Skip to content

fix ONNX support for bloom #18456

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
thomasw21 merged 5 commits into from
Aug 4, 2022
+14 −9
Merged

fix ONNX support for bloom #18456

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
5f4d510
replace triu operation for ONNX support
NouamaneTazi Aug 3, 2022
0fbe776
remove in-place addition
NouamaneTazi Aug 3, 2022
106a775
fix ONNX config past shapes
NouamaneTazi Aug 3, 2022
cf04a16
nit: refactor head_dim
NouamaneTazi Aug 3, 2022
5ff1ff2
add comment for why we replaced `triu`
NouamaneTazi Aug 3, 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
15 changes: 10 additions & 5 deletions src/transformers/models/bloom/configuration_bloom.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -214,14 +214,19 @@ def generate_dummy_inputs(
batch, seqlen = common_inputs["input_ids"].shape
# Not using the same length for past_key_values
past_key_values_length = seqlen + 2
past_shape = (
batch,
head_dim = self._config.hidden_size // self.num_attention_heads
past_key_shape = (
batch * self.num_attention_heads,
head_dim,
past_key_values_length,
self.num_attention_heads,
self._config.hidden_size // self.num_attention_heads,
)
past_value_shape = (
batch * self.num_attention_heads,
past_key_values_length,
head_dim,
)
ordered_inputs["past_key_values"] = [
(torch.zeros(past_shape), torch.zeros(past_shape)) for _ in range(self.num_layers)
(torch.zeros(past_key_shape), torch.zeros(past_value_shape)) for _ in range(self.num_layers)
]

ordered_inputs["attention_mask"] = common_inputs["attention_mask"]
Expand Down
8 changes: 4 additions & 4 deletions src/transformers/models/bloom/modeling_bloom.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -61,8 +61,9 @@ def _make_causal_mask(
"""
batch_size, target_length = input_ids_shape
mask = torch.empty((target_length, target_length + past_key_values_length), dtype=torch.bool, device=device)
mask[:, past_key_values_length:] = True
mask[:, past_key_values_length:].triu_(diagonal=1)
# ONNX doesn't support `torch.Tensor.triu` properly, thus we use this workaround
seq_ids = torch.arange(target_length, device=device)
mask[:, past_key_values_length:] = seq_ids[:, None] < seq_ids[None, :]

if past_key_values_length > 0:
mask[:, :past_key_values_length] = False
Expand Down Expand Up @@ -698,8 +699,7 @@ def forward(
past_key_values_length = 0
if past_key_values[0] is not None:
past_key_values_length = past_key_values[0][0].shape[2]
seq_length_with_past += past_key_values_length

seq_length_with_past = seq_length_with_past + past_key_values_length
if attention_mask is None:
attention_mask = torch.ones((batch_size, seq_length_with_past), device=hidden_states.device)
else:
Expand Down