List your company

Safety-aware fine-tuning on REDORCA selected (subset 1)

1.02Evaluation LossOnline dynamic
1.00841.08671.1651.2433Nov 26, 2025

Evaluation results

MethodLinks
Online dynamicsample ratio (R)=10%, G=1, backbone=LLAMA-8B2025.111.02
Onlinesample ratio (R)=10%, G=1, backbone=LLAMA-8B2025.111.2
Onlinesample ratio (R)=5%, G=1, backbone=LLAMA-8B2025.111.22
DSIRbackbone=LLAMA-8B2025.111.24
Random selectionbackbone=LLAMA-8B2025.111.25
Offline BDSbackbone=LLAMA-8B2025.111.25
Direct mixingmixing ratio (ρ)=1, backbone=LLAMA-8B2025.111.27
LESSbackbone=LLAMA-8B2025.111.28
Direct mixingmixing ratio (ρ)=0.5, backbone=LLAMA-8B2025.111.31