Multi-turn Dialogue on MT-Bench (Speedup and tau)
4.55SpeedupBASTION
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| BASTIONModel=Qwen3-4B, Decoding Mode=Stochastic Decoding, Temperature=12026.05 | 4.55 | 6.57 | |
| BASTIONModel=Qwen3-4B, Decoding Mode=Greedy Decoding, Temperature=02026.05 | 4.53 | 6.58 | |
| BASTIONModel=Qwen3-8B, Decoding Mode=Stochastic Decoding, Temperature=12026.05 | 4.32 | 6.32 | |
| BASTIONModel=Qwen3-8B, Decoding Mode=Greedy Decoding, Temperature=02026.05 | 4.31 | 6.28 | |
| DFlashModel=Qwen3-8B, Decoding Mode=Greedy Decoding, Temperature=02026.05 | 2.81 | 4.46 | |
| DFlashModel=Qwen3-4B, Decoding Mode=Greedy Decoding, Temperature=02026.05 | 2.74 | 4.63 | |
| EAGLE-3Model=Qwen3-4B, Decoding Mode=Greedy Decoding, Temperature=02026.05 | 2.69 | 2.48 | |
| EAGLE-3Model=Qwen3-8B, Decoding Mode=Greedy Decoding, Temperature=02026.05 | 2.68 | 2.24 | |
| DFlashModel=Qwen3-4B, Decoding Mode=Stochastic Decoding, Temperature=12026.05 | 2.65 | 4.21 | |
| DFlashModel=Qwen3-8B, Decoding Mode=Stochastic Decoding, Temperature=12026.05 | 2.6 | 4 | |
| EAGLE-3Model=Qwen3-4B, Decoding Mode=Stochastic Decoding, Temperature=12026.05 | 2.59 | 2.37 | |
| EAGLE-3Model=Qwen3-8B, Decoding Mode=Stochastic Decoding, Temperature=12026.05 | 2.45 | 2.04 |