Summarization on SAMSum (ROUGE)
27.2ROUGE ScoreVanilla
Evaluation Results
| Method | Links | |
|---|---|---|
| VanillaBackbone=Vicuna-7B-v1.52026.05 | 27.2 | |
| LoRABackbone=Vicuna-7B-v1.52026.05 | 26.8 | |
| LEDBackbone=Vicuna-7B-v1.52026.05 | 26.7 | |
| VanillaBackbone=Llama2-7B-chat2026.05 | 26.7 | |
| Circuit BreakersBackbone=Vicuna-7B-v1.52026.05 | 26.2 | |
| EVA (text)Backbone=Vicuna-7B-v1.52026.05 | 26 | |
| Circuit BreakersBackbone=Llama2-7B-chat2026.05 | 26 | |
| LoRABackbone=Llama2-7B-chat2026.05 | 25.9 | |
| LoRABackbone=Mistral-7B-Instruct-v0.22026.05 | 25.8 | |
| LoRABackbone=Qwen2.5-7B-Instruct2026.05 | 25.8 | |
| LEDBackbone=Llama2-7B-chat2026.05 | 25.6 | |
| EVA (text)Backbone=Llama3.1-8B-Instruct2026.05 | 25.6 | |
| EVA (text)Backbone=Qwen2.5-7B-Instruct2026.05 | 25.6 | |
| VanillaBackbone=Mistral-7B-Instruct-v0.22026.05 | 25.5 | |
| Circuit BreakersBackbone=Mistral-7B-Instruct-v0.22026.05 | 25.4 | |
| VanillaBackbone=Llama3.1-8B-Instruct2026.05 | 25.4 | |
| LoRABackbone=Llama3.1-8B-Instruct2026.05 | 25.4 | |
| VanillaBackbone=Qwen2.5-7B-Instruct2026.05 | 25.4 | |
| LEDBackbone=Qwen2.5-7B-Instruct2026.05 | 25.4 | |
| Circuit BreakersBackbone=Qwen2.5-7B-Instruct2026.05 | 25.4 | |
| Circuit BreakersBackbone=Llama3.1-8B-Instruct2026.05 | 25.2 | |
| EVA (text)Backbone=Llama2-7B-chat2026.05 | 25.1 | |
| LEDBackbone=Llama3.1-8B-Instruct2026.05 | 25.1 | |
| SafeDecodingBackbone=Qwen2.5-7B-Instruct2026.05 | 23.9 | |
| SafeDecodingBackbone=Llama2-7B-chat2026.05 | 22.7 | |
| LEDBackbone=Mistral-7B-Instruct-v0.22026.05 | 22.1 | |
| EVA (text)Backbone=Mistral-7B-Instruct-v0.22026.05 | 20.5 | |
| SafeDecodingBackbone=Mistral-7B-Instruct-v0.22026.05 | 20.4 | |
| SafeDecodingBackbone=Vicuna-7B-v1.52026.05 | 17.4 | |
| SafeDecodingBackbone=Llama3.1-8B-Instruct2026.05 | 13.6 |