[breaking|pipelines|tokenizers] Adding slow-fast tokenizers equivalence tests pipelines - Removing sentencepiece as a required dependency (#8073)
* Fixing roberta for slow-fast tests * WIP getting equivalence on pipelines * slow-to-fast equivalence - working on question-answering pipeline * optional FAISS tests * Pipeline Q&A * Move pipeline tests to their own test job again * update tokenizer to add sequence id methods * update to tokenizers 0.9.4 * set sentencepiecce as optional * clean up squad * clean up pipelines to use sequence_ids * style/quality * wording * Switch to use_fast = True by default * update tests for use_fast at True by default * fix rag tokenizer test * removing protobuf from required dependencies * fix NER test for use_fast = True by default * fixing example tests (Q&A examples use slow tokenizers for now) * protobuf in main deps extras["sentencepiece"] and example deps * fix protobug install test * try to fix seq2seq by switching to slow tokenizers for now * Update src/transformers/tokenization_utils_base.py Co-authored-by: Lysandre Debut <lysandre@huggingface.co> * Update src/transformers/tokenization_utils_base.py Co-authored-by: Lysandre Debut <lysandre@huggingface.co> Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
This commit is contained in:
Thomas Wolf
@@ -1,10 +1,10 @@
|
||||
from typing import List, Optional
|
||||
from unittest import mock
|
||||
|
||||
from transformers import is_tf_available, is_torch_available, pipeline
|
||||
|
||||
# from transformers.pipelines import DefaultArgumentHandler, Pipeline
|
||||
from transformers.pipelines import Pipeline
|
||||
from transformers.testing_utils import _run_slow_tests, is_pipeline_test, require_tf, require_torch, slow
|
||||
from transformers.tokenization_utils_base import to_py_obj
|
||||
|
||||
|
||||
VALID_INPUTS = ["A simple string", ["list of strings"]]
|
||||
@@ -13,9 +13,11 @@ VALID_INPUTS = ["A simple string", ["list of strings"]]
|
||||
@is_pipeline_test
|
||||
class CustomInputPipelineCommonMixin:
|
||||
pipeline_task = None
|
||||
pipeline_loading_kwargs = {}
|
||||
small_models = None # Models tested without the @slow decorator
|
||||
large_models = None # Models tested with the @slow decorator
|
||||
pipeline_loading_kwargs = {} # Additional kwargs to load the pipeline with
|
||||
pipeline_running_kwargs = {} # Additional kwargs to run the pipeline with
|
||||
small_models = [] # Models tested without the @slow decorator
|
||||
large_models = [] # Models tested with the @slow decorator
|
||||
valid_inputs = VALID_INPUTS # Some inputs which are valid to compare fast and slow tokenizers
|
||||
|
||||
def setUp(self) -> None:
|
||||
if not is_tf_available() and not is_torch_available():
|
||||
@@ -47,73 +49,11 @@ class CustomInputPipelineCommonMixin:
|
||||
@require_torch
|
||||
@slow
|
||||
def test_pt_defaults(self):
|
||||
pipeline(self.pipeline_task, framework="pt")
|
||||
|
||||
@require_tf
|
||||
@slow
|
||||
def test_tf_defaults(self):
|
||||
pipeline(self.pipeline_task, framework="tf")
|
||||
|
||||
@require_torch
|
||||
def test_torch_small(self):
|
||||
for model_name in self.small_models:
|
||||
nlp = pipeline(task=self.pipeline_task, model=model_name, tokenizer=model_name, framework="pt")
|
||||
self._test_pipeline(nlp)
|
||||
|
||||
@require_tf
|
||||
def test_tf_small(self):
|
||||
for model_name in self.small_models:
|
||||
nlp = pipeline(task=self.pipeline_task, model=model_name, tokenizer=model_name, framework="tf")
|
||||
self._test_pipeline(nlp)
|
||||
|
||||
@require_torch
|
||||
@slow
|
||||
def test_torch_large(self):
|
||||
for model_name in self.large_models:
|
||||
nlp = pipeline(task=self.pipeline_task, model=model_name, tokenizer=model_name, framework="pt")
|
||||
self._test_pipeline(nlp)
|
||||
|
||||
@require_tf
|
||||
@slow
|
||||
def test_tf_large(self):
|
||||
for model_name in self.large_models:
|
||||
nlp = pipeline(task=self.pipeline_task, model=model_name, tokenizer=model_name, framework="tf")
|
||||
self._test_pipeline(nlp)
|
||||
|
||||
def _test_pipeline(self, nlp: Pipeline):
|
||||
raise NotImplementedError
|
||||
|
||||
|
||||
@is_pipeline_test
|
||||
class MonoInputPipelineCommonMixin:
|
||||
pipeline_task = None
|
||||
pipeline_loading_kwargs = {} # Additional kwargs to load the pipeline with
|
||||
pipeline_running_kwargs = {} # Additional kwargs to run the pipeline with
|
||||
small_models = [] # Models tested without the @slow decorator
|
||||
large_models = [] # Models tested with the @slow decorator
|
||||
mandatory_keys = {} # Keys which should be in the output
|
||||
valid_inputs = VALID_INPUTS # inputs which are valid
|
||||
invalid_inputs = [None] # inputs which are not allowed
|
||||
expected_multi_result: Optional[List] = None
|
||||
expected_check_keys: Optional[List[str]] = None
|
||||
|
||||
def setUp(self) -> None:
|
||||
if not is_tf_available() and not is_torch_available():
|
||||
return # Currently no JAX pipelines
|
||||
|
||||
for model_name in self.small_models:
|
||||
pipeline(self.pipeline_task, model=model_name, tokenizer=model_name, **self.pipeline_loading_kwargs)
|
||||
for model_name in self.large_models:
|
||||
pipeline(self.pipeline_task, model=model_name, tokenizer=model_name, **self.pipeline_loading_kwargs)
|
||||
|
||||
@require_torch
|
||||
@slow
|
||||
def test_pt_defaults_loads(self):
|
||||
pipeline(self.pipeline_task, framework="pt", **self.pipeline_loading_kwargs)
|
||||
|
||||
@require_tf
|
||||
@slow
|
||||
def test_tf_defaults_loads(self):
|
||||
def test_tf_defaults(self):
|
||||
pipeline(self.pipeline_task, framework="tf", **self.pipeline_loading_kwargs)
|
||||
|
||||
@require_torch
|
||||
@@ -166,6 +106,95 @@ class MonoInputPipelineCommonMixin:
|
||||
)
|
||||
self._test_pipeline(nlp)
|
||||
|
||||
def _test_pipeline(self, nlp: Pipeline):
|
||||
raise NotImplementedError
|
||||
|
||||
@require_torch
|
||||
def test_compare_slow_fast_torch(self):
|
||||
for model_name in self.small_models:
|
||||
nlp_slow = pipeline(
|
||||
task=self.pipeline_task,
|
||||
model=model_name,
|
||||
tokenizer=model_name,
|
||||
framework="pt",
|
||||
use_fast=False,
|
||||
**self.pipeline_loading_kwargs,
|
||||
)
|
||||
nlp_fast = pipeline(
|
||||
task=self.pipeline_task,
|
||||
model=model_name,
|
||||
tokenizer=model_name,
|
||||
framework="pt",
|
||||
use_fast=True,
|
||||
**self.pipeline_loading_kwargs,
|
||||
)
|
||||
self._compare_slow_fast_pipelines(nlp_slow, nlp_fast, method="forward")
|
||||
|
||||
@require_tf
|
||||
def test_compare_slow_fast_tf(self):
|
||||
for model_name in self.small_models:
|
||||
nlp_slow = pipeline(
|
||||
task=self.pipeline_task,
|
||||
model=model_name,
|
||||
tokenizer=model_name,
|
||||
framework="tf",
|
||||
use_fast=False,
|
||||
**self.pipeline_loading_kwargs,
|
||||
)
|
||||
nlp_fast = pipeline(
|
||||
task=self.pipeline_task,
|
||||
model=model_name,
|
||||
tokenizer=model_name,
|
||||
framework="tf",
|
||||
use_fast=True,
|
||||
**self.pipeline_loading_kwargs,
|
||||
)
|
||||
self._compare_slow_fast_pipelines(nlp_slow, nlp_fast, method="call")
|
||||
|
||||
def _compare_slow_fast_pipelines(self, nlp_slow: Pipeline, nlp_fast: Pipeline, method: str):
|
||||
"""We check that the inputs to the models forward passes are identical for
|
||||
slow and fast tokenizers.
|
||||
"""
|
||||
with mock.patch.object(
|
||||
nlp_slow.model, method, wraps=getattr(nlp_slow.model, method)
|
||||
) as mock_slow, mock.patch.object(nlp_fast.model, method, wraps=getattr(nlp_fast.model, method)) as mock_fast:
|
||||
for inputs in self.valid_inputs:
|
||||
if isinstance(inputs, dict):
|
||||
inputs.update(self.pipeline_running_kwargs)
|
||||
_ = nlp_slow(**inputs)
|
||||
_ = nlp_fast(**inputs)
|
||||
else:
|
||||
_ = nlp_slow(inputs, **self.pipeline_running_kwargs)
|
||||
_ = nlp_fast(inputs, **self.pipeline_running_kwargs)
|
||||
|
||||
mock_slow.assert_called()
|
||||
mock_fast.assert_called()
|
||||
|
||||
self.assertEqual(len(mock_slow.call_args_list), len(mock_fast.call_args_list))
|
||||
for mock_slow_call_args, mock_fast_call_args in zip(
|
||||
mock_slow.call_args_list, mock_slow.call_args_list
|
||||
):
|
||||
slow_call_args, slow_call_kwargs = mock_slow_call_args
|
||||
fast_call_args, fast_call_kwargs = mock_fast_call_args
|
||||
|
||||
slow_call_args, slow_call_kwargs = to_py_obj(slow_call_args), to_py_obj(slow_call_kwargs)
|
||||
fast_call_args, fast_call_kwargs = to_py_obj(fast_call_args), to_py_obj(fast_call_kwargs)
|
||||
|
||||
self.assertEqual(slow_call_args, fast_call_args)
|
||||
self.assertDictEqual(slow_call_kwargs, fast_call_kwargs)
|
||||
|
||||
|
||||
@is_pipeline_test
|
||||
class MonoInputPipelineCommonMixin(CustomInputPipelineCommonMixin):
|
||||
"""A version of the CustomInputPipelineCommonMixin
|
||||
with a predefined `_test_pipeline` method.
|
||||
"""
|
||||
|
||||
mandatory_keys = {} # Keys which should be in the output
|
||||
invalid_inputs = [None] # inputs which are not allowed
|
||||
expected_multi_result: Optional[List] = None
|
||||
expected_check_keys: Optional[List[str]] = None
|
||||
|
||||
def _test_pipeline(self, nlp: Pipeline):
|
||||
self.assertIsNotNone(nlp)
|
||||
|
||||
@@ -199,76 +228,3 @@ class MonoInputPipelineCommonMixin:
|
||||
self.assertIn(key, result)
|
||||
|
||||
self.assertRaises(Exception, nlp, self.invalid_inputs)
|
||||
|
||||
|
||||
# @is_pipeline_test
|
||||
# class DefaultArgumentHandlerTestCase(unittest.TestCase):
|
||||
# def setUp(self) -> None:
|
||||
# self.handler = DefaultArgumentHandler()
|
||||
#
|
||||
# def test_kwargs_x(self):
|
||||
# mono_data = {"X": "This is a sample input"}
|
||||
# mono_args = self.handler(**mono_data)
|
||||
#
|
||||
# self.assertTrue(isinstance(mono_args, list))
|
||||
# self.assertEqual(len(mono_args), 1)
|
||||
#
|
||||
# multi_data = {"x": ["This is a sample input", "This is a second sample input"]}
|
||||
# multi_args = self.handler(**multi_data)
|
||||
#
|
||||
# self.assertTrue(isinstance(multi_args, list))
|
||||
# self.assertEqual(len(multi_args), 2)
|
||||
#
|
||||
# def test_kwargs_data(self):
|
||||
# mono_data = {"data": "This is a sample input"}
|
||||
# mono_args = self.handler(**mono_data)
|
||||
#
|
||||
# self.assertTrue(isinstance(mono_args, list))
|
||||
# self.assertEqual(len(mono_args), 1)
|
||||
#
|
||||
# multi_data = {"data": ["This is a sample input", "This is a second sample input"]}
|
||||
# multi_args = self.handler(**multi_data)
|
||||
#
|
||||
# self.assertTrue(isinstance(multi_args, list))
|
||||
# self.assertEqual(len(multi_args), 2)
|
||||
#
|
||||
# def test_multi_kwargs(self):
|
||||
# mono_data = {"data": "This is a sample input", "X": "This is a sample input 2"}
|
||||
# mono_args = self.handler(**mono_data)
|
||||
#
|
||||
# self.assertTrue(isinstance(mono_args, list))
|
||||
# self.assertEqual(len(mono_args), 2)
|
||||
#
|
||||
# multi_data = {
|
||||
# "data": ["This is a sample input", "This is a second sample input"],
|
||||
# "test": ["This is a sample input 2", "This is a second sample input 2"],
|
||||
# }
|
||||
# multi_args = self.handler(**multi_data)
|
||||
#
|
||||
# self.assertTrue(isinstance(multi_args, list))
|
||||
# self.assertEqual(len(multi_args), 4)
|
||||
#
|
||||
# def test_args(self):
|
||||
# mono_data = "This is a sample input"
|
||||
# mono_args = self.handler(mono_data)
|
||||
#
|
||||
# self.assertTrue(isinstance(mono_args, list))
|
||||
# self.assertEqual(len(mono_args), 1)
|
||||
#
|
||||
# mono_data = ["This is a sample input"]
|
||||
# mono_args = self.handler(mono_data)
|
||||
#
|
||||
# self.assertTrue(isinstance(mono_args, list))
|
||||
# self.assertEqual(len(mono_args), 1)
|
||||
#
|
||||
# multi_data = ["This is a sample input", "This is a second sample input"]
|
||||
# multi_args = self.handler(multi_data)
|
||||
#
|
||||
# self.assertTrue(isinstance(multi_args, list))
|
||||
# self.assertEqual(len(multi_args), 2)
|
||||
#
|
||||
# multi_data = ["This is a sample input", "This is a second sample input"]
|
||||
# multi_args = self.handler(*multi_data)
|
||||
#
|
||||
# self.assertTrue(isinstance(multi_args, list))
|
||||
# self.assertEqual(len(multi_args), 2)
|
||||
|
||||
Reference in New Issue
Block a user