List your company

Behavior selection on BiPO (test)

64Hallucination Pair AccuracyCOLD-FD
42.1647.8353.559.17Mar 6, 2026

Evaluation results

MethodLinks
COLD-FDBackbone=Llama-2-7b-chat-hf2026.03647049524950
DiffMeanBackbone=Llama-2-7b-hf2026.0361—49—50—
COLD-FDBackbone=Llama-2-7b-hf2026.03608154545853
Base(ICL)Backbone=Llama-2-7b-hf2026.03585851514545
ReFT(vector)Backbone=Llama-2-7b-hf2026.03585849495050
BaseBackbone=Llama-2-7b-hf2026.03575749495050
COLD-KernelBackbone=Llama-2-7b-hf2026.03575849495050
Base(ICL)Backbone=Llama-2-7b-chat-hf2026.03565650505050
ReFT(mlp)Backbone=Llama-2-7b-hf2026.03525649495050
DiffMeanBackbone=Llama-2-7b-chat-hf2026.0346—71—50—
BaseBackbone=Llama-2-7b-chat-hf2026.03434360604949
ReFT(mlp)Backbone=Llama-2-7b-chat-hf2026.03433954525250
ReFT(vector)Backbone=Llama-2-7b-chat-hf2026.03434357564847
COLD-KernelBackbone=Llama-2-7b-chat-hf2026.03434360604949