def get_contextual_embeddings()

in lama/modules/bert_connector.py [0:0]

• 13 lines of code
• 8 McCabe index (conditional complexity)

    def get_contextual_embeddings(self, sentences_list, try_cuda=True):

        # assume in input 1 or 2 sentences - in general, it considers only the first 2 sentences
        if not sentences_list:
            return None
        if try_cuda:
            self.try_cuda()

        tokens_tensor, segments_tensor, attention_mask_tensor, masked_indices_list, tokenized_text_list = self.__get_input_tensors_batch(sentences_list)

        with torch.no_grad():
            all_encoder_layers, _ = self.bert_model(
                tokens_tensor.to(self._model_device),
                segments_tensor.to(self._model_device))

        all_encoder_layers = [layer.cpu() for layer in all_encoder_layers]

        sentence_lengths = [len(x) for x in tokenized_text_list]

        # all_encoder_layers: a list of the full sequences of encoded-hidden-states at the end
        # of each attention block (i.e. 12 full sequences for BERT-base, 24 for BERT-large), each
        # encoded-hidden-state is a torch.FloatTensor of size [batch_size, sequence_length, hidden_size]
        return all_encoder_layers, sentence_lengths, tokenized_text_list