fastqa.py 文件源码

def __init__(self, shared_resources: SharedResources):
        super(FastQAPyTorchModule, self).__init__()
        self._shared_resources = shared_resources
        input_size = shared_resources.config["repr_dim_input"]
        size = shared_resources.config["repr_dim"]
        self._size = size
        self._with_char_embeddings = self._shared_resources.config.get("with_char_embeddings", False)

        # modules & parameters
        if self._with_char_embeddings:
            self._conv_char_embedding = embedding.ConvCharEmbeddingModule(
                len(shared_resources.char_vocab), size)
            self._embedding_projection = nn.Linear(size + input_size, size)
            self._embedding_highway = Highway(size, 1)
            self._v_wiq_w = nn.Parameter(torch.ones(1, 1, input_size + size))
            input_size = size
        else:
            self._v_wiq_w = nn.Parameter(torch.ones(1, 1, input_size))

        self._bilstm = BiLSTM(input_size + 2, size)
        self._answer_layer = FastQAAnswerModule(shared_resources)

        # [size, 2 * size]
        self._question_projection = nn.Parameter(torch.cat([torch.eye(size), torch.eye(size)], dim=1))
        self._support_projection = nn.Parameter(torch.cat([torch.eye(size), torch.eye(size)], dim=1))