SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

[ci] Re-run integration ground truth from fairseq

Browse Source 
Adopted best practice set by @patrickvonplaten of commenting lines run on fairseq, for easy comparison

also see #3020
This commit is contained in:
Julien Chaumond
2020-03-03 15:31:31 -05:00
parent 5b396457e5
commit f631e01d2c
4 changed files with 39 additions and 24 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
tests/test_modeling_xlm_roberta.py
View File

@@ -30,14 +30,13 @@ class XLMRobertaModelIntegrationTest(unittest.TestCase):
@slow
def test_xlm_roberta_base(self):
model = XLMRobertaModel.from_pretrained("xlm-roberta-base")
input_ids = torch.tensor([0, 581, 10269, 83, 99942, 136, 60742, 23, 70, 80583, 18276, 2]).unsqueeze(
0
) # The dog is cute and lives in the garden house
input_ids = torch.tensor([[0, 581, 10269, 83, 99942, 136, 60742, 23, 70, 80583, 18276, 2]])
# The dog is cute and lives in the garden house
expected_output_shape = torch.Size((1, 12, 768)) # batch_size, sequence_length, embedding_vector_dim
expected_output_values_last_dim = torch.tensor(
[-0.0101, 0.1218, -0.0803, 0.0801, 0.1327, 0.0776, -0.1215, 0.2383, 0.3338, 0.3106, 0.0300, 0.0252]
).unsqueeze(0)
[[-0.0101, 0.1218, -0.0803, 0.0801, 0.1327, 0.0776, -0.1215, 0.2383, 0.3338, 0.3106, 0.0300, 0.0252]]
)
# xlmr = torch.hub.load('pytorch/fairseq', 'xlmr.base')
# xlmr.eval()
# expected_output_values_last_dim = xlmr.extract_features(input_ids[0])[:, :, -1]
@@ -50,14 +49,13 @@ class XLMRobertaModelIntegrationTest(unittest.TestCase):
@slow
def test_xlm_roberta_large(self):
model = XLMRobertaModel.from_pretrained("xlm-roberta-large")
input_ids = torch.tensor([0, 581, 10269, 83, 99942, 136, 60742, 23, 70, 80583, 18276, 2]).unsqueeze(
0
) # The dog is cute and lives in the garden house
input_ids = torch.tensor([[0, 581, 10269, 83, 99942, 136, 60742, 23, 70, 80583, 18276, 2]])
# The dog is cute and lives in the garden house
expected_output_shape = torch.Size((1, 12, 1024)) # batch_size, sequence_length, embedding_vector_dim
expected_output_values_last_dim = torch.tensor(
[-0.0699, -0.0318, 0.0705, -0.1241, 0.0999, -0.0520, 0.1004, -0.1838, -0.4704, 0.1437, 0.0821, 0.0126]
).unsqueeze(0)
[[-0.0699, -0.0318, 0.0705, -0.1241, 0.0999, -0.0520, 0.1004, -0.1838, -0.4704, 0.1437, 0.0821, 0.0126]]
)
# xlmr = torch.hub.load('pytorch/fairseq', 'xlmr.large')
# xlmr.eval()
# expected_output_values_last_dim = xlmr.extract_features(input_ids[0])[:, :, -1]