Skip to content

handle numpy inputs in whole word mask data collator #22032

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
sgugger merged 1 commit into from
Mar 10, 2023
Merged

handle numpy inputs in whole word mask data collator #22032

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 3 additions & 1 deletion src/transformers/data/data_collator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -883,6 +883,8 @@ def torch_call(self, examples: List[Union[List[int], Any, Dict[str, Any]]]) -> D
return {"input_ids": inputs, "labels": labels}

def tf_call(self, examples: List[Union[List[int], Any, Dict[str, Any]]]) -> Dict[str, Any]:
import tensorflow as tf
Copy link
Copy Markdown
Contributor

@gante gante Mar 9, 2023

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Note for other reviewers: this import pattern is the one we are using in other functions in this file (we might want to reconsider it)


if isinstance(examples[0], Mapping):
input_ids = [e["input_ids"] for e in examples]
else:
Expand All @@ -907,7 +909,7 @@ def tf_call(self, examples: List[Union[List[int], Any, Dict[str, Any]]]) -> Dict
ref_tokens[i] = "##" + ref_tokens[i]
mask_labels.append(self._whole_word_mask(ref_tokens))
batch_mask = _tf_collate_batch(mask_labels, self.tokenizer, pad_to_multiple_of=self.pad_to_multiple_of)
inputs, labels = self.tf_mask_tokens(batch_input, batch_mask)
inputs, labels = self.tf_mask_tokens(tf.cast(batch_input, tf.int64), batch_mask)
return {"input_ids": inputs, "labels": labels}

def numpy_call(self, examples: List[Union[List[int], Any, Dict[str, Any]]]) -> Dict[str, Any]:
Expand Down
27 changes: 21 additions & 6 deletions tests/trainer/test_data_collator.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -271,12 +271,17 @@ def test_data_collator_for_language_modeling(self):
self._test_no_pad_and_pad(no_pad_features, pad_features)

def test_data_collator_for_whole_word_mask(self):
features = [{"input_ids": list(range(10))}, {"input_ids": list(range(10))}]

tokenizer = BertTokenizer(self.vocab_file)
data_collator = DataCollatorForWholeWordMask(tokenizer, return_tensors="pt")

features = [{"input_ids": list(range(10))}, {"input_ids": list(range(10))}]
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 10)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 10)))

# Features can already be tensors
features = [{"input_ids": np.arange(10)}, {"input_ids": np.arange(10)}]
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, torch.Size((2, 10)))
self.assertEqual(batch["labels"].shape, torch.Size((2, 10)))

Expand Down Expand Up @@ -553,12 +558,17 @@ def test_data_collator_for_language_modeling(self):
self._test_no_pad_and_pad(no_pad_features, pad_features)

def test_data_collator_for_whole_word_mask(self):
features = [{"input_ids": list(range(10))}, {"input_ids": list(range(10))}]

tokenizer = BertTokenizer(self.vocab_file)
data_collator = DataCollatorForWholeWordMask(tokenizer, return_tensors="tf")

features = [{"input_ids": list(range(10))}, {"input_ids": list(range(10))}]
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape.as_list(), [2, 10])
self.assertEqual(batch["labels"].shape.as_list(), [2, 10])

# Features can already be tensors
features = [{"input_ids": np.arange(10)}, {"input_ids": np.arange(10)}]
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape.as_list(), [2, 10])
self.assertEqual(batch["labels"].shape.as_list(), [2, 10])

Expand Down Expand Up @@ -825,12 +835,17 @@ def test_data_collator_for_language_modeling(self):
self._test_no_pad_and_pad(no_pad_features, pad_features)

def test_data_collator_for_whole_word_mask(self):
features = [{"input_ids": list(range(10))}, {"input_ids": list(range(10))}]

tokenizer = BertTokenizer(self.vocab_file)
data_collator = DataCollatorForWholeWordMask(tokenizer, return_tensors="np")

features = [{"input_ids": list(range(10))}, {"input_ids": list(range(10))}]
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, (2, 10))
self.assertEqual(batch["labels"].shape, (2, 10))

# Features can already be tensors
features = [{"input_ids": np.arange(10)}, {"input_ids": np.arange(10)}]
batch = data_collator(features)
self.assertEqual(batch["input_ids"].shape, (2, 10))
self.assertEqual(batch["labels"].shape, (2, 10))

Expand Down