paper
referenced-only
2025
paper:2025-announcingAnnouncing open-source saes for llama 3.3 70b and llama 3.1 8b
Similar preprints — Semantic Scholar
Cited by (2)
- Understanding and Steering Llama 3 with Sparse Autoencoders - Goodfire
A sparse autoencoder trained on layer-19 residual-stream activations of Llama-3-8B-Instruct, using LMSYS-Chat-1M, yields features that can be labeled automatically, ranked by causal relevance, and use
- Endogenous Resistance to Activation Steering in Language Models