agu18dec
/

final_olens

interpretability

activation-reading

Model card Files Files and versions

2.91 GB

Ctrl+K

Ctrl+K

1 contributor

History: 14 commits

agu18dec's picture

olens.py: optional decode-time template ban (read(..., ban_template=True)) and forced prefix; README: why and numbers

e1b1f4d verified about 14 hours ago

ar
final_olens lens B: lens (RL step 150), inverter, AR, standalone loader, curves, README 8 days ago
inverter
final_olens lens B: lens (RL step 150), inverter, AR, standalone loader, curves, README 8 days ago
lens_rl_step150
final_olens lens B: lens (RL step 150), inverter, AR, standalone loader, curves, README 8 days ago
lens_rl_step300
lens B RL steps 300 and 600 (+.015 at 600), curves over 600 RL steps, per-eval at step 600, hallucination tradeoff 7 days ago
lens_rl_step600
lens B RL steps 300 and 600 (+.015 at 600), curves over 600 RL steps, per-eval at step 600, hallucination tradeoff 7 days ago
p3_bullets_rl_step150
Add P3 bullet variant (RL step 150) 5 days ago
.gitattributes

1.63 kB
README: checkpoints (decay ends), AR fit, WSB-by-eval figure; curves 8 days ago
README.md

14.6 kB
olens.py: optional decode-time template ban (read(..., ban_template=True)) and forced prefix; README: why and numbers about 14 hours ago
olens.py

13.3 kB
olens.py: optional decode-time template ban (read(..., ban_template=True)) and forced prefix; README: why and numbers about 14 hours ago
olens_lensB.py

188 Bytes
Loader: olens.py with from_hub; README without A/B naming 5 days ago
training_curves.png

393 kB
xet

lens B RL steps 300 and 600 (+.015 at 600), curves over 600 RL steps, per-eval at step 600, hallucination tradeoff 7 days ago
wsb_by_eval.png

147 kB
xet

lens B RL steps 300 and 600 (+.015 at 600), curves over 600 RL steps, per-eval at step 600, hallucination tradeoff 7 days ago