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

Accelerate bug in specific versions #1007

Merged
merged 3 commits into from
Aug 28, 2023
Merged

Accelerate bug in specific versions #1007

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
0236131
Exclude accelerate versions with unwrap_model bug
BenjaminBossan Aug 14, 2023
941ea5e
Be more precise about the versions
BenjaminBossan Aug 14, 2023
8a27860
Use accelerate>=0.22.0
BenjaminBossan Aug 28, 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
2 changes: 1 addition & 1 deletion requirements-dev.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
accelerate>=0.11.0
accelerate>=0.11.0,!=0.20.*,!=0.21.0
Copy link
Member

Choose a reason for hiding this comment

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

At this point, what do you think of going with >=0.22 ?

Copy link
Collaborator Author

Choose a reason for hiding this comment

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

True, it's best to be on the latest version anyway. And since accelerate is not a strict dependency, users can still use skorch with different accelerate versions if they want.

fire
flaky
future>=0.17.1
Expand Down
22 changes: 22 additions & 0 deletions skorch/tests/test_hf.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -660,10 +660,32 @@ def test_mixed_precision_save_load_params(
accelerator = accelerator_cls(mixed_precision=mixed_precision)
net = net_cls(accelerator=accelerator)
net.initialize()

filename = tmp_path / 'accel-net-params.pth'
net.save_params(f_params=filename)
net.load_params(f_params=filename)

@pytest.mark.parametrize('mixed_precision', ['fp16', 'bf16', 'no'])
def test_mixed_precision_inference(
self, net_cls, accelerator_cls, data, mixed_precision, tmp_path
):
from accelerate.utils import is_bf16_available

if (mixed_precision != 'no') and not torch.cuda.is_available():
pytest.skip('skipping AMP test because device does not support it')
if (mixed_precision == 'bf16') and not is_bf16_available():
pytest.skip('skipping bf16 test because device does not support it')

X, y = data
accelerator = accelerator_cls(mixed_precision=mixed_precision)
net = net_cls(accelerator=accelerator)
net.fit(X, y)
net.predict(X)
net.predict_proba(X)

Xt = torch.from_numpy(X).to(net.device)
net.forward(Xt)

def test_force_cpu(self, net_cls, accelerator_cls, data):
accelerator = accelerator_cls(device_placement=False, cpu=True)
net = net_cls(accelerator=accelerator)
Expand Down