sentence_camembert_base.py

from typing import  Dict, List, Any
from optimum.onnxruntime import ORTModelForFeatureExtraction
from transformers import AutoTokenizer
import torch.nn.functional as F
import torch

# copied from the model card
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)


class sentence_embeddings(path = '.'):
    def __init__(self, path):
        # load the optimized model
        self.model = ORTModelForFeatureExtraction.from_pretrained(path, file_name="model_quantized.onnx")
        self.tokenizer = AutoTokenizer.from_pretrained(path)

    def __call__(self, data: Any) -> List[List[Dict[str, float]]]:
        """
        Args:
            data (:obj:):
                includes the input data and the parameters for the inference.
        Return:
            A :obj:`list`:. The list contains the embeddings of the inference inputs
        """
        inputs = data.get("inputs", data)

        # tokenize the input
        encoded_inputs = self.tokenizer(inputs, padding=True, truncation=True, return_tensors='pt')
        # run the model
        outputs = self.model(**encoded_inputs)
        # Perform pooling
        embeddings = mean_pooling(outputs, encoded_inputs['attention_mask'])
        # Normalize embeddings
        embeddings = F.normalize(embeddings, p=2, dim=1)
        # postprocess the prediction
        return {'embeddings': embeddings.tolist()}