Computational Cost (GFLOPs/token) on Natural Language Generation for XSum
5.38GFLOPs/tokenLTE
Evaluation results
| Method | Links | |
|---|---|---|
| LTEBackbone=LLaMA-7B, Permitted Quality Drop=ROUGE-L <= 0.052024.02 | 5.38 | |
| Deja VuBackbone=LLaMA-7B, Permitted Quality Drop=ROUGE-L <= 0.052024.02 | 7.92 | |
| R-LLaMA+MoEBackbone=LLaMA-7B, Permitted Quality Drop=ROUGE-L <= 0.052024.02 | 8.27 | |
| MoEficationBackbone=LLaMA-7B, Permitted Quality Drop=ROUGE-L <= 0.052024.02 | 10.45 | |
| FullBackbone=LLaMA-7B, Permitted Quality Drop=ROUGE-L <= 0.052024.02 | 12.06 |