Pruning a pretrained sparse autoencoder can produce an interpretability tool for a WANDA-pruned LLM that is roughly comparable to retraining an SAE on the pruned model, though with notable caveats in the reported metrics.
Towards efficient post-training quantization of pre-trained language models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
On the transferability of Sparse Autoencoders for interpreting compressed models
Pruning a pretrained sparse autoencoder can produce an interpretability tool for a WANDA-pruned LLM that is roughly comparable to retraining an SAE on the pruned model, though with notable caveats in the reported metrics.