Attribution Consistency and Downstream Performance on Spider-TAS
64F1 ScoreDeepSeek-V3.2
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| DeepSeek-V3.2Size=671B, Prompting Strategy=Prompting2026.04 | 64 | — | — | — | |
| ReTASSize=4B2026.04 | 63.5 | 61.4 | 21.9 | 10.2 | |
| GPT-5.1Size=Closed, Prompting Strategy=Prompting2026.04 | 61.5 | — | — | — | |
| Qwen3-30B-A3BSize=30B, Prompting Strategy=Reflection: Dual View2026.04 | 60.9 | 55.6 | 25 | 10.4 | |
| Qwen3-30B-A3BSize=30B, Prompting Strategy=Prompting2026.04 | 60.4 | — | — | — | |
| QwQ-32BSize=32B, Prompting Strategy=Reflection: Dual View2026.04 | 60.3 | 34.8 | 26.9 | 24.2 | |
| Qwen3-30B-A3BSize=30B, Prompting Strategy=Reflection: Single View2026.04 | 60.1 | 47.7 | — | — | |
| QwQ-32BSize=32B, Prompting Strategy=Prompting2026.04 | 58.2 | — | — | — | |
| QwQ-32BSize=32B, Prompting Strategy=Reflection: Single View2026.04 | 57.7 | 33.8 | — | — | |
| GLM-4.6Size=9B, Prompting Strategy=Reflection: Dual View2026.04 | 54.2 | 34.2 | 32.3 | 18.3 | |
| GLM-4.6Size=9B, Prompting Strategy=Reflection: Single View2026.04 | 50.7 | 35.1 | — | — | |
| GLM-4.6Size=9B, Prompting Strategy=Prompting2026.04 | 49.8 | — | — | — |