Backdoored Qwen2.5-3B models show attention deviations concentrated in layers 20-30, with single-token triggers localized and multi-token triggers diffuse.
Databricks dolly 15k
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
dataset 1
citation-polarity summary
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1roles
dataset 1polarities
use dataset 1representative citing papers
citing papers explorer
-
Mechanistic Exploration of Backdoored Large Language Model Attention Patterns
Backdoored Qwen2.5-3B models show attention deviations concentrated in layers 20-30, with single-token triggers localized and multi-token triggers diffuse.