Multimodal Conversational Question Answering on MMCoQA
38ROUGE-LFull Cache
Evaluation Results
| Method | Links | |
|---|---|---|
| Full CacheArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 38 | |
| Full CacheArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 35.5 | |
| RetentiveKVArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 33.1 | |
| SnapKVArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 32.5 | |
| Full CacheArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 32.5 | |
| SAINTArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 32 | |
| RetentiveKVArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 31.5 | |
| H2OArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 31.5 | |
| MedaArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 31.5 | |
| LOOK-MArchitecture=Qwen3-VL, Parameter Scale=8B2026.04 | 31.5 | |
| SnapKVArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 30.5 | |
| LOOK-MArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 29.5 | |
| SnapKVArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 29.5 | |
| RetentiveKVArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 29.5 | |
| MedaArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 28.5 | |
| SAINTArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 28.5 | |
| MedaArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 28.5 | |
| H2OArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 28 | |
| SAINTArchitecture=LLaVA-v1.5, Parameter Scale=7B2026.04 | 28 | |
| H2OArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 28 | |
| LOOK-MArchitecture=Qwen3-VL, Parameter Scale=4B2026.04 | 28 |