Financial Numerical Reasoning on ConvFinQA (dev)
85.67Execution AccuracyDSA-7B
Evaluation results
| Method | Links | |
|---|---|---|
| DSA-7BModel Scale=7B2026.05 | 85.67 | |
| FINDERBackbone=Mistral-7B Instruct v0.22026.05 | 81.95 | |
| APOLLO2026.05 | 78.76 | |
| DSA-3BModel Scale=3B2026.05 | 78.5 | |
| ENCORE2026.05 | 76 | |
| PoT-GPT-4Backbone=GPT-42026.05 | 74.77 | |
| ArgRecog2026.05 | 73.94 | |
| CBR-Ren2026.05 | 72.61 | |
| DSA-1.5BModel Scale=1.5B2026.05 | 71 | |
| FinQANet2026.05 | 68.32 | |
| Qwen2.5-7B-InsZero-shot=true, Model Scale=7B2026.05 | 59.2 | |
| Qwen2.5-3B-InsZero-shot=true, Model Scale=3B2026.05 | 44.8 | |
| Qwen2.5-1.5B-InsZero-shot=true, Model Scale=1.5B2026.05 | 30.15 |