CLVQ-VAE maps lower-layer transformer activations to higher-layer ones through a discrete codebook, yielding concept vectors evaluated with probe ablation and human annotation.
decent imitation of Brando’s godfather
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Cross-Layer Discrete Concept Discovery for Interpreting Language Models
CLVQ-VAE maps lower-layer transformer activations to higher-layer ones through a discrete codebook, yielding concept vectors evaluated with probe ablation and human annotation.