Prefill Latency on MMMU
1,797.4Prefill Latency (ms)Pruning Only
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Pruning OnlyRetain=25%, K_v (Experts)=8, Batch Size=8, Hardware=NVIDIA A800 GPU, Backbone=InternVL3.52025.11 | 1,797.4 | 1.56 | |
| FastMMoERetain=75%, K_v (Experts)=4, Batch Size=8, Hardware=NVIDIA A800 GPU, Backbone=InternVL3.52025.11 | 1,800.3 | 1.55 | |
| Pruning OnlyRetain=50%, K_v (Experts)=8, Batch Size=8, Hardware=NVIDIA A800 GPU, Backbone=InternVL3.52025.11 | 1,817.4 | 1.54 | |
| Pruning OnlyRetain=75%, K_v (Experts)=8, Batch Size=8, Hardware=NVIDIA A800 GPU, Backbone=InternVL3.52025.11 | 1,827.2 | 1.53 | |
| BaselineRetain=100%, K_v (Experts)=8 (Full), Batch Size=8, Hardware=NVIDIA A800 GPU, Backbone=InternVL3.52025.11 | 2,796.5 | 1 |