Modality Conflict Resolution on ALME
71Text AccuracyQwen2-Audio-7B-Instruct
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Qwen2-Audio-7B-InstructCondition=Abase => TCF2026.06 | 71 | 0.41 | |
| Qwen2-Audio-7B-InstructCondition=ACF => Tbase2026.06 | 70 | 0.39 | |
| Ultravox v0.6-llama-3.1-8bCondition=Tbase => ACF2026.06 | 66 | 0.32 | |
| Ultravox v0.6-llama-3.1-8bCondition=ACF => Tbase2026.06 | 62 | 0.25 | |
| Ultravox v0.6-llama-3.1-8bCondition=TCF => Abase2026.06 | 61 | 0.22 | |
| Ultravox v0.6-llama-3.1-8bCondition=Abase => TCF2026.06 | 56 | 0.12 | |
| Qwen2-Audio-7B-InstructCondition=Tbase => ACF2026.06 | 50 | 0 | |
| Qwen2-Audio-7B-InstructCondition=TCF => Abase2026.06 | 47 | -0.05 |