finding
active
finding:cross-layer-transcoder-features-form-a-mix-of-local-global-and-intermediate-features-by-decoder-normCross-layer transcoder features form a mix of local, global and intermediate features by decoder norm
Preliminary decoder norm analysis.
Source paper
extracted_fromRelated by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- Decoder norm profile across 18 layers.
- Discussion.
- MLP replacement with sparse features underlying the original attribution graphs.
- Motivation for DFC.
- Footnote 2.
- Conceptual difference.
- Mask virtual weights during training targeting interference.
- Decomposition into shared and model-specific features.