Stsb Xlm R Greek Transfer

🚀 Semantic Textual Similarity for the Greek language using Transformers and Transfer Learning

This model maps sentences and paragraphs to a 768-dimensional dense vector space, enabling tasks such as clustering and semantic search.

🚀 Quick Start

This is a sentence-transformers model that maps sentences & paragraphs to a 768-dimensional dense vector space. It can be used for tasks like clustering or semantic search. The model follows a Teacher-Student transfer learning approach described here to train an XLM-Roberta-base model on STS using parallel EN-EL sentence pairs.

✨ Features

• Maps sentences and paragraphs to a 768-dimensional dense vector space.
• Suitable for tasks like clustering and semantic search.
• Trained using a Teacher-Student transfer learning approach on parallel EN-EL sentence pairs.

📦 Installation

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

pip install -U sentence-transformers

💻 Usage Examples

Basic Usage

Using Sentence-Transformers

from sentence_transformers import SentenceTransformer
model = SentenceTransformer('{MODEL_NAME}')

sentences1 = ['Το κινητό έπεσε και έσπασε.',
             'Το κινητό έπεσε και έσπασε.',
             'Το κινητό έπεσε και έσπασε.']

sentences2 = ["H πτώση κατέστρεψε τη συσκευή.",
             "Το αυτοκίνητο έσπασε στα δυο.",
             "Ο υπουργός έπεσε και έσπασε το πόδι του."]


embeddings1 = model.encode(sentences1, convert_to_tensor=True)
embeddings2 = model.encode(sentences2, convert_to_tensor=True)

#Compute cosine-similarities (clone repo for util functions)
from sentence_transformers import util
cosine_scores = util.pytorch_cos_sim(embeddings1, embeddings2)

#Output the pairs with their score
for i in range(len(sentences1)):
    print("{} 		 {} 		 Score: {:.4f}".format(sentences1[i], sentences2[i], cosine_scores[i][i]))
    
#Outputs:
#Το κινητό έπεσε και έσπασε. 		 H πτώση κατέστρεψε τη συσκευή. 		 Score: 0.6741
#Το κινητό έπεσε και έσπασε. 		 Το αυτοκίνητο έσπασε στα δυο. 		 Score: 0.5067
#Το κινητό έπεσε και έσπασε. 		 Ο υπουργός έπεσε και έσπασε το πόδι του. 		 Score: 0.4548

Using HuggingFace Transformers

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, max pooling.
sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])

print("Sentence embeddings:")
print(sentence_embeddings)

📚 Documentation

Evaluation Results

Similarity Evaluation on STS.en-el.txt (translated manually for evaluation purposes)

We measure the semantic textual similarity (STS) between sentence pairs in different languages:

Translation

We measure the translation accuracy. Given a list with source sentences (e.g., 1000 English sentences) and a list with matching target (translated) sentences (e.g., 1000 Greek sentences), for each sentence pair, we check if their embeddings are the closest using cosine similarity. This evaluator reports accuracy (higher = better).

Training

The model was trained with the following parameters:

DataLoader: torch.utils.data.dataloader.DataLoader of length 135121 with parameters:

{'batch_size': 16, 'sampler': 'torch.utils.data.sampler.RandomSampler', 'batch_sampler': 'torch.utils.data.sampler.BatchSampler'}

Loss: sentence_transformers.losses.MSELoss.MSELoss

Parameters of the fit()-Method:

{
    "callback": null,
    "epochs": 4,
    "evaluation_steps": 1000,
    "evaluator": "sentence_transformers.evaluation.SequentialEvaluator.SequentialEvaluator",
    "max_grad_norm": 1,
    "optimizer_class": "<class 'transformers.optimization.AdamW'>",
    "optimizer_params": {
        "correct_bias": false,
        "eps": 1e-06,
        "lr": 2e-05
    },
    "scheduler": "WarmupLinear",
    "steps_per_epoch": null,
    "warmup_steps": 10000,
    "weight_decay": 0.01
}

Full Model Architecture

SentenceTransformer(
  (0): Transformer({'max_seq_length': 400, '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})
)

🔧 Technical Details

The model is based on the XLM-Roberta-base architecture and is trained using a Teacher-Student transfer learning approach on parallel EN-EL sentence pairs. It maps sentences and paragraphs to a 768-dimensional dense vector space, which can be used for various natural language processing tasks.

📄 License

This project is licensed under the Apache-2.0 license.

Acknowledgement

The research work was supported by the Hellenic Foundation for Research and Innovation (HFRI) under the HFRI PhD Fellowship grant (Fellowship Number:50, 2nd call)

Citing & Authors

Citation info for Greek model: TBD

Based on the transfer learning approach of Making Monolingual Sentence Embeddings Multilingual using Knowledge Distillation