Skip to content

fixes the gradient checkpointing of whisper #22019

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 4 commits into from
Mar 8, 2023
Merged

fixes the gradient checkpointing of whisper #22019

Changes from 2 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
562f741
fixing
soma2000-lang Mar 8, 2023
d8cf88f
Update modeling_whisper.py
soma2000-lang Mar 8, 2023
2d6687f
Update modeling_whisper.py
soma2000-lang Mar 8, 2023
674c1d3
Update src/transformers/models/whisper/modeling_whisper.py
gante Mar 8, 2023
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
12 changes: 6 additions & 6 deletions src/transformers/models/whisper/modeling_whisper.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1029,6 +1029,12 @@ def forward(
hidden_states = inputs_embeds + positions
hidden_states = nn.functional.dropout(hidden_states, p=self.dropout, training=self.training)

if self.gradient_checkpointing and self.training:
if use_cache:
logger.warning(

@sgugger sgugger Mar 8, 2023

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

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

Suggested change
logger.warning(
logger.warning(

This should be warn_once here (or warning_once) can't rememver the syntax.

@gante gante Mar 8, 2023 •
edited
Loading

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

warning_once :) (@soma2000-lang your fork is probably a bit stale, we moved from logger.warning to logger.warning_once on this statement ~2 weeks ago)

@soma2000-lang soma2000-lang Mar 8, 2023

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

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

@gante done !

"`use_cache = True` is incompatible with gradient checkpointing. Setting `use_cache =" " False`..."
Comment thread
gante marked this conversation as resolved.
Outdated
)
use_cache = False
# decoder layers
all_hidden_states = () if output_hidden_states else None
all_self_attns = () if output_attentions else None
Expand All @@ -1053,12 +1059,6 @@ def forward(
past_key_value = past_key_values[idx] if past_key_values is not None else None

if self.gradient_checkpointing and self.training:
if use_cache:
logger.warning(
"`use_cache = True` is incompatible with gradient checkpointing. Setting `use_cache ="
" False`..."
)
use_cache = False

def create_custom_forward(module):
def custom_forward(*inputs):
Expand Down