Skip to content

Update glu check #1689

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
merged 6 commits into from
Nov 2, 2022
Merged

Update glu check #1689

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
e62292a
update glu check
mvpatel2000 Nov 2, 2022
4b25d56
separate out type check
mvpatel2000 Nov 2, 2022
5c2fdd5
fix test
mvpatel2000 Nov 2, 2022
47ae086
update docs
mvpatel2000 Nov 2, 2022
f7cbee0
type ignore
mvpatel2000 Nov 2, 2022
ff9e3ee
Merge branch 'dev' into mvpatel2000/glu-check
mvpatel2000 Nov 2, 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
9 changes: 5 additions & 4 deletions composer/algorithms/gated_linear_units/gated_linear_units.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -100,17 +100,18 @@ def apply_gated_linear_units(model: torch.nn.Module,

# get the activation functions used
act_fns = {module.intermediate_act_fn for module in intermediate_modules}
if len(act_fns) == 0:
num_act_fns = len({type(act_fn) for act_fn in act_fns})
if num_act_fns == 0:
raise ValueError('Tried to get the activation function from the model, but none were found. '
'Please specify `act_fn` manually to use Gated Linear Units.')
elif len(act_fns) > 1:
elif num_act_fns > 1:
raise ValueError('Tried to get the activation function from the model, but multiple different '
'functions are used. This is currently unsupported with Gated Linear Units. '
'Please either use one activation function in BertIntermediate modules or '
'specify `act_fn` to manually override activation functions.')

# since our set is of 1, let's extract the only activation function remaining.
(act_fn,) = act_fns #type: ignore will fail below if None
# since our set is of 1, let's extract the activation function
act_fn = next(iter(act_fns)) # type: ignore will fail below if None

if act_fn is None:
raise ValueError(
Expand Down