Multi-modal Evaluation on MME (C, P, Avg % Reporting)
657.86Cognition Score (C)ToFu
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| ToFuModel=Qwen2.5-VL-7B-Instruct, Vision token retention rate=55%2026.05 | 657.86 | 1,683.05 | 100.75 | |
| LightKVModel=Qwen2.5-VL-7B-Instruct, Vision token retention rate=55%2026.05 | 647.5 | 1,706.38 | 101.37 | |
| ToMeModel=Qwen2.5-VL-7B-Instruct, Vision token retention rate=55%2026.05 | 640.71 | 1,687.75 | 100.04 | |
| VanillaModel=Qwen2.5-VL-7B-Instruct, Vision token retention rate=55%2026.05 | 638.21 | 1,695.25 | 100 | |
| FastVModel=Qwen2.5-VL-7B-Instruct, Vision token retention rate=55%2026.05 | 625.35 | 1,687.78 | 98.77 | |
| PiToMeModel=Qwen2.5-VL-7B-Instruct, Vision token retention rate=55%2026.05 | 624.64 | 1,671.09 | 100.24 | |
| VanillaBackbone=EVE-7B-v1-HD, Vision token retention rate=55%2026.05 | 304.6 | 1,314.1 | — | |
| LightKVBackbone=EVE-7B-v1-HD, Vision token retention rate=55%2026.05 | 291.4 | 1,308.9 | — | |
| FastVBackbone=EVE-7B-v1-HD, Vision token retention rate=55%2026.05 | 290.3 | 1,238.6 | — | |
| LightKVBackbone=EVE-7B-v1, Vision token retention rate=55%2026.05 | 269.3 | 1,203.1 | — | |
| VanillaBackbone=EVE-7B-v1, Vision token retention rate=55%2026.05 | 269.2 | 1,230.8 | — | |
| FastVBackbone=EVE-7B-v1, Vision token retention rate=55%2026.05 | 259.3 | 1,144.5 | — |