def test_no_token_type_layer(self):
params = copy.deepcopy(self.params_dict)
params["type_vocab_size"] = 0
params = Params(params)
module = TransformerEmbeddings.from_params(params)
assert len(module.embeddings) == 2
def setup_method(self):
super().setup_method()
self.params_dict = {key: val for key, val in PARAMS_DICT.items()}
params = Params(copy.deepcopy(self.params_dict))
self.transformer_embeddings = TransformerEmbeddings.from_params(params)
def test_output_size(params):
input_ids = torch.tensor([[1, 2]])
token_type_ids = torch.tensor([[1, 0]], dtype=torch.long)
position_ids = torch.tensor([[0, 1]])
params["output_size"] = 7
module = TransformerEmbeddings.from_params(params)
output = module(input_ids=input_ids,
token_type_ids=token_type_ids,
position_ids=position_ids)
assert output.shape[-1] == 7
def test_output_size(self):
input_ids = torch.tensor([[1, 2]])
token_type_ids = torch.tensor([[1, 0]], dtype=torch.long)
position_ids = torch.tensor([[0, 1]])
params = copy.deepcopy(self.params_dict)
params["output_size"] = 7
params = Params(params)
module = TransformerEmbeddings.from_params(params)
output = module.forward(
input_ids=input_ids, token_type_ids=token_type_ids, position_ids=position_ids
)
assert output.shape[-1] == 7
def transformer_embeddings(params):
return TransformerEmbeddings.from_params(params.duplicate())
def test_no_token_type_layer(params):
params["type_vocab_size"] = 0
module = TransformerEmbeddings.from_params(params)
assert len(module.embeddings) == 2
