List your company
ResearchFormerly WizWand

Triplet Alignment on Toxic (Accuracy)

63.53AccuracyGSTransform
51.008454.259257.5160.7608May 30, 2025Jul 29, 2025Sep 27, 2025Nov 27, 2025Jan 26, 2026Mar 27, 2026May 27, 2026
Updated 5mo ago

Evaluation results

MethodLinks
GSTransformBackbone=Mxbai2025.0563.53
GSTransformBackbone=UAE2025.0563.01
GSTransformBackbone=BGE2025.0561.2
LoRABackbone=LLaMA-3.2-1B, Evaluation Protocol=Fine-tuning2026.0560.71
LoRABackbone=Mistral-7B, Evaluation Protocol=Fine-tuning2026.0560.09
PromptEmbedderBackbone=LLaMA-3.2-3B, Evaluation Protocol=Frozen Backbone Methods2026.0559.58
PromptEmbedderBackbone=Qwen2.5-7B, Evaluation Protocol=Frozen Backbone Methods2026.0558.95
LoRABackbone=Qwen2.5-7B, Evaluation Protocol=Fine-tuning2026.0558.12
EchoBackbone=Mistral-7B, Evaluation Protocol=Training-free Prompting2026.0556.41
LoRABackbone=LLaMA-3.2-3B, Evaluation Protocol=Fine-tuning2026.0556.27
Pretend-CoTBackbone=Qwen2.5-7B, Evaluation Protocol=Training-free Prompting2026.0556.07
Mxbai2025.0555.67
Prompt-TuningBackbone=Mistral-7B, Evaluation Protocol=Frozen Backbone Methods2026.0555.55
InBedderBackbone=Llama22025.0555.41
UAE2025.0555.2
PromptEOLBackbone=Qwen2.5-7B, Evaluation Protocol=Training-free Prompting2026.0554.94
BGE2025.0554.88
Prompt-TuningBackbone=LLaMA-3.2-1B, Evaluation Protocol=Frozen Backbone Methods2026.0554.63
PromptEmbedderBackbone=LLaMA-3.2-1B, Evaluation Protocol=Frozen Backbone Methods2026.0554.59
Prompt-TuningBackbone=LLaMA-3.2-3B, Evaluation Protocol=Frozen Backbone Methods2026.0554.35
Pretend-CoTBackbone=LLaMA-3.2-3B, Evaluation Protocol=Training-free Prompting2026.0554.27
Pretend-CoTBackbone=LLaMA-3.2-1B, Evaluation Protocol=Training-free Prompting2026.0554.15
Pretend-CoTBackbone=Mistral-7B, Evaluation Protocol=Training-free Prompting2026.0553.94
PromptEOLBackbone=LLaMA-3.2-3B, Evaluation Protocol=Training-free Prompting2026.0553.79
PromptEmbedderBackbone=Mistral-7B, Evaluation Protocol=Frozen Backbone Methods2026.0553.78
PromptEOLBackbone=Mistral-7B, Evaluation Protocol=Training-free Prompting2026.0553.77
InstructOR2025.0553.75
EchoBackbone=Qwen2.5-7B, Evaluation Protocol=Training-free Prompting2026.0553.37
Prompt-TuningBackbone=Qwen2.5-7B, Evaluation Protocol=Frozen Backbone Methods2026.0552.81
EchoBackbone=LLaMA-3.2-1B, Evaluation Protocol=Training-free Prompting2026.0552.68
PromptEOLBackbone=LLaMA-3.2-1B, Evaluation Protocol=Training-free Prompting2026.0552.29
EchoBackbone=LLaMA-3.2-3B, Evaluation Protocol=Training-free Prompting2026.0552.28
InBedderBackbone=Roberta2025.0551.49