Model gemma
Training model: gemma3:27b
Inference model: gpt-4o-mini
Investigations
Investigation 10 (espionage)
Investigation 35 (potions)
Investigation 64 (southgermancredit)
Investigation 154 (wisconsin)
Performance
Dataset
Validation accuracy
Test accuracy
espionage
0.75
0.75
potions
0.9
0.6
southgermancredit
0.5096153846153846
0.5398230088495575
wisconsin
0.5873015873015873
0.42424242424242425