l3cube-pune/indic-sentence-bert-nli

IndicSBERT

This is a MuRIL model (google/muril-base-cased) trained on the NLI dataset of ten major Indian Languages.
The single model works for English, Hindi, Marathi, Kannada, Tamil, Telugu, Gujarati, Oriya, Punjabi, Malayalam, and Bengali. The model also has cross-lingual capabilities.
Released as a part of project MahaNLP: https://github.com/l3cube-pune/MarathiNLP

A better sentence similarity model (fine-tuned version of this model) is shared here: https://huggingface.co/l3cube-pune/indic-sentence-similarity-sbert

More details on the dataset, models, and baseline results can be found in our [paper] (https://arxiv.org/abs/2304.11434)

@article{deode2023l3cube,
  title={L3Cube-IndicSBERT: A simple approach for learning cross-lingual sentence representations using multilingual BERT},
  author={Deode, Samruddhi and Gadre, Janhavi and Kajale, Aditi and Joshi, Ananya and Joshi, Raviraj},
  journal={arXiv preprint arXiv:2304.11434},
  year={2023}
}

monolingual Indic SBERT paper
multilingual Indic SBERT paper

Other Monolingual Indic sentence BERT models are listed below:
Marathi SBERT
Hindi SBERT
Kannada SBERT
Telugu SBERT
Malayalam SBERT
Tamil SBERT
Gujarati SBERT
Oriya SBERT
Bengali SBERT
Punjabi SBERT
Indic SBERT (multilingual)

Other Monolingual similarity models are listed below:
Marathi Similarity
Hindi Similarity
Kannada Similarity
Telugu Similarity
Malayalam Similarity
Tamil Similarity
Gujarati Similarity
Oriya Similarity
Bengali Similarity
Punjabi Similarity
Indic Similarity (multilingual)

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

model = SentenceTransformer('{MODEL_NAME}')
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('{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)