jettjaniak teren issues

jettjaniak / teren

Linking activation space features to model behavior

Apache License 2.0

0 stars 1 forks source link

issues

Newest

Newest Most commented Recently updated Oldest Least commented Least recently updated

create core experiment script

#29 jettjaniak opened 4 months ago
0
core_experiment script, WIP

#28 jettjaniak closed 4 months ago
0
Adding Lindsey note perturbations

#27 ChatrikMangat closed 4 months ago
0
jupyter formating in pre-commit

#26 jettjaniak closed 4 months ago
0
add OAI's SAEs to Lindsey's repro nb

#25 jettjaniak closed 4 months ago
0
extend lindsey's to pert. at multiple scales & compute KL

#24 jettjaniak closed 3 months ago
1
save SAEFeatureExamples++ to persistent storage

#23 jettjaniak opened 4 months ago
1
make Lindsey's repro work with OAI SAEs

#22 jettjaniak closed 4 months ago
0
pre-commit should check notebooks with black

#21 jettjaniak closed 4 months ago
0
Add the functionality to specify in each perturbation if we perturb from activation or towards another point

#20 GiglemaAI opened 4 months ago
1
Create a separate clean experiment which replicates Stepan's results e2e for at least naive random

#19 GiglemaAI opened 4 months ago
0
Implement necessary perturbations

#18 GiglemaAI closed 4 months ago
1
migrate new perturbations

#17 jettjaniak closed 4 months ago
2
typing, separate fns/methods

#16 jettjaniak closed 4 months ago
0
how to turn inactive features on?

#15 jettjaniak opened 4 months ago
0
think how to abstract at what seq. pos. we perturb or calculate loss

#14 jettjaniak opened 4 months ago
0
split `get_pert_loss` function s.t. we can access perturbed activations if we want to

#13 jettjaniak closed 4 months ago
0
add method to perturbations that returns pert_by_feature_id

#12 jettjaniak closed 4 months ago
0
make it harder to confuse feature_idx and feature_id

#11 jettjaniak closed 4 months ago
0
Compute resid mean across the whole dataset

#10 ChatrikMangat closed 4 months ago
0
figure out how to compute KL divergence effectively

#9 jettjaniak closed 4 months ago
1
figure out how to compute "standard error of the mean"

#8 jettjaniak closed 4 months ago
0
initial base experiment

#7 GiglemaAI closed 4 months ago
0
perturb. revamp & notebook

#6 jettjaniak closed 4 months ago
0
update nvidia drivers on VMs

#5 jettjaniak closed 4 months ago
0
auto shutdown VMs

#4 jettjaniak closed 4 months ago
1
figure out how to deal with private keys on VMs

#3 jettjaniak closed 4 months ago
0
zotero

#2 jettjaniak closed 4 months ago
0
make HF org

#1 jettjaniak closed 4 months ago
0