anicka
/

nla-qwen3-4b-v2

natural-language-autoencoder

mechanistic-interpretability

activation-geometry

geometric-wellbeing

Model card Files Files and versions

nla-qwen3-4b-v2 / README.md

Commit History

Rewrite: credit Anthropic NLA as inspiration, explain our minimal replication

4190e69
verified

anicka commited on 20 days ago

NLA v2: axis-relevant training, FVE 0.94

c22cef7
verified

anicka commited on 20 days ago