List your company

Language Modeling on Safe Prompts (test)

5.52PPLSVGT
4.544411.129717.71524.3003May 12, 2026

Evaluation results

MethodLinks
SVGTBackbone=MISTRAL-7B2026.055.52
No GuidanceBackbone=MISTRAL-7B2026.055.6
System PromptBackbone=MISTRAL-7B2026.055.69
No GuidanceBackbone=LLAMA-3.2-3B2026.056.71
System PromptBackbone=LLAMA-3.2-3B2026.056.92
SVGTBackbone=LLAMA-3.2-3B2026.057.34
DPO (LoRA)Backbone=MISTRAL-7B2026.058.28
RE-ControlBackbone=MISTRAL-7B2026.058.77
DPO (LoRA)Backbone=LLAMA-3.2-3B2026.059.21
RE-ControlBackbone=LLAMA-3.2-3B2026.059.54
No GuidanceBackbone=QWEN2-1.5B2026.059.97
System PromptBackbone=QWEN2-1.5B2026.0510.08
SVGTBackbone=QWEN2-1.5B2026.0510.22
ITIBackbone=MISTRAL-7B2026.0510.31
ITIBackbone=LLAMA-3.2-3B2026.0511.01
RE-ControlBackbone=QWEN2-1.5B2026.0511.46
DPO (LoRA)Backbone=QWEN2-1.5B2026.0511.69
ITIBackbone=QWEN2-1.5B2026.0513.82
SVGTBackbone=GPT-22026.0523.89
No GuidanceBackbone=GPT-22026.0524.53
System PromptBackbone=GPT-22026.0525.01
RE-ControlBackbone=GPT-22026.0527.65
DPO (LoRA)Backbone=GPT-22026.0528.15
ITIBackbone=GPT-22026.0529.91