Edit model card
YAML Metadata Warning: empty or missing yaml metadata in repo card (https://huggingface.co/docs/hub/model-cards#model-card-metadata)

kwang2049/TSDAE-askubuntu2nli_stsb

This is a model from the paper "TSDAE: Using Transformer-based Sequential Denoising Auto-Encoder for Unsupervised Sentence Embedding Learning". This model was only trained with the TSDAE objective on AskUbuntu in an unsupervised manner. Training procedure of this model:

  1. Initialized with bert-base-uncased;
  2. Unsupervised training on AskUbuntu with the TSDAE objective;

The pooling method is CLS-pooling.

Usage

To use this model, an convenient way is through SentenceTransformers. So please install it via:

pip install sentence-transformers

And then load the model and use it to encode sentences:

from sentence_transformers import SentenceTransformer, models
dataset = 'askubuntu'
model_name_or_path = f'kwang2049/TSDAE-{dataset}'
model = SentenceTransformer(model_name_or_path)
model[1] = models.Pooling(model[0].get_word_embedding_dimension(), pooling_mode='cls')  # Note this model uses CLS-pooling
sentence_embeddings = model.encode(['This is the first sentence.', 'This is the second one.'])

Evaluation

To evaluate the model against the datasets used in the paper, please install our evaluation toolkit USEB:

pip install useb  # Or git clone and pip install .
python -m useb.downloading all  # Download both training and evaluation data

And then do the evaluation:

from sentence_transformers import SentenceTransformer, models
import torch
from useb import run_on
dataset = 'askubuntu'
model_name_or_path = f'kwang2049/TSDAE-{dataset}'
model = SentenceTransformer(model_name_or_path)
model[1] = models.Pooling(model[0].get_word_embedding_dimension(), pooling_mode='cls')  # Note this model uses CLS-pooling
@torch.no_grad()
def semb_fn(sentences) -> torch.Tensor:
   return torch.Tensor(model.encode(sentences, show_progress_bar=False))
result = run_on(
   dataset,
   semb_fn=semb_fn,
   eval_type='test',
   data_eval_path='data-eval'
)

Training

Please refer to the page of TSDAE training in SentenceTransformers.

Cite & Authors

If you use the code for evaluation, feel free to cite our publication TSDAE: Using Transformer-based Sequential Denoising Auto-Encoderfor Unsupervised Sentence Embedding Learning:

@article{wang-2021-TSDAE,
    title = "TSDAE: Using Transformer-based Sequential Denoising Auto-Encoderfor Unsupervised Sentence Embedding Learning",
    author = "Wang, Kexin and Reimers, Nils and  Gurevych, Iryna", 
    journal= "arXiv preprint arXiv:2104.06979",
    month = "4",
    year = "2021",
    url = "https://arxiv.org/abs/2104.06979",
}
Downloads last month
6