List your company

Free-form answer generation on KLAR (OOD)

38.13AccuracyGRPO
6.586814.775922.96531.1541Jun 4, 2026

Evaluation results

MethodLinks
GRPOBackbone=Qwen-2.5-7B2026.0638.13
BaselineBackbone=Qwen-2.5-7B2026.0635.76
CPT + GRPOBackbone=Qwen-2.5-7B2026.0630.75
CPT + SFTBackbone=Qwen-2.5-7B2026.0630.21
SFTBackbone=Qwen-2.5-7B2026.0629.96
CPTBackbone=Qwen-2.5-7B2026.0623.75
CPT + GRPOBackbone=OLMo-2-1124-7B2026.0617.6
GRPOBackbone=OLMo-2-1124-7B2026.0616.7
BaselineBackbone=OLMo-2-1124-7B2026.0613.3
CPT + SFTBackbone=OLMo-2-1124-7B2026.068.6
CPTBackbone=OLMo-2-1124-7B2026.068.3
SFTBackbone=OLMo-2-1124-7B2026.067.8