metadata

pipeline_tag: sentence-similarity
tags:
  - sentence-transformers
  - feature-extraction
  - sentence-similarity
language:
  - es
datasets:
  - hackathon-pln-es/nli-es
widget:
  - text: >-
      A ver si nos tenemos que poner todos en huelga hasta cobrar lo que
      queramos.
  - text: >-
      La huelga es el método de lucha más eficaz para conseguir mejoras en el
      salario.
  - text: Tendremos que optar por hacer una huelga para cobrar lo que queremos.
  - text: Queda descartada la huelga aunque no cobremos lo que queramos.

bertin-roberta-base-finetuning-esnli

This is a sentence-transformers model trained on a collection of NLI tasks for Spanish. It maps sentences & paragraphs to a 768 dimensional dense vector space and can be used for tasks like clustering or semantic search.

Based around the siamese networks approach from this paper.

You can see a demo for this model here.

You can find our other model, paraphrase-spanish-distilroberta here and its demo here.

Usage (Sentence-Transformers)

Using this model becomes easy when you have sentence-transformers installed:

pip install -U sentence-transformers

Then you can use the model like this:

from sentence_transformers import SentenceTransformer
sentences = ["Este es un ejemplo", "Cada oración es transformada"]

model = SentenceTransformer('hackathon-pln-es/bertin-roberta-base-finetuning-esnli')
embeddings = model.encode(sentences)
print(embeddings)

Usage (HuggingFace Transformers)

Without sentence-transformers, you can use the model like this: First, you pass your input through the transformer model, then you have to apply the right pooling-operation on-top of the contextualized word embeddings.

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 = ['This is an example sentence', 'Each sentence is converted']

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('hackathon-pln-es/bertin-roberta-base-finetuning-esnli')
model = AutoModel.from_pretrained('hackathon-pln-es/bertin-roberta-base-finetuning-esnli')

# 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)

Evaluation Results

Our model was evaluated on the task of Semantic Textual Similarity using the SemEval-2015 Task for Spanish. We measure

	BETO STS	BERTIN STS (this model)	Relative improvement
cosine_pearson	0.609803	0.683188	+12.03
cosine_spearman	0.528776	0.615916	+16.48
euclidean_pearson	0.590613	0.672601	+13.88
euclidean_spearman	0.526529	0.611539	+16.15
manhattan_pearson	0.589108	0.672040	+14.08
manhattan_spearman	0.525910	0.610517	+16.09
dot_pearson	0.544078	0.600517	+10.37
dot_spearman	0.460427	0.521260	+13.21

Training

The model was trained with the parameters:

Dataset

We used a collection of datasets of Natural Language Inference as training data:

ESXNLI, only the part in spanish
SNLI, automatically translated
MultiNLI, automatically translated

The whole dataset used is available here.

Here we leave the trick we used to increase the amount of data for training here:

  for row in reader:
    if row['language'] == 'es':
      
      sent1 = row['sentence1'].strip()
      sent2 = row['sentence2'].strip()
    
      add_to_samples(sent1, sent2, row['gold_label'])
      add_to_samples(sent2, sent1, row['gold_label'])  #Also add the opposite

DataLoader:

sentence_transformers.datasets.NoDuplicatesDataLoader.NoDuplicatesDataLoader of length 1818 with parameters:

{'batch_size': 64}

Loss:

sentence_transformers.losses.MultipleNegativesRankingLoss.MultipleNegativesRankingLoss with parameters:

{'scale': 20.0, 'similarity_fct': 'cos_sim'}

Parameters of the fit()-Method:

{
    "epochs": 10,
    "evaluation_steps": 0,
    "evaluator": "sentence_transformers.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'transformers.optimization.AdamW'>",
    "optimizer_params": {
        "lr": 2e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 909,
    "weight_decay": 0.01
}

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 514, 'do_lower_case': False}) with Transformer model: RobertaModel 
  (1): Pooling({'word_embedding_dimension': 768, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False})
)

Authors