Weakening Neurons: An Input-Output Functionality in Transformers with Outsize Influence
The learned input-output behavior of GLU-based neurons in large language models (LLMs) is analyzed and it is found that weakening neurons display surprising behavior: even though there are few, they activate often and have a large influence on model behavior.
Sebastian Gerstner, Hilal AlQuabeh, Kentaro Inui et al.
· 0 citations