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

Use concept-erasure implementation of LEACE and SAL #252

Merged
merged 8 commits into from
Jul 10, 2023
Merged

Use concept-erasure implementation of LEACE and SAL #252

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
dc2cc49
Use concept-erasure implementation of LEACE and SAL
norabelrose Jun 8, 2023
0a70094
fix parameter name in ccs
lauritowal Jun 11, 2023
280343c
Fix test failures
norabelrose Jun 14, 2023
703844c
Merge branch 'leace' of github.com:EleutherAI/elk into leace
norabelrose Jun 14, 2023
fac6247
Be picky about the concept-erasure version
norabelrose Jun 14, 2023
0f6f120
Merge remote-tracking branch 'origin/main' into leace
norabelrose Jul 6, 2023
0f8d0a1
Refactor to support concept-erasure v0.1
norabelrose Jul 6, 2023
3db2cc8
Fix test failure
norabelrose Jul 6, 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
Prev Previous commit
Fix test failure
  • Loading branch information
@norabelrose
norabelrose committed Jul 6, 2023
commit 3db2cc8419aa4d6d4ecab0826e2e8d8cbb631fdc
1 change: 0 additions & 1 deletion elk/evaluation/evaluate.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,7 +40,6 @@ def apply_to_layer(

reporter_path = experiment_dir / "reporters" / f"layer_{layer}.pt"
reporter = torch.load(reporter_path, map_location=device)
reporter.eval()
Copy link
Collaborator

Choose a reason for hiding this comment

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

Isn't the eval() here still neded?

Copy link
Member Author

Choose a reason for hiding this comment

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

No because CcsReporter doesn't actually have any submodules like nn.BatchNorm or nn.Dropout whose behavior changes due to eval()


row_bufs = defaultdict(list)
for ds_name, (val_h, val_gt, _) in val_output.items():
Expand Down
2 changes: 1 addition & 1 deletion elk/training/eigen_reporter.py
View file
Open in desktop
Copy link
Collaborator

Choose a reason for hiding this comment

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

nice :) no need for pseudolabels at inference time

Original file line number Diff line number Diff line change
Expand Up @@ -42,7 +42,7 @@ def __post_init__(self):


class EigenFitter:
"""A linear reporter whose weights are computed via eigendecomposition.
"""Fit a linear reporter with eigendecomposition.

Args:
cfg: The reporter configuration.
Expand Down