List your company

RLHF Alignment on HH-RLHF (held-out)

78Win RateCHI-REBEL
71.86473.45775.0576.643Sep 23, 2025

Evaluation results

MethodLinks
CHI-REBELBackbone=Llama-8B2025.097819.7
KL-REBELBackbone=Llama-8B2025.0977.320.5
CHI-REBELBackbone=Llama-1B2025.0977.320.8
W-REBELBackbone=Llama-8B2025.0976.920.9
KL-REBELBackbone=Llama-1B2025.0976.421.6
W-REBELBackbone=Llama-1B2025.0976.121.9
KL-DPOBackbone=Llama-8B2025.0975.421.9
REBELBackbone=Llama-8B2025.097522.2
WDPOBackbone=Llama-8B2025.0974.222.9
KL-DPOBackbone=Llama-1B2025.0974.123.7
REBELBackbone=Llama-1B2025.0973.824
WDPOBackbone=Llama-1B2025.0973.524.1
DPOBackbone=Llama-8B2025.0972.923.7
DPOBackbone=Llama-1B2025.0972.125.6