---
pipeline_tag: sentence-similarity
tags:
  - sentence-transformers
  - feature-extraction
  - sentence-similarity
  - transformers
  - dpr
widget:
- source_sentence: "আমি বাংলায় গান গাই"
  sentences:
    - "I sing in Bangla"
    - "I sing in Bengali"
    - "I sing in English"
    - "আমি গান গাই না "
  example_title: "Singing"
---

# `semantic_xlmr`

This is a [sentence-transformers](https://www.SBERT.net) model: It maps sentences & paragraphs to a 768 dimensional dense vector space and can be used for tasks like **clustering** or **semantic search**.

<!--- Describe your model here -->

## Model Details

- Model name: semantic_xlmr
- Model version: 1.0
- Architecture: Sentence Transformer
- Language: Multilingual ( fine-tuned for Bengali Language)

## Training

The model was fine-tuned using  **Multilingual Knowledge Distillation** method. We took `paraphrase-distilroberta-base-v2` as the teacher model and  `xlm-roberta-large` as the student model.



![image](https://i.ibb.co/8Xrgnfr/sentence-transformer-model.png)

## Intended Use:

- **Primary Use Case:** Semantic similarity, clustering, and semantic searches
- **Potential Use Cases:** Document retrieval, information retrieval, recommendation systems, chatbot systems , FAQ system

## Usage

### Using Sentence-Transformers

Using this model becomes easy when you have [sentence-transformers](https://www.SBERT.net) installed:

```
pip install -U sentence-transformers
```

Then you can use the model like this:

```python
from sentence_transformers import SentenceTransformer
sentences = ["I sing in bengali", "আমি বাংলায় গান গাই"]

model = SentenceTransformer('headlesstech/semantic_xlmr')
embeddings = model.encode(sentences)
print(embeddings)
```

### Using HuggingFace Transformers

Without [sentence-transformers](https://www.SBERT.net), 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.

```python
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 = ["I sing in bengali", "আমি বাংলায় গান গাই"]

# Load model from HuggingFace Hub
tokenizer = AutoTokenizer.from_pretrained('headlesstech/semantic_xlmr')
model = AutoModel.from_pretrained('headlesstech/semantic_xlmr')

# 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)
```

## Full Model Architecture

```
SentenceTransformer(
  (0): Transformer({'max_seq_length': 512, '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})
)
```