Table Question Answering on Financial TableQA
85.51Execution AccuracyQwen-2.5-7B
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen-2.5-7BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=100%, Sketch Guidance=with sketch2026.04 | 85.51 | |
| LLaMA-3.1-8BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=100%, Sketch Guidance=with sketch2026.04 | 85.38 | |
| LLaMA-3.2-3BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=100%, Sketch Guidance=with sketch2026.04 | 82.31 | |
| Qwen-2.5-3BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=100%, Sketch Guidance=with sketch2026.04 | 81.92 | |
| LLaMA-3.1-8BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=10%, Sketch Guidance=with sketch2026.04 | 80.13 | |
| Qwen-2.5-7BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=10%, Sketch Guidance=with sketch2026.04 | 79.62 | |
| GPT-5Model Scale=>70B, Training Protocol=Off-the-shelf, Sketch Guidance=with sketch2026.04 | 77.82 | |
| Claude-4.5-SonnetModel Scale=>70B, Training Protocol=Off-the-shelf, Sketch Guidance=with sketch2026.04 | 76.79 | |
| Gemini-2.5-ProModel Scale=>70B, Training Protocol=Off-the-shelf, Sketch Guidance=with sketch2026.04 | 76.41 | |
| Qwen-2.5-7BModel Scale=<10B, Training Protocol=SFT, Data Percentage=100%, Sketch Guidance=with sketch2026.04 | 76.15 | |
| Qwen-2.5-7BModel Scale=<10B, Training Protocol=SFT, Data Percentage=100%, Sketch Guidance=without sketch2026.04 | 75.26 | |
| Claude-4.5-SonnetModel Scale=>70B, Training Protocol=Off-the-shelf, Sketch Guidance=without sketch2026.04 | 73.97 | |
| Qwen-2.5-3BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=10%, Sketch Guidance=with sketch2026.04 | 73.97 | |
| LLaMA-3.1-8BModel Scale=<10B, Training Protocol=SFT, Data Percentage=100%, Sketch Guidance=with sketch2026.04 | 73.97 | |
| Qwen-2.5-7BModel Scale=<10B, Training Protocol=SFT, Data Percentage=10%, Sketch Guidance=with sketch2026.04 | 72.56 | |
| LLaMA-3.1-8BModel Scale=<10B, Training Protocol=SFT, Data Percentage=100%, Sketch Guidance=without sketch2026.04 | 72.18 | |
| GPT-5Model Scale=>70B, Training Protocol=Off-the-shelf, Sketch Guidance=without sketch2026.04 | 71.28 | |
| LLaMA-3.2-3BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=10%, Sketch Guidance=with sketch2026.04 | 70.38 | |
| Gemini-2.5-ProModel Scale=>70B, Training Protocol=Off-the-shelf, Sketch Guidance=without sketch2026.04 | 70 | |
| LLaMA-3.2-3BModel Scale=<10B, Training Protocol=SFT, Data Percentage=100%, Sketch Guidance=with sketch2026.04 | 69.74 | |
| LLaMA-3.2-3BModel Scale=<10B, Training Protocol=SFT, Data Percentage=100%, Sketch Guidance=without sketch2026.04 | 69.74 | |
| Qwen-2.5-7BModel Scale=<10B, Training Protocol=SFT, Data Percentage=10%, Sketch Guidance=without sketch2026.04 | 69.1 | |
| Qwen-2.5-3BModel Scale=<10B, Training Protocol=SFT, Data Percentage=100%, Sketch Guidance=with sketch2026.04 | 68.33 | |
| LLaMA-3.1-8BModel Scale=<10B, Training Protocol=SFT, Data Percentage=10%, Sketch Guidance=with sketch2026.04 | 66.67 | |
| Qwen-2.5-3BModel Scale=<10B, Training Protocol=SFT, Data Percentage=100%, Sketch Guidance=without sketch2026.04 | 62.69 | |
| LLaMA-3.1-8BModel Scale=<10B, Training Protocol=SFT, Data Percentage=10%, Sketch Guidance=without sketch2026.04 | 62.44 | |
| Qwen-2.5-72BModel Scale=>70B, Training Protocol=Off-the-shelf, Sketch Guidance=with sketch2026.04 | 60.26 | |
| Qwen-2.5-3BModel Scale=<10B, Training Protocol=SFT, Data Percentage=10%, Sketch Guidance=without sketch2026.04 | 59.23 | |
| LLaMA-3.2-3BModel Scale=<10B, Training Protocol=SFT, Data Percentage=10%, Sketch Guidance=with sketch2026.04 | 57.56 | |
| Qwen-2.5-72BModel Scale=>70B, Training Protocol=Off-the-shelf, Sketch Guidance=without sketch2026.04 | 56.54 | |
| Qwen-2.5-3BModel Scale=<10B, Training Protocol=SFT, Data Percentage=10%, Sketch Guidance=with sketch2026.04 | 55.9 | |
| Qwen-2.5-7BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=100%, Sketch Guidance=without sketch2026.04 | 53.21 | |
| LLaMA-3.2-3BModel Scale=<10B, Training Protocol=SFT, Data Percentage=10%, Sketch Guidance=without sketch2026.04 | 52.95 | |
| Qwen-2.5-3BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=100%, Sketch Guidance=without sketch2026.04 | 45.64 | |
| Qwen-2.5-3BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=10%, Sketch Guidance=without sketch2026.04 | 43.85 | |
| LLaMA-3.1-8BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=10%, Sketch Guidance=without sketch2026.04 | 41.15 | |
| LLaMA-3.1-8BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=100%, Sketch Guidance=without sketch2026.04 | 41.03 | |
| Qwen-2.5-7BModel Scale=<10B, Training Protocol=Off-the-shelf, Sketch Guidance=with sketch2026.04 | 38.08 | |
| Qwen-2.5-7BModel Scale=<10B, Training Protocol=Off-the-shelf, Sketch Guidance=without sketch2026.04 | 35.9 | |
| Qwen-2.5-7BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=10%, Sketch Guidance=without sketch2026.04 | 34.62 | |
| LLaMA-3.2-3BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=100%, Sketch Guidance=without sketch2026.04 | 31.67 | |
| LLaMA-3.2-3BModel Scale=<10B, Training Protocol=TaNOS, Data Percentage=10%, Sketch Guidance=without sketch2026.04 | 31.54 | |
| LLaMA-3.1-8BModel Scale=<10B, Training Protocol=Off-the-shelf, Sketch Guidance=with sketch2026.04 | 23.59 | |
| LLaMA-3.1-8BModel Scale=<10B, Training Protocol=Off-the-shelf, Sketch Guidance=without sketch2026.04 | 21.15 | |
| Qwen-2.5-3BModel Scale=<10B, Training Protocol=Off-the-shelf, Sketch Guidance=without sketch2026.04 | 13.21 | |
| Qwen-2.5-3BModel Scale=<10B, Training Protocol=Off-the-shelf, Sketch Guidance=with sketch2026.04 | 12.44 | |
| LLaMA-3.2-3BModel Scale=<10B, Training Protocol=Off-the-shelf, Sketch Guidance=with sketch2026.04 | 10.64 | |
| LLaMA-3.2-3BModel Scale=<10B, Training Protocol=Off-the-shelf, Sketch Guidance=without sketch2026.04 | 7.56 |