Video Multimodal Understanding on MMBench-Video
1.5MMBench Coarse ScorePrumerge
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| PrumergeVLM=Qwen32026.05 | 1.5 | 1.53 | |
| BaselineVLM=Qwen3, KV Cache=all KV2026.05 | 1.49 | 1.56 | |
| KVCapsuleVLM=Qwen32026.05 | 1.49 | 1.58 | |
| FastVVLM=Qwen22026.05 | 1.3 | 1.24 | |
| SparseVLMsVLM=Qwen22026.05 | 1.24 | 1.22 | |
| BaselineVLM=Qwen2, KV Cache=all KV2026.05 | 1.23 | 1.23 | |
| KVCapsuleVLM=Qwen22026.05 | 1.23 | 1.23 | |
| PrumergeVLM=Qwen22026.05 | 1.23 | 1.22 | |
| SparseVLMsVLM=Qwen32026.05 | 1.2 | 1.24 | |
| BaselineVLM=Qwen2.5, KV Cache=all KV2026.05 | 1.18 | 1.22 | |
| KVCapsuleVLM=Qwen2.52026.05 | 0.9 | 1.01 | |
| PrumergeVLM=Qwen2.52026.05 | 0.9 | 0.96 | |
| SparseVLMsVLM=Qwen2.52026.05 | 0.69 | 0.73 | |
| FastVVLM=Qwen32026.05 | 0.56 | 0.59 | |
| FastVVLM=Qwen2.52026.05 | 0.17 | 0.08 |