Mathematical Reasoning on GSM8K (Accuracy, Accuracy degradation)
0.3Accuracy Degradation (Δ)GTD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GTDFailure Scenario=Single-agent failure2025.10 | 0.3 | — | |
| GTDFailure Scenario=Noisy agent (50% err.)2025.10 | 1.4 | — | |
| MaASFailure Scenario=Single-agent failure2025.10 | 1.8 | — | |
| GTDFailure Scenario=Two-agent failure2025.10 | 2.1 | — | |
| G-Des.Failure Scenario=Single-agent failure2025.10 | 2.1 | — | |
| MaASFailure Scenario=Noisy agent (50% err.)2025.10 | 3.2 | — | |
| G-Des.Failure Scenario=Noisy agent (50% err.)2025.10 | 3.9 | — | |
| MaASFailure Scenario=Two-agent failure2025.10 | 5.6 | — | |
| G-Des.Failure Scenario=Two-agent failure2025.10 | 6.8 | — | |
| G-Des.Failure Scenario=No failure2025.10 | — | 92.09 | |
| GTDFailure Scenario=No failure2025.10 | — | 94.14 | |
| MaASFailure Scenario=No failure2025.10 | — | 92.3 |