print(embeddings)

💻 Usage Examples

Basic Usage

from sentence_transformers import SentenceTransformer
sentences = ["This is an example sentence", "Each sentence is converted"]

model = SentenceTransformer('fangxq/XYZ-embedding-zh')
embeddings = model.encode(sentences)
print(embeddings)

📚 Documentation

Evaluation Results

For an automated evaluation of this model, see the Sentence Embeddings Benchmark: https://seb.sbert.net

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})
  (2): Dense({'in_features': 1024, 'out_features': 1792, 'bias': True, 'activation_function': 'torch.nn.modules.linear.Identity'})
)

Model Metrics

Dataset	Task	MAP	MRR	Main Score
MTEB CMedQAv1	Reranking	89.61792115239176	91.46722222222222	89.61792115239176
MTEB CMedQAv2	Reranking	89.22040591564271	91.2995238095238	89.22040591564271
MTEB CmedqaRetrieval	Retrieval	Multiple metrics available	Multiple metrics available	47.899
MTEB CovidRetrieval	Retrieval	Multiple metrics available	Multiple metrics available	88.237
MTEB DuRetrieval	Retrieval	Multiple metrics available	Multiple metrics available	90.794
MTEB EcomRetrieval	Retrieval	Multiple metrics available	Multiple metrics available	69.719
MTEB MMarcoReranking	Reranking	35.30260957061897	34.098015873015875	35.30260957061897
MTEB MMarcoRetrieval	Retrieval	Multiple metrics available	Multiple metrics available	82.479
MTEB MedicalRetrieval	Retrieval	Multiple metrics available	Multiple metrics available	67.415
MTEB T2Reranking	Reranking	69.06615146698508	79.7588755091294	69.06615146698508
MTEB T2Retrieval	Retrieval	Multiple metrics available	Multiple metrics available	85.931
MTEB VideoRetrieval	Retrieval	Multiple metrics available	Multiple metrics available	80.03099999999999

🔧 Technical Details

The model architecture consists of a Transformer layer, a Pooling layer, and a Dense layer. The Transformer layer uses a BertModel with a maximum sequence length of 512 and no lower - casing. The Pooling layer uses mean token pooling, and the Dense layer maps the 1024 - dimensional input features to 1792 - dimensional output features.

📄 License

No license information provided in the original document.

Citing & Authors

No specific citation and author details are provided in a structured way in the given text. If there are specific requirements for citation, please refer to the official documentation or relevant resources of the model.

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