List your company
ResearchFormerly WizWand

Corpus Composition Inference on LLMScan Mid-Grained 1.0 (test)

65.98Overlap AccuracyLLMSurgeon
31.774440.654749.53558.4153May 28, 2026
Updated 5mo ago

Evaluation results

MethodLinks
LLMSurgeonTarget Model=Pythia-12B2026.0565.98
LLMSurgeonTarget Model=Pythia-2.8B2026.0563.2
LLMSurgeonTarget Model=GPT-Neo-2.7B2026.0561.86
GradNormTarget Model=GPT-Neo-2.7B2026.0558.78
DC-PDDTarget Model=GPT-Neo-2.7B2026.0556.07
GradNormTarget Model=Pythia-2.8B2026.0555.66
DC-PDDTarget Model=Pythia-2.8B2026.0555.43
RecallTarget Model=Pythia-2.8B2026.0555.23
ZlibTarget Model=GPT-Neo-2.7B2026.0555.16
Min-K%++Target Model=Pythia-2.8B2026.0553.97
ZlibTarget Model=Pythia-2.8B2026.0553.86
LossTarget Model=GPT-Neo-2.7B2026.0553.54
Min-K%Target Model=GPT-Neo-2.7B2026.0553.47
Min-K%++Target Model=GPT-Neo-2.7B2026.0553.47
LossTarget Model=Pythia-2.8B2026.0553.27
Min-K%Target Model=Pythia-2.8B2026.0553.24
NeighborTarget Model=Pythia-12B2026.0552.9
DUCITarget Model=GPT-Neo-2.7B2026.0552.86
RecallTarget Model=Pythia-12B2026.0552.63
DUCITarget Model=Pythia-2.8B2026.0552.62
DUCITarget Model=Pythia-12B2026.0552.62
DC-PDDTarget Model=Pythia-12B2026.0552.53
LossTarget Model=Pythia-12B2026.0552.36
RefTarget Model=GPT-Neo-2.7B2026.0552.33
RefTarget Model=Pythia-2.8B2026.0552.33
RefTarget Model=Pythia-12B2026.0552.33
Joint-LogitTarget Model=Pythia-12B2026.0552.26
Joint-LogitTarget Model=Pythia-2.8B2026.0552.23
NeighborTarget Model=Pythia-2.8B2026.0552.21
Joint-LogitTarget Model=GPT-Neo-2.7B2026.0552.19
NeighborTarget Model=GPT-Neo-2.7B2026.0551.85
Min-K%Target Model=Pythia-12B2026.0550.23
RecallTarget Model=GPT-Neo-2.7B2026.0549.04
ZlibTarget Model=Pythia-12B2026.0547.8
GradNormTarget Model=Pythia-12B2026.0535.98
Min-K%++Target Model=Pythia-12B2026.0533.09