Extra Long Video Retrieval on VideoNIAH
97.5AccuracyAdaSpark
Evaluation Results
| Method | Links | |
|---|---|---|
| AdaSparkModel Size=Mid Size, Backbone=Qwen-2.5-VL-7B, SFT=true2026.04 | 97.5 | |
| AdaSparkModel Size=Small Size, Backbone=Qwen-2.5-VL-3B, SFT=true2026.04 | 95.5 | |
| FrameFusionModel Size=Mid Size, Backbone=Qwen-2.5-VL-7B, SFT=true2026.04 | 92.5 | |
| VideoChat-Flash-2BModel Size=Small Size2026.04 | 92 | |
| FrameFusionModel Size=Small Size, Backbone=Qwen-2.5-VL-3B, SFT=true2026.04 | 90 | |
| Video-XL-7BModel Size=Mid Size2026.04 | 90 | |
| FastVModel Size=Mid Size, Backbone=Qwen-2.5-VL-7B, SFT=true2026.04 | 88.5 | |
| Qwen-2.5-VL-3B + SFTModel Size=Small Size, SFT=true2026.04 | 88 | |
| FastVModel Size=Small Size, Backbone=Qwen-2.5-VL-3B, SFT=true2026.04 | 88 | |
| Qwen-2.5-VL-3BModel Size=Small Size, SFT=false2026.04 | 86.5 | |
| ToMeModel Size=Small Size, Backbone=Qwen-2.5-VL-3B, SFT=true2026.04 | 86.5 | |
| Qwen-2.5-VL-7B + SFTModel Size=Mid Size, SFT=true2026.04 | 86 | |
| ToMeModel Size=Mid Size, Backbone=Qwen-2.5-VL-7B, SFT=true2026.04 | 85 | |
| Qwen-2.5-VL-7BModel Size=Mid Size, SFT=false2026.04 | 83 | |
| MoBAModel Size=Mid Size, Backbone=Qwen-2.5-VL-7B, SFT=true2026.04 | 70.5 | |
| MoBAModel Size=Small Size, Backbone=Qwen-2.5-VL-3B, SFT=true2026.04 | 65.5 | |
| LongVA-7BModel Size=Mid Size2026.04 | 58 |