Edit model card

ST-NLI-ca_paraphrase-multilingual-mpnet-base

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

It has been developed through further training of a multilingual fine-tuned model, paraphrase-multilingual-mpnet-base-v2 using NLI data. Specifically, it was trained on two Catalan NLI datasets: TE-ca and the professional translation of XNLI into Catalan. The training employed the Multiple Negatives Ranking Loss with Hard Negatives, which leverages triplets composed of a premise, an entailed hypothesis, and a contradiction. It is important to note that, given this format, neutral hypotheses from the NLI datasets were not used for training. Additionally, as a form of data augmentation, the model's training set was expanded by duplicating the triplets, wherein the order of the premise and entailed hypothesis was reversed, resulting in a total of 18,928 triplets.

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, util
sentences = ["This is an example sentence", "Each sentence is converted"]

model = SentenceTransformer('{MODEL_NAME}')
embeddings = model.encode(sentences)
print(embeddings)

For instance, to sort a list of sentences by their similarity to a reference sentence, the following code can be used:

reference_sent = "Avui és un bon dia."
sentences = [
    "M'agrada el dia que fa.",
    "Tothom té un mal dia.",
    "És dijous.",
    "Fa un dia realment dolent",
]

reference_sent_embedding = model.encode(reference_sent)
similarity_scores = {}
for sentence in sentences:
    sent_embedding = model.encode(sentence)
    cosine_similarity = util.pytorch_cos_sim(reference_sent_embedding, sent_embedding)
    similarity_scores[float(cosine_similarity.data[0][0])] = sentence

print(f"Sentences in order of similarity to '{reference_sent}' (from max to min):")
for i, (cosine_similarity,sent) in enumerate(dict(sorted(similarity_scores.items(), reverse=True)).items()):
    print(f"{i}) '{sent}': {cosine_similarity}")

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('{MODEL_NAME}')
model = AutoModel.from_pretrained('{MODEL_NAME}')

# 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

We evaluated the model on the test set of the Catalan Semantic Text Similarity (STS-ca), and on two paraphrase identification tasks in Catalan: Parafraseja and the professional translation of PAWS into Catalan.

STS-ca (Pearson) Parafraseja (acc) PAWS-ca (acc)
0.65 0.72 0.65

Training

The model was trained with the parameters:

DataLoader:

sentence_transformers.datasets.NoDuplicatesDataLoader.NoDuplicatesDataLoader of length 147 with parameters:

{'batch_size': 128}

Loss:

sentence_transformers.losses.MultipleNegativesRankingLoss.MultipleNegativesRankingLoss with parameters:

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

Parameters of the fit()-Method:

{
    "epochs": 1,
    "evaluation_steps": 14,
    "evaluator": "sentence_transformers.evaluation.EmbeddingSimilarityEvaluator.EmbeddingSimilarityEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'torch.optim.adamw.AdamW'>",
    "optimizer_params": {
        "lr": 2e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 15,
    "weight_decay": 0.01
}

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 128, 'do_lower_case': False}) with Transformer model: XLMRobertaModel 
  (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})
)

Citing & Authors

For further information, send an email to aina@bsc.es

Downloads last month
321
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Collection including projecte-aina/ST-NLI-ca_paraphrase-multilingual-mpnet-base