Few-Shot learning on LongBench Samsum (test)
1.31Avg Accepted Draft Tokens per RoundLlama 3.1 8B (Post-norm)
Evaluation Results
| Method | Links | |
|---|---|---|
| Llama 3.1 8B (Post-norm)SWA mode=SWA + BoS, Window=10242026.05 | 1.31 | |
| Llama 3.1 8B (Post-norm)SWA mode=SWA + Sys Prompt, Window=10242026.05 | 1.31 | |
| Llama 3.1 8B (Pre-norm)SWA mode=SWA + BoS, Window=10242026.05 | 1.04 | |
| Llama 3.1 8B (Pre-norm)SWA mode=SWA + Sys Prompt, Window=10242026.05 | 0.39 |