In-Context Learning Stability Analysis on UniICL-Bench (test)
2.1Random Replace Error (Und.)UniICL
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| UniICL2026.03 | 2.1 | 10.3 | 1.4 | 6.1 | 1.6 | 3.4 | 1.7 | 6.6 | |
| BAGEL2026.03 | 7.1 | 22 | 2.8 | 10.9 | 7.9 | 7.8 | 5.9 | 13.6 | |
| MLLM Avg.2026.03 | 7.3 | — | 1.8 | — | 6.3 | — | 5.1 | — | |
| Unified Avg.2026.03 | 17.2 | 15.7 | 8.5 | 5.7 | 11.3 | 10.4 | 12.4 | 10.6 |