Generalization Performance Analysis on ID and OOD Evaluation Suite Combined
80.2Average AccuracyICR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ICRBackbone=Qwen2.5-7B2025.09 | 80.2 | 0 | |
| Few-shot*Backbone=Qwen2.5-7B2025.09 | 77.5 | 1 | |
| M²IVBackbone=Qwen2.5-7B2025.09 | 76.2 | 3 | |
| LIVEBackbone=Qwen2.5-7B2025.09 | 75.7 | 4 | |
| I2CLBackbone=Qwen2.5-7B2025.09 | 75.6 | 3 | |
| Zero-shotBackbone=Qwen2.5-7B2025.09 | 71.3 | — | |
| Few-shot*Backbone=Llama2-7B2025.09 | 66.8 | 1 | |
| ICRBackbone=Llama2-7B2025.09 | 59.9 | 0 | |
| M²IVBackbone=Llama2-7B2025.09 | 56.9 | 0 | |
| LIVEBackbone=Llama2-7B2025.09 | 56.6 | 2 | |
| I2CLBackbone=Llama2-7B2025.09 | 55 | 2 | |
| Zero-shotBackbone=Llama2-7B2025.09 | 50.5 | — |