2026-09-20 · America/Los_Angeles · 社区动态 · #20

Jev vs classical ML: results from 8 classification datasets

Jev vs classical ML: results from 8 classification datasets I tested Jev in zero-shot and few-shot settings against 11 classical ML models, including SVM, XGBoost and logistic regression. Scores in the screenshot are balanced accuracy. This mostly went as I expected. If I already have labelled data, I’d start by training a classical model. They were more accurate on Banking77 and the business tabular datasets, and running a trained model locally avoids the per-call API cost. Jev did well on IMDb, so it depends on the task. But I don’t see a reason to replace a working ML model based on these…

热度 44.5 / 100；排名与评分保留该期记录。