Question Answering on SQuAD (Acc, TNFT, Tokens, Delay)
75.5Accuracy (Acc)Base
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| BaseBackbone=Qwen3-4B2026.05 | 75.5 | 239.92 | 786.71 | 11.58 | |
| BaseModel=Qwen3-4B2026.05 | 75.5 | 239.92 | 786.71 | 11.58 | |
| VanillaBackbone=Qwen3-4B2026.05 | 75.1 | 242.92 | 713.53 | 9.68 | |
| VanillaModel=Qwen3-4B2026.05 | 75.1 | 242.92 | 713.53 | 9.68 | |
| OursBackbone=Qwen3-4B2026.05 | 74.74 | 0 | 248.8 | 5.48 | |
| Stream (Ours)Model=Qwen3-4B2026.05 | 74.74 | 0 | 248.8 | 5.48 | |
| BaseBackbone=Qwen3-1.7B2026.05 | 53.8 | 239.92 | 778.18 | 8.92 | |
| BaseModel=Qwen3-1.7B2026.05 | 53.8 | 239.92 | 778.18 | 8.92 | |
| OursBackbone=Qwen3-1.7B2026.05 | 53.5 | 0 | 277.45 | 4.62 | |
| Stream (Ours)Model=Qwen3-1.7B2026.05 | 53.5 | 0 | 277.45 | 4.62 | |
| VanillaBackbone=Qwen3-1.7B2026.05 | 51.7 | 242.92 | 710.66 | 7.79 | |
| VanillaModel=Qwen3-1.7B2026.05 | 51.7 | 242.92 | 710.66 | 7.79 |