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

Add magic method to our TF models to convert datasets with column inference (#17160)

Browse Source 
* Add method to call to_tf_dataset() with column inference

* Add test for dataset creation

* Add a default arg for data collator

* Fix test

* Fix call with non-dev version of datasets

* Test correct column removal too

* make fixup

* More tests to make sure we remove unwanted columns

* Fix test to avoid predicting on unbuilt models

* Fix test to avoid predicting on unbuilt models

* Fix test to remove unwanted head mask columns from inputs

* Stop pushing your debug breakpoints to the main repo of the $2bn company you work for

* Skip the test in convnext because no grouped conv support

* Drop bools from the dataset dict

* Make style

* Skip the training test for models whose input dicts don't give us labels

* Skip transformerXL in the test because it doesn't return a simple loss

* Skip TFTapas because of some odd NaN losses

* make style

* make fixup

* Add docstring

* fixup

* Update src/transformers/modeling_tf_utils.py

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Update src/transformers/modeling_tf_utils.py

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Update src/transformers/modeling_tf_utils.py

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Update src/transformers/modeling_tf_utils.py

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Update src/transformers/modeling_tf_utils.py

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Remove breakpoint from tests

* Fix assert, add requires_backends

* Protect tokenizer import with if TYPE_CHECKING

* make fixup

* Add noqa, more fixup

* More rearranging for ~* aesthetics *~

* Adding defaults for shuffle and batch_size to match to_tf_dataset()

* Update src/transformers/modeling_tf_utils.py

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
This commit is contained in:
Matt
2022-06-06 15:53:49 +01:00
committed by GitHub
parent d28b7aa8cb
commit 19a8a3036d
5 changed files with 162 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

52
tests/test_modeling_tf_common.py
View File

@@ -25,6 +25,8 @@ import unittest.mock as mock
from importlib import import_module
from typing import List, Tuple
from datasets import Dataset
from huggingface_hub import delete_repo, login
from requests.exceptions import HTTPError
from transformers import is_tf_available, is_torch_available
@@ -1509,6 +1511,56 @@ class TFModelTesterMixin:
observed_main_input_name = list(model_signature.parameters.keys())[1]
self.assertEqual(model_class.main_input_name, observed_main_input_name)
def test_dataset_conversion(self):
config, inputs_dict = self.model_tester.prepare_config_and_inputs_for_common()
for model_class in self.all_model_classes:
model = model_class(config)
tf_inputs_dict = self._prepare_for_class(inputs_dict, model_class, return_labels=False)
tf_inputs_dict = {
key: val
for key, val in tf_inputs_dict.items()
if "head_mask" not in key and isinstance(val, tf.Tensor)
}
tf_inputs_dict["extra_unwanted_column"] = list(tf_inputs_dict.values())[0] # Use a random other tensor
input_dataset = Dataset.from_dict(tf_inputs_dict)
tf_dataset = model.prepare_tf_dataset(
input_dataset, batch_size=len(input_dataset), drop_remainder=False, shuffle=False
)
test_batch = next(iter(tf_dataset))
if isinstance(test_batch, tf.Tensor):
self.assertEqual(len(test_batch), len(input_dataset)) # Assert we didn't lose any data
else:
# Assert we discarded the unwanted extra column but kept everything else
self.assertEqual(len(test_batch), len(input_dataset.features) - 1)
self.assertNotIn("extra_unwanted_column", test_batch)
for tensor in test_batch.values():
self.assertTrue(isinstance(tensor, tf.Tensor))
self.assertEqual(len(tensor), len(input_dataset)) # Assert we didn't lose any data
model(test_batch, training=False)
if "labels" in inspect.signature(model_class.call).parameters.keys():
tf_inputs_dict = self._prepare_for_class(inputs_dict, model_class, return_labels=True)
if "labels" not in tf_inputs_dict:
return # This model isn't giving us labels after all, don't try training with it
tf_inputs_dict = {key: val for key, val in tf_inputs_dict.items() if "head_mask" not in key}
tf_inputs_dict["extra_unwanted_column"] = list(tf_inputs_dict.values())[0] # Use a random other tensor
input_dataset = Dataset.from_dict(tf_inputs_dict)
tf_dataset = model.prepare_tf_dataset(
input_dataset, batch_size=len(input_dataset), drop_remainder=False, shuffle=False
)
test_batch, test_batch_labels = next(iter(tf_dataset))
self.assertGreater(len(test_batch_labels), 0) # Assert the labels are present
feature_columns = 1 if isinstance(test_batch, tf.Tensor) else len(test_batch)
label_columns = 1 if isinstance(test_batch_labels, tf.Tensor) else len(test_batch_labels)
# Assert we discarded the unwanted extra column but kept everything else
self.assertEqual(feature_columns + label_columns, len(input_dataset.features) - 1)
if isinstance(test_batch, dict):
self.assertNotIn("extra_unwanted_column", test_batch)
if isinstance(test_batch_labels, dict):
self.assertNotIn("extra_unwanted_column", test_batch_labels)
model.compile(optimizer="sgd", run_eagerly=True)
model.train_on_batch(test_batch, test_batch_labels)
def _generate_random_bad_tokens(self, num_bad_tokens, model):
# special tokens cannot be bad tokens
special_tokens = []