hypothesis
active
hypothesis:with-sparse-kernels-causal-crosscoders-may-beat-per-layer-saes-on-flopsWith sparse kernels, causal crosscoders may beat per-layer SAEs on FLOPs
Decoder cost dominates.
Source paper
extracted_fromNeighborhood — ranked by edge-count
Findings (1)
finding
- Naive implementation results.
Related 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.
- Crosscoders are about 2x less FLOP-efficient than per-layer SAEs at equal eval loss at large computefinding0.838Cost of identifying cross-layer structure.
- Indicates redundant structure across layers.
- Circuit simplification theory.
- Motivating example with 13 vs 3 features.
- Decoder norm profile across 18 layers.
- Crosscoders give a dramatically better whole-model (MSE, L0) tradeoff than summed per-layer SAEsfinding0.792Consolidating shared structure gives concise decomposition.
- Downside of abstraction.
- Motivation for DFC.