AHA: Aligning Large Audio-Language Models for Reasoning Hallucinations via Counterfactual Hard Negatives

Benchmark Model Rank Results
audio-understanding-reasoning-on-mmau-testQwen-Audio-AHA#2Accuracy: 72.5
audio-understanding-reasoning-on-mmau-proQwen-Audio-AHA#3MMAU-Pro Overall Weighted Performance: 54.8
audio-understanding-reasoning-on-mmarQwen-Audio-AHA#6Accuracy: 59.7