Erlangshen SimCSE 110M Chinese

🚀 Erlangshen-SimCSE-110M-Chinese

Based on the unsupervised version of SimCSE, this model is trained on supervised tasks using collected and sorted Chinese NLI data, achieving good results on Chinese sentence pair tasks.

• Main Page: Fengshenbang
• Github: Fengshenbang-LM

🚀 Quick Start

This section provides a quick overview of the Erlangshen-SimCSE-110M-Chinese model, including its features, installation, and basic usage.

✨ Features

• Supervised Training: Based on the unsupervised version of SimCSE, it is trained on supervised tasks using Chinese NLI data.
• Good Performance: Achieves good results on Chinese sentence pair tasks.
• General Sentence Embedding: Capable of extracting general sentence vectors without fine-tuning.

📦 Installation

To use the Erlangshen-SimCSE-110M-Chinese model, you need to install the transformers library. You can install it using pip:

pip install transformers

💻 Usage Examples

Basic Usage

from transformers import AutoTokenizer, AutoModelForMaskedLM

# Load the model and tokenizer
model = AutoModelForMaskedLM.from_pretrained('IDEA-CCNL/Erlangshen-SimCSE-110M-Chinese')
tokenizer = AutoTokenizer.from_pretrained('IDEA-CCNL/Erlangshen-SimCSE-110M-Chinese')

import torch
from sklearn.metrics.pairwise import cosine_similarity

# Define two sentences
texta = '今天天气真不错，我们去散步吧！'
textb = '今天天气真糟糕，还是在宅家里写bug吧！'

# Tokenize the sentences
inputs_a = tokenizer(texta, return_tensors="pt")
inputs_b = tokenizer(textb, return_tensors="pt")

# Get the model outputs
outputs_a = model(**inputs_a, output_hidden_states=True)
texta_embedding = outputs_a.hidden_states[-1][:, 0, :].squeeze()

outputs_b = model(**inputs_b, output_hidden_states=True)
textb_embedding = outputs_b.hidden_states[-1][:, 0, :].squeeze()

# Calculate the cosine similarity
with torch.no_grad():
    similarity_score = cosine_similarity(texta_embedding.reshape(1, -1), textb_embedding.reshape(1, -1))[0][0]

print(similarity_score)

📚 Documentation

Model Taxonomy

Model Information

In order to obtain a general sentence-embedding model, we used a large number of unsupervised and supervised data for contrastive learning based on the Bert-base model. Finally, we obtained a model that can use the [CLS] output from the model to judge the similarity without fine-tuning. Different from the sentence similarity task after fine-tuning the BERT model, our model has the ability to extract sentence vectors directly after pre-training. The evaluation results on some tasks are as follows:

Note: Our model uses [cls] directly without whitening; other models use last avg + whitening.

📄 License

This project is licensed under the Apache-2.0 license.

📖 Citation

If you use our model in your work, please cite our paper:

@article{fengshenbang,
  author    = {Jiaxing Zhang and Ruyi Gan and Junjie Wang and Yuxiang Zhang and Lin Zhang and Ping Yang and Xinyu Gao and Ziwei Wu and Xiaoqun Dong and Junqing He and Jianheng Zhuo and Qi Yang and Yongfeng Huang and Xiayu Li and Yanghan Wu and Junyu Lu and Xinyu Zhu and Weifeng Chen and Ting Han and Kunhao Pan and Rui Wang and Hao Wang and Xiaojun Wu and Zhongshen Zeng and Chongpei Chen},
  title     = {Fengshenbang 1.0: Being the Foundation of Chinese Cognitive Intelligence},
  journal   = {CoRR},
  volume    = {abs/2209.02970},
  year      = {2022}
}

You can also cite our website:

@misc{Fengshenbang-LM,
  title={Fengshenbang-LM},
  author={IDEA-CCNL},
  year={2021},
  howpublished={\url{https://github.com/IDEA-CCNL/Fengshenbang-LM}},
}