List your company

General Capability on MTBench

9.14MTBench ScoreREPBEND
3.80485.18996.5757.9601Apr 2, 2025Apr 11, 2025Apr 21, 2025May 1, 2025May 10, 2025May 20, 2025May 30, 2025

Evaluation results

MethodLinks
REPBENDModel=Qwen2.5 14B Instruct2025.049.14
Original Weight ModelModel=Qwen2.5 14B Instruct2025.048.71
WHP (W/O safe set)Backbone=Llama3 8B Instruct2025.048.01
WHP (W/O safe set)Backbone=Mistral 7B Instruct-v0.22025.047.88
OriginalBackbone=Llama3 8B Instruct2025.047.84
NPOBackbone=Llama3 8B Instruct2025.047.79
CBBackbone=Llama3 8B Instruct2025.047.74
OriginalBackbone=Mistral 7B Instruct-v0.22025.047.72
CB*Backbone=Llama3 8B Instruct, publicly_available_safety_tuned=true2025.047.72
NPO (W/O safe set)Backbone=Llama3 8B Instruct2025.047.71
REPBENDBackbone=Llama3 8B Instruct2025.047.71
APL (RM)Model=Llama-3-8B2025.057.65
WHPBackbone=Llama3 8B Instruct2025.047.63
CAPOModel=Llama-3-8B2025.057.63
APLModel=Llama-3-8B2025.057.59
CBBackbone=Mistral 7B Instruct-v0.22025.047.58
DPOBackbone=Mistral 7B Instruct-v0.22025.047.57
CB*Backbone=Mistral 7B Instruct-v0.2, publicly_available_safety_tuned=true2025.047.57
APL (w/o Adv)Model=Llama-3-8B2025.057.56
DPOBackbone=Llama3 8B Instruct2025.047.54
DPOModel=Llama-3-8B2025.057.54
BaseModel=Llama-3-8B2025.057.53
Safety PromptingBackbone=Mistral 7B Instruct-v0.22025.047.5
REPBENDBackbone=Mistral 7B Instruct-v0.22025.047.5
REPBENDModel=Gemma2 2B Instruct2025.047.37
Original Weight ModelModel=Gemma2 2B Instruct2025.047.35
NPOBackbone=Mistral 7B Instruct-v0.22025.047.31
Safety PromptingBackbone=Llama3 8B Instruct2025.047.18
TABackbone=Llama3 8B Instruct2025.047
TABackbone=Mistral 7B Instruct-v0.22025.046.93
SFTBackbone=Llama3 8B Instruct2025.046.88
APL (RM)Model=Mistral-7B2025.056.88
DPOModel=Mistral-7B2025.056.85
BaseModel=Mistral-7B2025.056.78
CAPOModel=Mistral-7B2025.056.66
APLModel=Mistral-7B2025.056.59
APL (w/o Adv)Model=Mistral-7B2025.056.53
WHPBackbone=Mistral 7B Instruct-v0.22025.046.29
SFTBackbone=Mistral 7B Instruct-v0.22025.045.98
R2D2*Backbone=Mistral 7B Instruct-v0.2, publicly_available_safety_tuned=true2025.045.97
RMUBackbone=Mistral 7B Instruct-v0.22025.045.77
NPO (W/O safe set)Backbone=Mistral 7B Instruct-v0.22025.044.67
RMUBackbone=Llama3 8B Instruct2025.044.01