Compare commits
6
Commits
tapas-final
..
tapas
| Author | SHA1 | Date | |
|---|---|---|---|
 LysandreJik
|
005fcce7cc | ||
|
Lysandre
|
5102527eee | ||
|
Lysandre
|
6861aa9cc4 | ||
|
Lysandre
|
0691e49741 | ||
|
LysandreJik
|
ca2725ee79 | ||
|
LysandreJik
|
ba66c4c817 |
File diff suppressed because it is too large
Load Diff
@@ -12,16 +12,15 @@
|
||||
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
# See the License for the specific language governing permissions and
|
||||
# limitations under the License.
|
||||
import inspect
|
||||
import functools
|
||||
import os
|
||||
import shutil
|
||||
import tempfile
|
||||
import unittest
|
||||
from typing import List, Tuple
|
||||
|
||||
import pandas as pd
|
||||
|
||||
from transformers import AddedToken
|
||||
import regex as re
|
||||
from transformers import AddedToken, PreTrainedTokenizerBase
|
||||
from transformers.testing_utils import require_tokenizers, slow
|
||||
from transformers.tokenization_tapas import (
|
||||
VOCAB_FILES_NAMES,
|
||||
@@ -44,65 +43,24 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
space_between_special_tokens = True
|
||||
from_pretrained_filter = filter_non_english
|
||||
|
||||
def get_table(
|
||||
self,
|
||||
tokenizer: TapasTokenizer,
|
||||
length=5,
|
||||
):
|
||||
toks = [tokenizer.decode([i], clean_up_tokenization_spaces=False) for i in range(len(tokenizer))]
|
||||
|
||||
if length == 0:
|
||||
data = {}
|
||||
else:
|
||||
data = {toks[0]: [toks[tok] for tok in range(1, length)]}
|
||||
|
||||
table = pd.DataFrame.from_dict(data)
|
||||
|
||||
return table
|
||||
|
||||
def get_table_and_query(
|
||||
self,
|
||||
tokenizer: TapasTokenizer,
|
||||
add_special_tokens: bool = True,
|
||||
length=5,
|
||||
):
|
||||
toks = [tokenizer.decode([i], clean_up_tokenization_spaces=False) for i in range(len(tokenizer))]
|
||||
table = self.get_table(tokenizer, length=length - 3)
|
||||
query = " ".join(toks[:3])
|
||||
|
||||
return table, query
|
||||
|
||||
def get_clean_sequence(
|
||||
self,
|
||||
tokenizer: TapasTokenizer,
|
||||
with_prefix_space=False,
|
||||
max_length=20,
|
||||
min_length=5,
|
||||
empty_table: bool = False,
|
||||
add_special_tokens: bool = True,
|
||||
return_table_and_query: bool = False,
|
||||
):
|
||||
|
||||
toks = [tokenizer.decode([i], clean_up_tokenization_spaces=False) for i in range(len(tokenizer))]
|
||||
|
||||
self, tokenizer: TapasTokenizer, empty_table: bool = False, add_special_tokens: bool = True
|
||||
) -> Tuple[pd.DataFrame, str, List[int]]:
|
||||
if empty_table:
|
||||
table = pd.DataFrame.from_dict({})
|
||||
query = " ".join(toks[:min_length])
|
||||
else:
|
||||
data = {toks[0]: [toks[tok] for tok in range(1, min_length - 3)]}
|
||||
data = {
|
||||
"Actors": ["Brad Pitt", "Leonardo Di Caprio", "George Clooney"],
|
||||
"Age": ["56", "45", "59"],
|
||||
"Number of movies": ["87", "53", "69"],
|
||||
"Date of birth": ["18 december 1963", "11 november 1974", "6 may 1961"],
|
||||
}
|
||||
table = pd.DataFrame.from_dict(data)
|
||||
query = " ".join(toks[:3])
|
||||
query = "Which actor appeared in the least number of movies?"
|
||||
|
||||
output_ids = tokenizer.encode(table, query, add_special_tokens=add_special_tokens)
|
||||
output_txt = tokenizer.decode(output_ids)
|
||||
inputs = tokenizer.encode(table, query, add_special_tokens=add_special_tokens)
|
||||
|
||||
assert len(output_ids) >= min_length, "Update the code to generate the sequences so that they are larger"
|
||||
assert len(output_ids) <= max_length, "Update the code to generate the sequences so that they are smaller"
|
||||
|
||||
if return_table_and_query:
|
||||
return output_txt, output_ids, table, query
|
||||
|
||||
return output_txt, output_ids
|
||||
return table, query, inputs
|
||||
|
||||
# def get_clean_sequence(self, tokenizer, with_prefix_space=False, max_length=20, min_length=5) -> Tuple[str, list]:
|
||||
# data = {
|
||||
@@ -376,164 +334,13 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
)
|
||||
self.assertEqual([e[0] for e in expected_results], tokens["offset_mapping"])
|
||||
|
||||
def test_tapas_integration_test(self):
|
||||
data = {
|
||||
"Actors": ["Brad Pitt", "Leonardo Di Caprio", "George Clooney"],
|
||||
"Age": ["56", "45", "59"],
|
||||
"Number of movies": ["87", "53", "69"],
|
||||
"Date of birth": ["18 december 1963", "11 november 1974", "6 may 1961"],
|
||||
}
|
||||
queries = [
|
||||
"When was Brad Pitt born?",
|
||||
"Which actor appeared in the least number of movies?",
|
||||
"What is the average number of movies?",
|
||||
]
|
||||
table = pd.DataFrame.from_dict(data)
|
||||
|
||||
# TODO: Should update this in the future
|
||||
tokenizer = TapasTokenizer.from_pretrained("lysandre/tapas-temporary-repo", model_max_length=512)
|
||||
|
||||
expected_results = {
|
||||
"input_ids": [
|
||||
101,
|
||||
2043,
|
||||
2001,
|
||||
8226,
|
||||
15091,
|
||||
2141,
|
||||
1029,
|
||||
102,
|
||||
5889,
|
||||
2287,
|
||||
2193,
|
||||
1997,
|
||||
5691,
|
||||
3058,
|
||||
1997,
|
||||
4182,
|
||||
8226,
|
||||
15091,
|
||||
5179,
|
||||
6584,
|
||||
2324,
|
||||
2285,
|
||||
3699,
|
||||
14720,
|
||||
4487,
|
||||
6178,
|
||||
9488,
|
||||
3429,
|
||||
5187,
|
||||
2340,
|
||||
2281,
|
||||
3326,
|
||||
2577,
|
||||
18856,
|
||||
7828,
|
||||
3240,
|
||||
5354,
|
||||
6353,
|
||||
1020,
|
||||
2089,
|
||||
3777,
|
||||
],
|
||||
"attention_mask": [
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
],
|
||||
"token_type_ids": [
|
||||
[0, 0, 0, 0, 0, 0, 0],
|
||||
[0, 0, 0, 0, 0, 0, 0],
|
||||
[0, 0, 0, 0, 0, 0, 0],
|
||||
[0, 0, 0, 0, 0, 0, 0],
|
||||
[0, 0, 0, 0, 0, 0, 0],
|
||||
[0, 0, 0, 0, 0, 0, 0],
|
||||
[0, 0, 0, 0, 0, 0, 0],
|
||||
[0, 0, 0, 0, 0, 0, 0],
|
||||
[1, 1, 0, 0, 0, 0, 0],
|
||||
[1, 2, 0, 0, 0, 0, 0],
|
||||
[1, 3, 0, 0, 0, 0, 0],
|
||||
[1, 3, 0, 0, 0, 0, 0],
|
||||
[1, 3, 0, 0, 0, 0, 0],
|
||||
[1, 4, 0, 0, 0, 0, 0],
|
||||
[1, 4, 0, 0, 0, 0, 0],
|
||||
[1, 4, 0, 0, 0, 0, 0],
|
||||
[1, 1, 1, 0, 0, 0, 0],
|
||||
[1, 1, 1, 0, 0, 0, 0],
|
||||
[1, 2, 1, 0, 2, 2, 0],
|
||||
[1, 3, 1, 0, 3, 1, 0],
|
||||
[1, 4, 1, 0, 2, 2, 0],
|
||||
[1, 4, 1, 0, 2, 2, 0],
|
||||
[1, 4, 1, 0, 2, 2, 0],
|
||||
[1, 1, 2, 0, 0, 0, 0],
|
||||
[1, 1, 2, 0, 0, 0, 0],
|
||||
[1, 1, 2, 0, 0, 0, 0],
|
||||
[1, 1, 2, 0, 0, 0, 0],
|
||||
[1, 2, 2, 0, 1, 3, 0],
|
||||
[1, 3, 2, 0, 1, 3, 0],
|
||||
[1, 4, 2, 0, 3, 1, 0],
|
||||
[1, 4, 2, 0, 3, 1, 0],
|
||||
[1, 4, 2, 0, 3, 1, 0],
|
||||
[1, 1, 3, 0, 0, 0, 0],
|
||||
[1, 1, 3, 0, 0, 0, 0],
|
||||
[1, 1, 3, 0, 0, 0, 0],
|
||||
[1, 1, 3, 0, 0, 0, 0],
|
||||
[1, 2, 3, 0, 3, 1, 0],
|
||||
[1, 3, 3, 0, 2, 2, 0],
|
||||
[1, 4, 3, 0, 1, 3, 0],
|
||||
[1, 4, 3, 0, 1, 3, 0],
|
||||
[1, 4, 3, 0, 1, 3, 0],
|
||||
],
|
||||
}
|
||||
|
||||
new_encoded_inputs = tokenizer.encode_plus(table=table, query=queries[0], padding="max_length")
|
||||
|
||||
self.assertDictEqual(new_encoded_inputs, expected_results)
|
||||
|
||||
def test_add_special_tokens(self):
|
||||
tokenizers: List[TapasTokenizer] = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
input_table = self.get_table(tokenizer, length=0)
|
||||
input_table, input_query, ids = self.get_clean_sequence(
|
||||
tokenizer, empty_table=True, add_special_tokens=False
|
||||
)
|
||||
|
||||
special_token = "[SPECIAL_TOKEN]"
|
||||
|
||||
@@ -541,14 +348,21 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
encoded_special_token = tokenizer.encode(input_table, special_token, add_special_tokens=False)
|
||||
self.assertEqual(len(encoded_special_token), 1)
|
||||
|
||||
decoded = tokenizer.decode(encoded_special_token, skip_special_tokens=True)
|
||||
text = tokenizer.decode(ids + encoded_special_token, clean_up_tokenization_spaces=False)
|
||||
encoded = tokenizer.encode(input_table, text, add_special_tokens=False)
|
||||
|
||||
input_encoded = tokenizer.encode(input_table, input_query, add_special_tokens=False)
|
||||
special_token_id = tokenizer.encode(input_table, special_token, add_special_tokens=False)
|
||||
self.assertEqual(encoded, input_encoded + special_token_id)
|
||||
|
||||
decoded = tokenizer.decode(encoded, skip_special_tokens=True)
|
||||
self.assertTrue(special_token not in decoded)
|
||||
|
||||
def test_add_tokens_tokenizer(self):
|
||||
tokenizers: List[TapasTokenizer] = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
table = self.get_table(tokenizer, length=0)
|
||||
input_table, input_query, ids = self.get_clean_sequence(tokenizer, empty_table=True)
|
||||
vocab_size = tokenizer.vocab_size
|
||||
all_size = len(tokenizer)
|
||||
|
||||
@@ -568,7 +382,9 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
self.assertEqual(added_toks, len(new_toks))
|
||||
self.assertEqual(all_size_2, all_size + len(new_toks))
|
||||
|
||||
tokens = tokenizer.encode(table, "aaaaa bbbbbb low cccccccccdddddddd l", add_special_tokens=False)
|
||||
tokens = tokenizer.encode(
|
||||
input_table, "aaaaa bbbbbb low cccccccccdddddddd l", add_special_tokens=False
|
||||
)
|
||||
|
||||
self.assertGreaterEqual(len(tokens), 4)
|
||||
self.assertGreater(tokens[0], tokenizer.vocab_size - 1)
|
||||
@@ -585,7 +401,7 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
self.assertEqual(all_size_3, all_size_2 + len(new_toks_2))
|
||||
|
||||
tokens = tokenizer.encode(
|
||||
table,
|
||||
input_table,
|
||||
">>>>|||<||<<|<< aaaaabbbbbb low cccccccccdddddddd <<<<<|||>|>>>>|> l",
|
||||
add_special_tokens=False,
|
||||
)
|
||||
@@ -603,7 +419,7 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
tokenizers = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
table = self.get_table(tokenizer, length=0)
|
||||
input_table, input_query, ids = self.get_clean_sequence(tokenizer, empty_table=True)
|
||||
|
||||
# new_toks = ["[ABC]", "[DEF]"] # TODO(thom) add this one back when Rust toks are ready: , "GHI IHG"]
|
||||
new_toks = [AddedToken("[ABC]", normalized=False), AddedToken("[DEF]", normalized=False)]
|
||||
@@ -613,7 +429,7 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
output = "[ABC] [DEF] [ABC] [DEF]"
|
||||
else:
|
||||
output = input
|
||||
encoded = tokenizer.encode(table, input, add_special_tokens=False)
|
||||
encoded = tokenizer.encode(input_table, input, add_special_tokens=False)
|
||||
decoded = tokenizer.decode(encoded, spaces_between_special_tokens=self.space_between_special_tokens)
|
||||
self.assertIn(decoded, [output, output.lower()])
|
||||
|
||||
@@ -621,7 +437,7 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
tokenizers = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
table = self.get_table(tokenizer, length=0)
|
||||
input_table, input_query, ids = self.get_clean_sequence(tokenizer, empty_table=True)
|
||||
sequence = "Sequence"
|
||||
|
||||
# check correct behaviour if no pad_token_id exists and add it eventually
|
||||
@@ -631,7 +447,7 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
padding_idx = tokenizer.pad_token_id
|
||||
token_type_padding_idx = tokenizer.pad_token_type_id
|
||||
|
||||
encoded_sequence = tokenizer.encode_plus(table, sequence, return_special_tokens_mask=True)
|
||||
encoded_sequence = tokenizer.encode_plus(input_table, sequence, return_special_tokens_mask=True)
|
||||
input_ids = encoded_sequence["input_ids"]
|
||||
special_tokens_mask = encoded_sequence["special_tokens_mask"]
|
||||
sequence_length = len(input_ids)
|
||||
@@ -640,7 +456,7 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
tokenizer.padding_side = "right"
|
||||
|
||||
not_padded_sequence = tokenizer.encode_plus(
|
||||
table,
|
||||
input_table,
|
||||
sequence,
|
||||
padding=True,
|
||||
return_special_tokens_mask=True,
|
||||
@@ -655,7 +471,7 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
assert special_tokens_mask == not_padded_special_tokens_mask
|
||||
|
||||
not_padded_sequence = tokenizer.encode_plus(
|
||||
table,
|
||||
input_table,
|
||||
sequence,
|
||||
padding=False,
|
||||
return_special_tokens_mask=True,
|
||||
@@ -673,7 +489,7 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
tokenizer.padding_side = "right"
|
||||
|
||||
right_padded_sequence = tokenizer.encode_plus(
|
||||
table,
|
||||
input_table,
|
||||
sequence,
|
||||
max_length=sequence_length + padding_size,
|
||||
padding="max_length",
|
||||
@@ -691,7 +507,7 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
# Test left padding
|
||||
tokenizer.padding_side = "left"
|
||||
left_padded_sequence = tokenizer.encode_plus(
|
||||
table,
|
||||
input_table,
|
||||
sequence,
|
||||
max_length=sequence_length + padding_size,
|
||||
padding="max_length",
|
||||
@@ -727,12 +543,12 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
tokenizers = self.get_tokenizers()
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
table = self.get_table(tokenizer, length=0)
|
||||
input_table, input_query, ids = self.get_clean_sequence(tokenizer, empty_table=True)
|
||||
input_text, output_text = self.get_input_output_texts(tokenizer)
|
||||
|
||||
tokens = tokenizer.tokenize(input_text)
|
||||
ids = tokenizer.convert_tokens_to_ids(tokens)
|
||||
ids_2 = tokenizer.encode(table, input_text, add_special_tokens=False)
|
||||
ids_2 = tokenizer.encode(input_table, input_text, add_special_tokens=False)
|
||||
self.assertListEqual(ids, ids_2)
|
||||
|
||||
tokens_2 = tokenizer.convert_ids_to_tokens(ids)
|
||||
@@ -746,450 +562,12 @@ class TapasTokenizationTest(TokenizerTesterMixin, unittest.TestCase):
|
||||
tokenizers = self.get_tokenizers(fast=False, do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
table, query = self.get_table_and_query(tokenizer)
|
||||
input_table, input_query, ids = self.get_clean_sequence(tokenizer)
|
||||
|
||||
if (
|
||||
tokenizer.build_inputs_with_special_tokens.__qualname__.split(".")[0] != "PreTrainedTokenizer"
|
||||
and "token_type_ids" in tokenizer.model_input_names
|
||||
):
|
||||
information = tokenizer.encode_plus(table, query, add_special_tokens=True)
|
||||
information = tokenizer.encode_plus(input_table, input_query, add_special_tokens=True)
|
||||
sequences, mask = information["input_ids"], information["token_type_ids"]
|
||||
self.assertEqual(len(sequences), len(mask))
|
||||
|
||||
@unittest.skip("TAPAS tokenizer only handles two sequences.")
|
||||
def test_maximum_encoding_length_pair_input(self):
|
||||
pass
|
||||
|
||||
@unittest.skip("TAPAS tokenizer only handles two sequences.")
|
||||
def test_maximum_encoding_length_single_input(self):
|
||||
pass
|
||||
|
||||
def test_number_of_added_tokens(self):
|
||||
tokenizers = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
|
||||
table, query = self.get_table_and_query(tokenizer)
|
||||
|
||||
sequences = tokenizer.encode(table, query, add_special_tokens=False)
|
||||
attached_sequences = tokenizer.encode(table, query, add_special_tokens=True)
|
||||
|
||||
# Method is implemented (e.g. not GPT-2)
|
||||
if len(attached_sequences) != 2:
|
||||
self.assertEqual(
|
||||
tokenizer.num_special_tokens_to_add(pair=True), len(attached_sequences) - len(sequences)
|
||||
)
|
||||
|
||||
def test_padding_to_max_length(self):
|
||||
"""We keep this test for backward compatibility but it should be removed when `pad_to_max_length` will be deprecated"""
|
||||
tokenizers = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
table = self.get_table(tokenizer)
|
||||
sequence = "Sequence"
|
||||
padding_size = 10
|
||||
|
||||
# check correct behaviour if no pad_token_id exists and add it eventually
|
||||
self._check_no_pad_token_padding(tokenizer, sequence)
|
||||
|
||||
padding_idx = tokenizer.pad_token_id
|
||||
|
||||
# Check that it correctly pads when a maximum length is specified along with the padding flag set to True
|
||||
tokenizer.padding_side = "right"
|
||||
encoded_sequence = tokenizer.encode(table, sequence)
|
||||
sequence_length = len(encoded_sequence)
|
||||
# FIXME: the next line should be padding(max_length) to avoid warning
|
||||
padded_sequence = tokenizer.encode(
|
||||
table, sequence, max_length=sequence_length + padding_size, pad_to_max_length=True
|
||||
)
|
||||
padded_sequence_length = len(padded_sequence)
|
||||
assert sequence_length + padding_size == padded_sequence_length
|
||||
assert encoded_sequence + [padding_idx] * padding_size == padded_sequence
|
||||
|
||||
# Check that nothing is done when a maximum length is not specified
|
||||
encoded_sequence = tokenizer.encode(table, sequence)
|
||||
sequence_length = len(encoded_sequence)
|
||||
|
||||
tokenizer.padding_side = "right"
|
||||
padded_sequence_right = tokenizer.encode(table, sequence, pad_to_max_length=True)
|
||||
padded_sequence_right_length = len(padded_sequence_right)
|
||||
assert sequence_length == padded_sequence_right_length
|
||||
assert encoded_sequence == padded_sequence_right
|
||||
|
||||
def test_padding_to_multiple_of(self):
|
||||
tokenizers = self.get_tokenizers()
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
if tokenizer.pad_token is None:
|
||||
self.skipTest("No padding token.")
|
||||
else:
|
||||
empty_tokens = tokenizer("", padding=True, pad_to_multiple_of=8)
|
||||
normal_tokens = tokenizer("This is a sample input", padding=True, pad_to_multiple_of=8)
|
||||
for key, value in empty_tokens.items():
|
||||
self.assertEqual(len(value) % 8, 0, "BatchEncoding.{} is not multiple of 8".format(key))
|
||||
for key, value in normal_tokens.items():
|
||||
self.assertEqual(len(value) % 8, 0, "BatchEncoding.{} is not multiple of 8".format(key))
|
||||
|
||||
normal_tokens = tokenizer("This", pad_to_multiple_of=8)
|
||||
for key, value in normal_tokens.items():
|
||||
self.assertNotEqual(len(value) % 8, 0, "BatchEncoding.{} is not multiple of 8".format(key))
|
||||
|
||||
# Should also work with truncation
|
||||
normal_tokens = tokenizer("This", padding=True, truncation=True, pad_to_multiple_of=8)
|
||||
for key, value in normal_tokens.items():
|
||||
self.assertEqual(len(value) % 8, 0, "BatchEncoding.{} is not multiple of 8".format(key))
|
||||
|
||||
# truncation to something which is not a multiple of pad_to_multiple_of raises an error
|
||||
self.assertRaises(
|
||||
ValueError,
|
||||
tokenizer.__call__,
|
||||
"This",
|
||||
padding=True,
|
||||
truncation=True,
|
||||
max_length=12,
|
||||
pad_to_multiple_of=8,
|
||||
)
|
||||
|
||||
def test_call(self):
|
||||
# Tests that all call wrap to encode_plus and batch_encode_plus
|
||||
tokenizers = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
sequences = [
|
||||
"Testing batch encode plus",
|
||||
"Testing batch encode plus with different sequence lengths",
|
||||
"Testing batch encode plus with different sequence lengths correctly pads",
|
||||
]
|
||||
|
||||
# Test not batched
|
||||
table = self.get_table(tokenizer, length=0)
|
||||
encoded_sequences_1 = tokenizer.encode_plus(table, sequences[0])
|
||||
encoded_sequences_2 = tokenizer(table, sequences[0])
|
||||
self.assertEqual(encoded_sequences_1, encoded_sequences_2)
|
||||
|
||||
# Test not batched pairs
|
||||
table = self.get_table(tokenizer, length=10)
|
||||
encoded_sequences_1 = tokenizer.encode_plus(table, sequences[1])
|
||||
encoded_sequences_2 = tokenizer(table, sequences[1])
|
||||
self.assertEqual(encoded_sequences_1, encoded_sequences_2)
|
||||
|
||||
# Test batched
|
||||
table = self.get_table(tokenizer, length=0)
|
||||
encoded_sequences_1 = tokenizer.batch_encode_plus(table, sequences)
|
||||
encoded_sequences_2 = tokenizer(table, sequences)
|
||||
self.assertEqual(encoded_sequences_1, encoded_sequences_2)
|
||||
|
||||
def test_batch_encode_plus_batch_sequence_length(self):
|
||||
# Tests that all encoded values have the correct size
|
||||
tokenizers = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
table = self.get_table(tokenizer, length=0)
|
||||
sequences = [
|
||||
"Testing batch encode plus",
|
||||
"Testing batch encode plus with different sequence lengths",
|
||||
"Testing batch encode plus with different sequence lengths correctly pads",
|
||||
]
|
||||
|
||||
encoded_sequences = [tokenizer.encode_plus(table, sequence) for sequence in sequences]
|
||||
encoded_sequences_batch = tokenizer.batch_encode_plus(table, sequences, padding=False)
|
||||
self.assertListEqual(
|
||||
encoded_sequences, self.convert_batch_encode_plus_format_to_encode_plus(encoded_sequences_batch)
|
||||
)
|
||||
|
||||
maximum_length = len(
|
||||
max([encoded_sequence["input_ids"] for encoded_sequence in encoded_sequences], key=len)
|
||||
)
|
||||
|
||||
# check correct behaviour if no pad_token_id exists and add it eventually
|
||||
self._check_no_pad_token_padding(tokenizer, sequences)
|
||||
|
||||
encoded_sequences_padded = [
|
||||
tokenizer.encode_plus(table, sequence, max_length=maximum_length, padding="max_length")
|
||||
for sequence in sequences
|
||||
]
|
||||
|
||||
encoded_sequences_batch_padded = tokenizer.batch_encode_plus(table, sequences, padding=True)
|
||||
self.assertListEqual(
|
||||
encoded_sequences_padded,
|
||||
self.convert_batch_encode_plus_format_to_encode_plus(encoded_sequences_batch_padded),
|
||||
)
|
||||
|
||||
# check 'longest' is unsensitive to a max length
|
||||
encoded_sequences_batch_padded_1 = tokenizer.batch_encode_plus(table, sequences, padding=True)
|
||||
encoded_sequences_batch_padded_2 = tokenizer.batch_encode_plus(
|
||||
table, sequences, max_length=maximum_length + 10, padding="longest"
|
||||
)
|
||||
for key in encoded_sequences_batch_padded_1.keys():
|
||||
self.assertListEqual(
|
||||
encoded_sequences_batch_padded_1[key],
|
||||
encoded_sequences_batch_padded_2[key],
|
||||
)
|
||||
|
||||
# check 'no_padding' is unsensitive to a max length
|
||||
encoded_sequences_batch_padded_1 = tokenizer.batch_encode_plus(table, sequences, padding=False)
|
||||
encoded_sequences_batch_padded_2 = tokenizer.batch_encode_plus(
|
||||
table, sequences, max_length=maximum_length + 10, padding=False
|
||||
)
|
||||
for key in encoded_sequences_batch_padded_1.keys():
|
||||
self.assertListEqual(
|
||||
encoded_sequences_batch_padded_1[key],
|
||||
encoded_sequences_batch_padded_2[key],
|
||||
)
|
||||
|
||||
def test_batch_encode_plus_overflowing_tokens(self):
|
||||
tokenizers = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
table = self.get_table(tokenizer, length=0)
|
||||
string_sequences = ["Testing the prepare_for_model method.", "Test"]
|
||||
|
||||
if tokenizer.pad_token is None:
|
||||
tokenizer.add_special_tokens({"pad_token": "[PAD]"})
|
||||
|
||||
tokenizer.batch_encode_plus(
|
||||
table, string_sequences, return_overflowing_tokens=True, truncation=True, padding=True, max_length=3
|
||||
)
|
||||
|
||||
def test_batch_encode_plus_padding(self):
|
||||
# Test that padded sequences are equivalent between batch_encode_plus and encode_plus
|
||||
|
||||
# Right padding tests
|
||||
tokenizers = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
table = self.get_table(tokenizer, length=0)
|
||||
sequences = [
|
||||
"Testing batch encode plus",
|
||||
"Testing batch encode plus with different sequence lengths",
|
||||
"Testing batch encode plus with different sequence lengths correctly pads",
|
||||
]
|
||||
|
||||
max_length = 100
|
||||
|
||||
# check correct behaviour if no pad_token_id exists and add it eventually
|
||||
self._check_no_pad_token_padding(tokenizer, sequences)
|
||||
|
||||
encoded_sequences = [
|
||||
tokenizer.encode_plus(table, sequence, max_length=max_length, padding="max_length")
|
||||
for sequence in sequences
|
||||
]
|
||||
encoded_sequences_batch = tokenizer.batch_encode_plus(
|
||||
table, sequences, max_length=max_length, padding="max_length"
|
||||
)
|
||||
self.assertListEqual(
|
||||
encoded_sequences, self.convert_batch_encode_plus_format_to_encode_plus(encoded_sequences_batch)
|
||||
)
|
||||
|
||||
# Left padding tests
|
||||
tokenizers = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
tokenizer.padding_side = "left"
|
||||
sequences = [
|
||||
"Testing batch encode plus",
|
||||
"Testing batch encode plus with different sequence lengths",
|
||||
"Testing batch encode plus with different sequence lengths correctly pads",
|
||||
]
|
||||
|
||||
max_length = 100
|
||||
|
||||
# check correct behaviour if no pad_token_id exists and add it eventually
|
||||
self._check_no_pad_token_padding(tokenizer, sequences)
|
||||
|
||||
encoded_sequences = [
|
||||
tokenizer.encode_plus(table, sequence, max_length=max_length, padding="max_length")
|
||||
for sequence in sequences
|
||||
]
|
||||
encoded_sequences_batch = tokenizer.batch_encode_plus(
|
||||
table, sequences, max_length=max_length, padding="max_length"
|
||||
)
|
||||
self.assertListEqual(
|
||||
encoded_sequences, self.convert_batch_encode_plus_format_to_encode_plus(encoded_sequences_batch)
|
||||
)
|
||||
|
||||
def test_padding_to_multiple_of(self):
|
||||
tokenizers = self.get_tokenizers()
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
table = self.get_table(tokenizer, length=0)
|
||||
if tokenizer.pad_token is None:
|
||||
self.skipTest("No padding token.")
|
||||
else:
|
||||
empty_tokens = tokenizer(table, padding=True, pad_to_multiple_of=8)
|
||||
normal_tokens = tokenizer(table, "This is a sample input", padding=True, pad_to_multiple_of=8)
|
||||
for key, value in empty_tokens.items():
|
||||
self.assertEqual(len(value) % 8, 0, "BatchEncoding.{} is not multiple of 8".format(key))
|
||||
for key, value in normal_tokens.items():
|
||||
self.assertEqual(len(value) % 8, 0, "BatchEncoding.{} is not multiple of 8".format(key))
|
||||
|
||||
normal_tokens = tokenizer(table, "This", pad_to_multiple_of=8)
|
||||
for key, value in normal_tokens.items():
|
||||
self.assertNotEqual(len(value) % 8, 0, "BatchEncoding.{} is not multiple of 8".format(key))
|
||||
|
||||
# Should also work with truncation
|
||||
normal_tokens = tokenizer(table, "This", padding=True, truncation=True, pad_to_multiple_of=8)
|
||||
for key, value in normal_tokens.items():
|
||||
self.assertEqual(len(value) % 8, 0, "BatchEncoding.{} is not multiple of 8".format(key))
|
||||
|
||||
# truncation to something which is not a multiple of pad_to_multiple_of raises an error
|
||||
self.assertRaises(
|
||||
ValueError,
|
||||
tokenizer.__call__,
|
||||
table,
|
||||
"This",
|
||||
padding=True,
|
||||
truncation=True,
|
||||
max_length=12,
|
||||
pad_to_multiple_of=8,
|
||||
)
|
||||
|
||||
@unittest.skip("TAPAS cannot handle `prepare_for_model` without passing by `encode_plus` or `batch_encode_plus`")
|
||||
def test_prepare_for_model(self):
|
||||
pass
|
||||
|
||||
def test_tokenizer_slow_store_full_signature(self):
|
||||
signature = inspect.signature(self.tokenizer_class.__init__)
|
||||
tokenizer = self.get_tokenizer()
|
||||
|
||||
for parameter_name, parameter in signature.parameters.items():
|
||||
if parameter.default != inspect.Parameter.empty:
|
||||
self.assertIn(parameter_name, tokenizer.init_kwargs)
|
||||
|
||||
def test_special_tokens_mask_input_pairs(self):
|
||||
tokenizers = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
sequence_0 = "Encode this."
|
||||
empty_table = self.get_table(tokenizer, length=0)
|
||||
table = self.get_table(tokenizer, length=10)
|
||||
encoded_sequence = tokenizer.encode(empty_table, sequence_0, add_special_tokens=False)
|
||||
encoded_sequence += tokenizer.encode(table, "", add_special_tokens=False)
|
||||
encoded_sequence_dict = tokenizer.encode_plus(
|
||||
table,
|
||||
sequence_0,
|
||||
add_special_tokens=True,
|
||||
return_special_tokens_mask=True,
|
||||
# add_prefix_space=False,
|
||||
)
|
||||
encoded_sequence_w_special = encoded_sequence_dict["input_ids"]
|
||||
special_tokens_mask = encoded_sequence_dict["special_tokens_mask"]
|
||||
self.assertEqual(len(special_tokens_mask), len(encoded_sequence_w_special))
|
||||
|
||||
filtered_sequence = [
|
||||
(x if not special_tokens_mask[i] else None) for i, x in enumerate(encoded_sequence_w_special)
|
||||
]
|
||||
filtered_sequence = [x for x in filtered_sequence if x is not None]
|
||||
self.assertEqual(encoded_sequence, filtered_sequence)
|
||||
|
||||
def test_special_tokens_mask(self):
|
||||
tokenizers = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
table = self.get_table(tokenizer, length=0)
|
||||
sequence_0 = "Encode this."
|
||||
# Testing single inputs
|
||||
encoded_sequence = tokenizer.encode(table, sequence_0, add_special_tokens=False)
|
||||
encoded_sequence_dict = tokenizer.encode_plus(
|
||||
table, sequence_0, add_special_tokens=True, return_special_tokens_mask=True
|
||||
)
|
||||
encoded_sequence_w_special = encoded_sequence_dict["input_ids"]
|
||||
special_tokens_mask = encoded_sequence_dict["special_tokens_mask"]
|
||||
self.assertEqual(len(special_tokens_mask), len(encoded_sequence_w_special))
|
||||
|
||||
filtered_sequence = [x for i, x in enumerate(encoded_sequence_w_special) if not special_tokens_mask[i]]
|
||||
self.assertEqual(encoded_sequence, filtered_sequence)
|
||||
|
||||
def test_save_and_load_tokenizer(self):
|
||||
# safety check on max_len default value so we are sure the test works
|
||||
tokenizers = self.get_tokenizers()
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
self.assertNotEqual(tokenizer.model_max_length, 42)
|
||||
|
||||
# Now let's start the test
|
||||
tokenizers = self.get_tokenizers()
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
# Isolate this from the other tests because we save additional tokens/etc
|
||||
table = self.get_table(tokenizer, length=0)
|
||||
tmpdirname = tempfile.mkdtemp()
|
||||
|
||||
sample_text = " He is very happy, UNwant\u00E9d,running"
|
||||
before_tokens = tokenizer.encode(table, sample_text, add_special_tokens=False)
|
||||
before_vocab = tokenizer.get_vocab()
|
||||
tokenizer.save_pretrained(tmpdirname)
|
||||
|
||||
after_tokenizer = tokenizer.__class__.from_pretrained(tmpdirname)
|
||||
after_tokens = after_tokenizer.encode(table, sample_text, add_special_tokens=False)
|
||||
after_vocab = after_tokenizer.get_vocab()
|
||||
self.assertListEqual(before_tokens, after_tokens)
|
||||
self.assertDictEqual(before_vocab, after_vocab)
|
||||
|
||||
shutil.rmtree(tmpdirname)
|
||||
|
||||
def test_right_and_left_padding(self):
|
||||
tokenizers = self.get_tokenizers(do_lower_case=False)
|
||||
for tokenizer in tokenizers:
|
||||
with self.subTest(f"{tokenizer.__class__.__name__}"):
|
||||
table = self.get_table(tokenizer, length=0)
|
||||
sequence = "Sequence"
|
||||
padding_size = 10
|
||||
|
||||
# check correct behaviour if no pad_token_id exists and add it eventually
|
||||
self._check_no_pad_token_padding(tokenizer, sequence)
|
||||
|
||||
padding_idx = tokenizer.pad_token_id
|
||||
|
||||
# RIGHT PADDING - Check that it correctly pads when a maximum length is specified along with the padding flag set to True
|
||||
tokenizer.padding_side = "right"
|
||||
encoded_sequence = tokenizer.encode(table, sequence)
|
||||
sequence_length = len(encoded_sequence)
|
||||
padded_sequence = tokenizer.encode(
|
||||
table, sequence, max_length=sequence_length + padding_size, padding="max_length"
|
||||
)
|
||||
padded_sequence_length = len(padded_sequence)
|
||||
assert sequence_length + padding_size == padded_sequence_length
|
||||
assert encoded_sequence + [padding_idx] * padding_size == padded_sequence
|
||||
|
||||
# LEFT PADDING - Check that it correctly pads when a maximum length is specified along with the padding flag set to True
|
||||
tokenizer.padding_side = "left"
|
||||
encoded_sequence = tokenizer.encode(table, sequence)
|
||||
sequence_length = len(encoded_sequence)
|
||||
padded_sequence = tokenizer.encode(
|
||||
table, sequence, max_length=sequence_length + padding_size, padding="max_length"
|
||||
)
|
||||
padded_sequence_length = len(padded_sequence)
|
||||
assert sequence_length + padding_size == padded_sequence_length
|
||||
assert [padding_idx] * padding_size + encoded_sequence == padded_sequence
|
||||
|
||||
# RIGHT & LEFT PADDING - Check that nothing is done for 'longest' and 'no_padding'
|
||||
encoded_sequence = tokenizer.encode(table, sequence)
|
||||
sequence_length = len(encoded_sequence)
|
||||
|
||||
tokenizer.padding_side = "right"
|
||||
padded_sequence_right = tokenizer.encode(table, sequence, padding=True)
|
||||
padded_sequence_right_length = len(padded_sequence_right)
|
||||
assert sequence_length == padded_sequence_right_length
|
||||
assert encoded_sequence == padded_sequence_right
|
||||
|
||||
tokenizer.padding_side = "left"
|
||||
padded_sequence_left = tokenizer.encode(table, sequence, padding="longest")
|
||||
padded_sequence_left_length = len(padded_sequence_left)
|
||||
assert sequence_length == padded_sequence_left_length
|
||||
assert encoded_sequence == padded_sequence_left
|
||||
|
||||
tokenizer.padding_side = "right"
|
||||
padded_sequence_right = tokenizer.encode(table, sequence)
|
||||
padded_sequence_right_length = len(padded_sequence_right)
|
||||
assert sequence_length == padded_sequence_right_length
|
||||
assert encoded_sequence == padded_sequence_right
|
||||
|
||||
tokenizer.padding_side = "left"
|
||||
padded_sequence_left = tokenizer.encode(table, sequence, padding=False)
|
||||
padded_sequence_left_length = len(padded_sequence_left)
|
||||
assert sequence_length == padded_sequence_left_length
|
||||
assert encoded_sequence == padded_sequence_left
|
||||
|
||||
@unittest.skip("TAPAS doesn't handle pre-tokenized inputs.")
|
||||
def test_pretokenized_inputs(self):
|
||||
pass
|
||||
|
||||
Reference in New Issue
Block a user