Multimodal Perception on BLINK
64.95AccuracyFull Cache
Evaluation Results
| Method | Links | |
|---|---|---|
| Full CacheArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 64.95 | |
| Full CacheArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 64.65 | |
| H2OArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 64.42 | |
| SnapKVArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 64.31 | |
| RetentiveKVArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 64.22 | |
| Full CacheArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 64.04 | |
| RetentiveKVArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 63.84 | |
| LOOK-MArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 63.83 | |
| SAINTArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 63.82 | |
| SnapKVArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 63.79 | |
| MedaArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 63.62 | |
| MedaArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 63.55 | |
| H2OArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 63.55 | |
| SAINTArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 63.48 | |
| LOOK-MArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 63.21 | |
| SnapKVArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 63.14 | |
| RetentiveKVArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 63.12 | |
| MedaArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 63.04 | |
| SAINTArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 62.64 | |
| H2OArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 62.62 | |
| LOOK-MArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 62.49 |