-
Notifications
You must be signed in to change notification settings - Fork 32
Pull requests: EleutherAI/elk
Author
Label
Projects
Milestones
Reviews
Assignee
Sort
Pull requests list
Make extract_hiddens usable programmatically
#52
by norabelrose
was merged Feb 10, 2023
Loading…
updated Feb 12, 2023
Move changes over from mother-of-all-refactors
#45
by norabelrose
was merged Feb 10, 2023
Loading…
updated Feb 12, 2023
Initial implementation of semi-supervised training
#58
by norabelrose
was merged Feb 14, 2023
Loading…
updated Feb 14, 2023
Move tests outside the module
#85
by norabelrose
was merged Feb 18, 2023
Loading…
updated Feb 18, 2023
Train reporters for different layers in parallel
#86
by norabelrose
was merged Feb 19, 2023
Loading…
updated Feb 19, 2023
Add few-shot prompt support; small PromptCollator improvements
#84
by norabelrose
was merged Feb 19, 2023
Loading…
updated Feb 19, 2023
Migrate CLI to
simple_parsing
, allowing for YAML based configs
#92
by norabelrose
was merged Feb 20, 2023
Loading…
updated Feb 20, 2023
Migrate to HF datasets for storing extracted hidden states
refactor
Code change for clarity/extensibility/etc.
Remove promptsource submodule; move essential bits into repo proper
#101
by norabelrose
was merged Mar 6, 2023
Loading…
updated Mar 6, 2023
Fit logistic regression models on the GPU
#103
by norabelrose
was merged Mar 7, 2023
Loading…
updated Mar 7, 2023
Switch back to imap_unordered in training while ensuring CSV is still sorted
enhancement
New feature or request
#114
by norabelrose
was merged Mar 8, 2023
Loading…
updated Mar 8, 2023
Factor out pseudo-classifier into Reporter method
#115
by norabelrose
was merged Mar 8, 2023
Loading…
updated Mar 8, 2023
PCA init MVP
enhancement
New feature or request
#117
by norabelrose
was merged Mar 8, 2023
Loading…
updated Mar 8, 2023
EigenReporter and VINC algorithm
#124
by norabelrose
was merged Mar 15, 2023
Loading…
updated Mar 15, 2023
Add a calibration error statistic
#126
by norabelrose
was merged Mar 16, 2023
Loading…
updated Mar 16, 2023
Replace held_out_split with select_train_val_splits
#133
by norabelrose
was merged Mar 17, 2023
Loading…
updated Mar 17, 2023
Go back to using 1000 datapoints as the default
#139
by norabelrose
was merged Mar 18, 2023
Loading…
updated Mar 18, 2023
Custom PyTorch implementation of eigsh
#137
by norabelrose
was merged Mar 18, 2023
Loading…
updated Mar 18, 2023
Support L2 regularization & cross validation for Classifier
#135
by norabelrose
was merged Mar 19, 2023
Loading…
updated Mar 19, 2023
Add data_dir to PromptDataset
#138
by norabelrose
was merged Mar 20, 2023
Loading…
updated Mar 20, 2023
Set --balance to True by default
#144
by norabelrose
was closed Apr 5, 2023
Loading…
updated Apr 5, 2023
Previous Next
ProTip!
no:milestone will show everything without a milestone.