Language Modeling on Safe Prompts (test)
5.52PPLSVGT
Evaluation Results
| Method | Links | |
|---|---|---|
| SVGTBackbone=MISTRAL-7B2026.05 | 5.52 | |
| No GuidanceBackbone=MISTRAL-7B2026.05 | 5.6 | |
| System PromptBackbone=MISTRAL-7B2026.05 | 5.69 | |
| No GuidanceBackbone=LLAMA-3.2-3B2026.05 | 6.71 | |
| System PromptBackbone=LLAMA-3.2-3B2026.05 | 6.92 | |
| SVGTBackbone=LLAMA-3.2-3B2026.05 | 7.34 | |
| DPO (LoRA)Backbone=MISTRAL-7B2026.05 | 8.28 | |
| RE-ControlBackbone=MISTRAL-7B2026.05 | 8.77 | |
| DPO (LoRA)Backbone=LLAMA-3.2-3B2026.05 | 9.21 | |
| RE-ControlBackbone=LLAMA-3.2-3B2026.05 | 9.54 | |
| No GuidanceBackbone=QWEN2-1.5B2026.05 | 9.97 | |
| System PromptBackbone=QWEN2-1.5B2026.05 | 10.08 | |
| SVGTBackbone=QWEN2-1.5B2026.05 | 10.22 | |
| ITIBackbone=MISTRAL-7B2026.05 | 10.31 | |
| ITIBackbone=LLAMA-3.2-3B2026.05 | 11.01 | |
| RE-ControlBackbone=QWEN2-1.5B2026.05 | 11.46 | |
| DPO (LoRA)Backbone=QWEN2-1.5B2026.05 | 11.69 | |
| ITIBackbone=QWEN2-1.5B2026.05 | 13.82 | |
| SVGTBackbone=GPT-22026.05 | 23.89 | |
| No GuidanceBackbone=GPT-22026.05 | 24.53 | |
| System PromptBackbone=GPT-22026.05 | 25.01 | |
| RE-ControlBackbone=GPT-22026.05 | 27.65 | |
| DPO (LoRA)Backbone=GPT-22026.05 | 28.15 | |
| ITIBackbone=GPT-22026.05 | 29.91 |