fastNLP/tests/modules/torch/encoder/test_seq2seq_encoder.py

import pytest

from fastNLP.envs.imports import _NEED_IMPORT_TORCH

if _NEED_IMPORT_TORCH:
    import torch

    from fastNLP.modules.torch.encoder.seq2seq_encoder import TransformerSeq2SeqEncoder, LSTMSeq2SeqEncoder
    from fastNLP import Vocabulary
    from fastNLP.embeddings.torch import StaticEmbedding


@pytest.mark.torch
class TestTransformerSeq2SeqEncoder:
    def test_case(self):
        vocab = Vocabulary().add_word_lst("This is a test .".split())
        embed = StaticEmbedding(vocab, embedding_dim=5)
        encoder = TransformerSeq2SeqEncoder(embed, num_layers=2, d_model=10, n_head=2)
        words_idx = torch.LongTensor([0, 1, 2]).unsqueeze(0)
        seq_len = torch.LongTensor([3])
        encoder_output, encoder_mask = encoder(words_idx, seq_len)
        assert (encoder_output.size() == (1, 3, 10))


class TestBiLSTMEncoder:
    def test_case(self):
        vocab = Vocabulary().add_word_lst("This is a test .".split())
        embed = StaticEmbedding(vocab, embedding_dim=5)
        encoder = LSTMSeq2SeqEncoder(embed, hidden_size=5, num_layers=1)
        words_idx = torch.LongTensor([0, 1, 2]).unsqueeze(0)
        seq_len = torch.LongTensor([3])

        encoder_output, encoder_mask = encoder(words_idx, seq_len)
        assert (encoder_mask.size() == (1, 3))