List your company

Multimodal Video Comprehension on Video-MME

0Average SparsityFullAttention
-2.34413.47829.345.122Feb 5, 2026

Evaluation results

MethodLinks
FullAttentionFPS=1fps, Model=Qwen2-VL-7B-Instruct, Subtitles=false2026.0200.7290.6360.5520.639
FullAttentionFPS=0.5fps, Model=Qwen2-VL-7B-Instruct, Subtitles=true2026.0200.7210.6880.6240.678
FlexPrefillFPS=1fps, Model=Qwen2-VL-7B-Instruct, Sparsity Threshold=0.99, Subtitles=false2026.0217.80.7240.6360.5580.639
FlexPrefillFPS=0.5fps, Model=Qwen2-VL-7B-Instruct, Sparsity Threshold=0.99, Subtitles=true2026.0228.80.7170.6840.6190.673
RRAttentionFPS=1fps, Model=Qwen2-VL-7B-Instruct, Sparsity Threshold=0.95, Subtitles=false2026.0234.70.7260.640.5620.643
RRAttentionFPS=0.5fps, Model=Qwen2-VL-7B-Instruct, Sparsity Threshold=0.95, Subtitles=true2026.0237.40.7220.6870.6230.677
XAttentionFPS=1fps, Model=Qwen2-VL-7B-Instruct, Sparsity Threshold=0.95, Subtitles=false2026.0237.50.7270.6360.5610.641
XAttentionFPS=0.5fps, Model=Qwen2-VL-7B-Instruct, Sparsity Threshold=0.95, Subtitles=true2026.0239.90.7230.6890.6230.679
FlexPrefillFPS=1fps, Model=Qwen2-VL-7B-Instruct, Sparsity Threshold=0.95, Subtitles=false2026.0246.70.7280.620.540.629
RRAttentionFPS=1fps, Model=Qwen2-VL-7B-Instruct, Sparsity Threshold=0.90, Subtitles=false2026.02470.7220.6360.5620.64
XAttentionFPS=1fps, Model=Qwen2-VL-7B-Instruct, Sparsity Threshold=0.90, Subtitles=false2026.0249.30.7180.6280.5570.634
RRAttentionFPS=0.5fps, Model=Qwen2-VL-7B-Instruct, Sparsity Threshold=0.90, Subtitles=true2026.0249.30.7170.6870.6220.675
XAttentionFPS=0.5fps, Model=Qwen2-VL-7B-Instruct, Sparsity Threshold=0.90, Subtitles=true2026.0251.50.720.680.6110.67
FlexPrefillFPS=0.5fps, Model=Qwen2-VL-7B-Instruct, Sparsity Threshold=0.95, Subtitles=true2026.0258.60.7080.670.5940.657