feat(backward_lens): add vocabulary readout of GPT-2 MLP gradient factors - #1723

Open

janmenjayap wants to merge 10 commits into

TransformerLensOrg:devfrom

janmenjayap:feat/backward-lens

Commits on Aug 25, 2026

feat(backward_lens): add gradient factorization contracts
janmenjayap
committed
0d32745
feat(backward_lens): capture GPT-2 MLP gradient factors
janmenjayap
committed
00035db
feat(backward_lens): project GPT-2 MLP gradient factors into vocabulary space
janmenjayap
committed
c49c1cd
docs(backward_lens): document GPT-2 gradient factor analysis
janmenjayap
committed
948cd92
docs(backward_lens): add executed GPT-2 gradient factor demo
janmenjayap
committed
a18f775
fix(backward_lens): stabilize demo notebook validation
janmenjayap
committed
6fc4b1f
fix(backward_lens): stabilize rank percentile output
janmenjayap
committed
38702a4
fix(backward_lens): support beartype test instrumentation
janmenjayap
committed
5344ff3

Commits on Aug 26, 2026