Model gemini

Training model: gemini-2.0-flash
Inference model: gpt-4o-mini

Investigations

Performance

DatasetValidation accuracyTest accuracy
espionage0.550.55
potions0.550.65
southgermancredit0.52884615384615390.5398230088495575
timetravel_insurance0.650.35
titanic0.72549019607843130.6666666666666666
wisconsin0.7460317460317460.696969696969697