Difference between revisions of "2025:Music Reasoning QA Results"

From MIREX Wiki
(MMAR Results)
(OMniBench Results)
 
Line 30: Line 30:
 
|- style="font-weight:bold;"
 
|- style="font-weight:bold;"
 
! System
 
! System
! Methods Used
 
 
! style="text-align:right;" | ACC
 
! style="text-align:right;" | ACC
 
! style="text-align:right;" | music ACC
 
! style="text-align:right;" | music ACC
 
|-
 
|-
| Baseline 1
 
 
| SAR-LM (w/ Qwen2.5-Omni)
 
| SAR-LM (w/ Qwen2.5-Omni)
 
| style="text-align:right;" | 31.26%
 
| style="text-align:right;" | 31.26%
 
| style="text-align:right;" | 41.50%
 
| style="text-align:right;" | 41.50%
 
|-
 
|-
| Baseline 2
 
 
| Qwen2-Audio-7B-Instruct
 
| Qwen2-Audio-7B-Instruct
 
| style="text-align:right;" | 40.72%
 
| style="text-align:right;" | 40.72%

Latest revision as of 02:32, 16 September 2025

MMAR Results

System ACC music ACC mix-sound-music mix-music-speech mix-sound-music-speech
SAR-LM (w/ Qwen2.5-Omni) 40.00% 33.98% 27.27% 48.78% 37.50%
Qwen2.5-Omni 56.70% 40.78% 54.55% 67.07% 58.33%

OMniBench Results

System ACC music ACC
SAR-LM (w/ Qwen2.5-Omni) 31.26% 41.50%
Qwen2-Audio-7B-Instruct 40.72% 38.68%