Question Answering on MS-MARCO (Acc, Rouge-L, PPL)
24.25AccuracyBaseline
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| BaselineModel=Qwen 2 7B2026.05 | 24.25 | 46.34 | 7.01 | |
| +ScramblerModel=Qwen 2 7B2026.05 | 24.23 | 46.34 | 7.05 | |
| +ScramblerModel=Llama 3.1 8B2026.05 | 24.18 | 48.16 | 2.05 | |
| +DistributedModel=Llama 3.1 8B2026.05 | 24.13 | 48.11 | 2.05 | |
| BaselineModel=Llama 3.1 8B2026.05 | 24.11 | 48.14 | 2.05 | |
| +DistributedModel=Qwen 2 7B2026.05 | 24.11 | 46.25 | 7.02 | |
| BaselineModel=Qwen 2.5 32B2026.05 | 22.25 | 46.75 | 5.5 | |
| +DistributedModel=Qwen 2.5 32B2026.05 | 22 | 46.56 | 5.51 | |
| +ScramblerModel=Qwen 2.5 32B2026.05 | 22 | 46.49 | 5.5 | |
| +DistributedModel=Mistral 7B2026.05 | 18.09 | 42.46 | 2.35 | |
| BaselineModel=Mistral 7B2026.05 | 18.03 | 42.44 | 2.35 | |
| +ScramblerModel=Mistral 7B2026.05 | 18.02 | 42.35 | 2.35 |