Dj

🚀 DMetaSoul/sbert-chinese-general-v2

This model is based on the BERT model of the bert-base-chinese version and is trained on the million-level semantic similarity dataset SimCLUE. It is suitable for general semantic matching scenarios. In terms of performance, this model has better generalization ability on various tasks.

⚠️ Important Note

A lightweight version of this model has also been open-sourced!

🚀 Quick Start

✨ Features

• Based on the bert-base-chinese model.
• Trained on the SimCLUE dataset.
• Suitable for general semantic matching scenarios.
• Better generalization ability on various tasks.

📦 Installation

You can install the necessary library via the following command:

pip install -U sentence-transformers

💻 Usage Examples

Basic Usage

Use the sentence-transformers framework to load the model and extract text representation vectors:

from sentence_transformers import SentenceTransformer
sentences = ["我的儿子！他猛然间喊道，我的儿子在哪儿？", "我的儿子呢！他突然喊道，我的儿子在哪里？"]

model = SentenceTransformer('DMetaSoul/sbert-chinese-general-v2')
embeddings = model.encode(sentences)
print(embeddings)

Advanced Usage

If you don't want to use sentence-transformers, you can load the model and extract text vectors through HuggingFace Transformers:

from transformers import AutoTokenizer, AutoModel
import torch


#Mean Pooling - Take attention mask into account for correct averaging
def mean_pooling(model_output, attention_mask):
    token_embeddings = model_output[0] #First element of model_output contains all token embeddings
    input_mask_expanded = attention_mask.unsqueeze(-1).expand(token_embeddings.size()).float()
    return torch.sum(token_embeddings * input_mask_expanded, 1) / torch.clamp(input_mask_expanded.sum(1), min=1e-9)


# Sentences we want sentence embeddings for
sentences = ["我的儿子！他猛然间喊道，我的儿子在哪儿？", "我的儿子呢！他突然喊道，我的儿子在哪里？"]

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('DMetaSoul/sbert-chinese-general-v2')
model = AutoModel.from_pretrained('DMetaSoul/sbert-chinese-general-v2')

# Tokenize sentences
encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')

# Compute token embeddings
with torch.no_grad():
    model_output = model(**encoded_input)

# Perform pooling. In this case, mean pooling.
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])

print("Sentence embeddings:")
print(sentence_embeddings)

📚 Documentation

Evaluation

The model was evaluated on several public semantic matching datasets, and the correlation coefficient between vector similarity and real labels was calculated:

Here, the differences between this model and the previously released sbert-chinese-general-v1 are compared. It can be seen that this model has better generalization ability on multiple tasks.

📄 License

Citing & Authors

E-mail: xiaowenbin@dmetasoul.com