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

fixed all tests, still need to check ctrl tf and pt and xlm tf

Browse Source
This commit is contained in:
patrickvonplaten
2020-03-08 21:45:55 +01:00
parent b4a3a64744
commit fbd02d4693
7 changed files with 51 additions and 49 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/test_modeling_ctrl.py
View File

@@ -246,4 +246,4 @@ class CTRLModelLanguageGenerationTest(unittest.TestCase):
] # Legal My neighbor is refusing to pay rent after 2 years and we are having to force him to pay
output_ids = model.generate(input_ids, do_sample=False)
self.assertListEqual(output_ids[0].tolist(), expected_output_ids)
self.assertListEqual(output_ids[0].numpy().tolist(), expected_output_ids)

2
tests/test_modeling_tf_gpt2.py
View File

@@ -356,7 +356,7 @@ class TFGPT2ModelLanguageGenerationTest(unittest.TestCase):
3290,
] # The dog was found in a field near the intersection of West and West Streets.\n\nThe dog
output_ids = model.generate(input_ids, do_sample=False)
self.assertListEqual(output_ids[0].tolist(), expected_output_ids)
self.assertListEqual(output_ids[0].numpy().tolist(), expected_output_ids)
@slow
def test_lm_generate_distilgpt2(self):

2
tests/test_modeling_tf_openai_gpt.py
View File

@@ -269,4 +269,4 @@ class TFOPENAIGPTModelLanguageGenerationTest(unittest.TestCase):
] # the president is a very good man. " \n " i\'m sure he is, " said the
output_ids = model.generate(input_ids, do_sample=False)
self.assertListEqual(output_ids[0].tolist(), expected_output_ids)
self.assertListEqual(output_ids[0].numpy().tolist(), expected_output_ids)

2
tests/test_modeling_tf_transfo_xl.py
View File

@@ -573,4 +573,4 @@ class TFTransfoXLModelLanguageGenerationTest(unittest.TestCase):
# TODO: add this test when trasnfo-xl-lmhead is implemented
with self.assertRaises(NotImplementedError):
model.generate(input_ids, max_length=200, do_sample=False)
# self.assertListEqual(output_ids[0].tolist(), expected_output_ids) TODO: (PVP) to add when transfo-xl is implemented
# self.assertListEqual(output_ids[0].numpy().tolist(), expected_output_ids) TODO: (PVP) to add when transfo-xl is implemented

43
tests/test_modeling_tf_xlm.py
View File

@@ -317,28 +317,29 @@ class TFXLMModelLanguageGenerationTest(unittest.TestCase):
@slow
def test_lm_generate_xlm_mlm_en_2048(self):
model = TFXLMWithLMHeadModel.from_pretrained("xlm-mlm-en-2048")
input_ids = tf.convert_to_tensor([[1, 14, 2232, 26, 1]], dtype=tf.int32) # the dog is cute
input_ids = tf.convert_to_tensor([[14, 447]], dtype=tf.int32) # the president
expected_output_ids = [
1,
14,
2232,
26,
1,
567,
26,
32,
149,
149,
149,
149,
149,
149,
149,
149,
149,
149,
149,
149,
] # The dog is nothing is it!!!!!!!!!!!! TODO (PVP): this sentence (and others I tried) does not make much sense, there seems to be a problem with xlm language generation.
447,
14,
447,
14,
447,
14,
447,
14,
447,
14,
447,
14,
447,
14,
447,
14,
447,
14,
447,
] # the president the president the president the president the president the president the president the president the president the president
# TODO(PVP): this and other input_ids I tried for generation give pretty bad results. Not sure why. Model might just not be made for auto-regressive inference
output_ids = model.generate(input_ids)
self.assertListEqual(output_ids[0].tolist(), expected_output_ids, do_sample=False)

2
tests/test_modeling_tf_xlnet.py
View File

@@ -814,4 +814,4 @@ class TFXLNetModelLanguageGenerationTest(unittest.TestCase):
output_ids = model.generate(input_ids, max_length=200, do_sample=False)
self.assertListEqual(output_ids[0].tolist(), expected_output_ids)
self.assertListEqual(output_ids[0].numpy().tolist(), expected_output_ids)

47
tests/test_modeling_xlm.py
View File

@@ -403,28 +403,29 @@ class XLMModelLanguageGenerationTest(unittest.TestCase):
@slow
def test_lm_generate_xlm_mlm_en_2048(self):
model = XLMWithLMHeadModel.from_pretrained("xlm-mlm-en-2048")
input_ids = torch.tensor([[1, 14, 2232, 26, 1]], dtype=torch.long, device=torch_device) # The dog is cute
input_ids = torch.tensor([[14, 447]], dtype=torch.long, device=torch_device) # the president
expected_output_ids = [
1,
14,
2232,
26,
1,
567,
26,
32,
149,
149,
149,
149,
149,
149,
149,
149,
149,
149,
149,
149,
] # The dog is nothing is it!!!!!!!!!!!! TODO (PVP): this sentence (and others I tried) does not make much sense, there seems to be a problem with xlm language generation.
output_ids = model.generate(input_ids)
self.assertListEqual(output_ids[0].tolist(), expected_output_ids, do_sample=False)
447,
14,
447,
14,
447,
14,
447,
14,
447,
14,
447,
14,
447,
14,
447,
14,
447,
14,
447,
] # the president the president the president the president the president the president the president the president the president the president
# TODO(PVP): this and other input_ids I tried for generation give pretty bad results. Not sure why. Model might just not be made for auto-regressive inference
output_ids = model.generate(input_ids, do_sample=False)
self.assertListEqual(output_ids[0].numpy().tolist(), expected_output_ids)