Skip to content
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

Jump to bottom

add Bigbird ONNX config #16427

Merged
merged 3 commits into from
Apr 12, 2022
Merged

add Bigbird ONNX config #16427

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
3057beb
add Bigbird ONNX config
vumichien Mar 26, 2022
aeccd04
Merge branch 'main' of https://github.com/vumichien/transformers into…
vumichien Apr 12, 2022
bff19aa
remove multiple choice classification mark
vumichien Apr 12, 2022
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
1 change: 1 addition & 0 deletions docs/source/en/serialization.mdx
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,6 +49,7 @@ Ready-made configurations include the following architectures:
- BART
- BEiT
- BERT
- BigBird
- Blenderbot
- BlenderbotSmall
- CamemBERT
Expand Down
4 changes: 2 additions & 2 deletions src/transformers/models/big_bird/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -28,7 +28,7 @@


_import_structure = {
"configuration_big_bird": ["BIG_BIRD_PRETRAINED_CONFIG_ARCHIVE_MAP", "BigBirdConfig"],
"configuration_big_bird": ["BIG_BIRD_PRETRAINED_CONFIG_ARCHIVE_MAP", "BigBirdConfig", "BigBirdOnnxConfig"],
}

if is_sentencepiece_available():
Expand Down Expand Up @@ -66,7 +66,7 @@
]

if TYPE_CHECKING:
from .configuration_big_bird import BIG_BIRD_PRETRAINED_CONFIG_ARCHIVE_MAP, BigBirdConfig
from .configuration_big_bird import BIG_BIRD_PRETRAINED_CONFIG_ARCHIVE_MAP, BigBirdConfig, BigBirdOnnxConfig

if is_sentencepiece_available():
from .tokenization_big_bird import BigBirdTokenizer
Expand Down
14 changes: 14 additions & 0 deletions src/transformers/models/big_bird/configuration_big_bird.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,8 +13,11 @@
# See the License for the specific language governing permissions and
# limitations under the License.
""" BigBird model configuration"""
from collections import OrderedDict
from typing import Mapping

from ...configuration_utils import PretrainedConfig
from ...onnx import OnnxConfig
from ...utils import logging


Expand Down Expand Up @@ -160,3 +163,14 @@ def __init__(
self.block_size = block_size
self.num_random_blocks = num_random_blocks
self.classifier_dropout = classifier_dropout


class BigBirdOnnxConfig(OnnxConfig):
@property
def inputs(self) -> Mapping[str, Mapping[int, str]]:
return OrderedDict(
[
("input_ids", {0: "batch", 1: "sequence"}),
("attention_mask", {0: "batch", 1: "sequence"}),
]
)
4 changes: 2 additions & 2 deletions src/transformers/models/big_bird/modeling_big_bird.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2994,7 +2994,7 @@ def forward(
# setting lengths logits to `-inf`
logits_mask = self.prepare_question_mask(question_lengths, seqlen)
if token_type_ids is None:
token_type_ids = (~logits_mask).long()
token_type_ids = torch.ones(logits_mask.size(), dtype=int) - logits_mask
Copy link
Member

Choose a reason for hiding this comment

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

Are these changes needed to enable inference with ONNX Runtime? Just trying to understand the need to change the modeling code

Copy link
Contributor Author

Choose a reason for hiding this comment

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

@lewtun Yes, you are right. I have changed the type of logits_mask here from boolean to integer to enable inference with ONNX Runtime. The error looks like this if we use the original modeling code.

RuntimeError: 0INTERNAL ASSERT FAILED at "../torch/csrc/jit/ir/alias_analysis.cpp":611, please report a bug to PyTorch. We don't have an op for aten::fill_ but it isn't a special case.  Argument types: Tensor, bool, 

Candidates:
    aten::fill_.Scalar(Tensor(a!) self, Scalar value) -> (Tensor(a!))
    aten::fill_.Tensor(Tensor(a!) self, Tensor value) -> (Tensor(a!))

Copy link
Member

Choose a reason for hiding this comment

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

Thanks for the clarification! This looks fine to me :)

logits_mask = logits_mask
logits_mask[:, 0] = False
logits_mask.unsqueeze_(2)
Expand Down Expand Up @@ -3057,5 +3057,5 @@ def prepare_question_mask(q_lengths: torch.Tensor, maxlen: int):
# q_lengths -> (bz, 1)
mask = torch.arange(0, maxlen).to(q_lengths.device)
mask.unsqueeze_(0) # -> (1, maxlen)
mask = mask < q_lengths
mask = torch.where(mask < q_lengths, 1, 0)
Copy link
Contributor Author

Choose a reason for hiding this comment

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

Also in here, I have to change the mask type

return mask
10 changes: 10 additions & 0 deletions src/transformers/onnx/features.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -6,6 +6,7 @@
from ..models.bart import BartOnnxConfig
from ..models.beit import BeitOnnxConfig
from ..models.bert import BertOnnxConfig
from ..models.big_bird import BigBirdOnnxConfig
from ..models.blenderbot import BlenderbotOnnxConfig
from ..models.blenderbot_small import BlenderbotSmallOnnxConfig
from ..models.camembert import CamembertOnnxConfig
Expand Down Expand Up @@ -156,6 +157,15 @@ class FeaturesManager:
"question-answering",
onnx_config_cls=BertOnnxConfig,
),
"bigbird": supported_features_mapping(
"default",
"masked-lm",
"causal-lm",
"sequence-classification",
"token-classification",
"question-answering",
onnx_config_cls=BigBirdOnnxConfig,
),
"ibert": supported_features_mapping(
"default",
"masked-lm",
Expand Down
1 change: 1 addition & 0 deletions tests/onnx/test_onnx_v2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -172,6 +172,7 @@ def test_values_override(self):
PYTORCH_EXPORT_MODELS = {
("albert", "hf-internal-testing/tiny-albert"),
("bert", "bert-base-cased"),
("bigbird", "google/bigbird-roberta-base"),
("ibert", "kssteven/ibert-roberta-base"),
("camembert", "camembert-base"),
("distilbert", "distilbert-base-cased"),
Expand Down