paraphrase Tiny B E R T L6 v2

Introduction

The paraphrase-TinyBERT-L6-v2 is a model from the Sentence Transformers library. It is designed to map sentences and paragraphs to a 768-dimensional dense vector space, enabling tasks such as clustering and semantic search.

Architecture

The model architecture employs a Transformer with a maximum sequence length of 128. It includes a Pooling layer configured to use mean pooling for sentence embeddings, leveraging a BERT model for token embedding.

Training

This model follows the Sentence-BERT framework, which uses a Siamese BERT-Network to derive sentence embeddings. It is optimized for sentence similarity tasks.

Guide: Running Locally

Basic Steps

1. Install Sentence Transformers
   Ensure that the sentence-transformers library is installed:

   pip install -U sentence-transformers
   

2. Use the Model
   Load and use the model for generating sentence embeddings:

   from sentence_transformers import SentenceTransformer
   
   model = SentenceTransformer('sentence-transformers/paraphrase-TinyBERT-L6-v2')
   sentences = ["This is an example sentence", "Each sentence is converted"]
   embeddings = model.encode(sentences)
   print(embeddings)
   

3. Alternative Using Hugging Face Transformers
   You can also use the model without the Sentence Transformers library by manually handling the token embeddings and pooling:

   from transformers import AutoTokenizer, AutoModel
   import torch
   
   tokenizer = AutoTokenizer.from_pretrained('sentence-transformers/paraphrase-TinyBERT-L6-v2')
   model = AutoModel.from_pretrained('sentence-transformers/paraphrase-TinyBERT-L6-v2')
   
   sentences = ['This is an example sentence', 'Each sentence is converted']
   encoded_input = tokenizer(sentences, padding=True, truncation=True, return_tensors='pt')
   
   with torch.no_grad():
       model_output = model(**encoded_input)
   
   def mean_pooling(model_output, attention_mask):
       token_embeddings = model_output[0]
       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)
   
   sentence_embeddings = mean_pooling(model_output, encoded_input['attention_mask'])
   print("Sentence embeddings:")
   print(sentence_embeddings)
   

Cloud GPUs

For faster processing, consider using cloud GPU services such as AWS, Google Cloud, or Azure.

License

The model is licensed under the Apache-2.0 License.