Question Answering on MARCO Unfiltered (Acc, Rouge-L, PPL)
18.77Accuracy+Distributed
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| +DistributedModel=Llama 3.1 8B2026.05 | 18.77 | 34.04 | 2.22 | |
| BaselineModel=Llama 3.1 8B2026.05 | 18.66 | 34.04 | 2.22 | |
| +ScramblerModel=Llama 3.1 8B2026.05 | 18.66 | 34.03 | 2.22 | |
| BaselineModel=Qwen 2 7B2026.05 | 17.58 | 29.43 | 12.85 | |
| +ScramblerModel=Qwen 2 7B2026.05 | 17.52 | 29.39 | 12.95 | |
| +DistributedModel=Qwen 2 7B2026.05 | 17.48 | 29.42 | 12.84 | |
| BaselineModel=Qwen 2.5 32B2026.05 | 17.08 | 32.64 | 7.19 | |
| +ScramblerModel=Qwen 2.5 32B2026.05 | 16.92 | 32.69 | 7.2 | |
| +DistributedModel=Qwen 2.5 32B2026.05 | 16.85 | 32.68 | 7.19 | |
| +DistributedModel=Mistral 7B2026.05 | 14.34 | 27.34 | 2.99 | |
| +ScramblerModel=Mistral 7B2026.05 | 14.25 | 27.3 | 2.99 | |
| BaselineModel=Mistral 7B2026.05 | 14.24 | 27.31 | 2.99 |