Visual Question Answering on GQA (Score and Relative Score)
64.2ScoreVanilla
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VanillaRetained Tokens=2880, Backbone=LLaVA-NeXT-7B2026.01 | 64.2 | 100 | |
| VisionTrimRetained Tokens=640, Backbone=LLaVA-NeXT-7B2026.01 | 63.2 | 98.4 | |
| PDropRetained Tokens=640, Backbone=LLaVA-NeXT-7B2026.01 | 62.9 | 98 | |
| VisionTrimRetained Tokens=320, Backbone=LLaVA-NeXT-7B2026.01 | 61.7 | 96.1 | |
| VisionZipRetained Tokens=640, Backbone=LLaVA-NeXT-7B2026.01 | 61.3 | 95.5 | |
| SparseVLMRetained Tokens=640, Backbone=LLaVA-NeXT-7B2026.01 | 60.3 | 93.9 | |
| VisionZipRetained Tokens=320, Backbone=LLaVA-NeXT-7B2026.01 | 59.3 | 92.4 | |
| PDropRetained Tokens=320, Backbone=LLaVA-NeXT-7B2026.01 | 58.5 | 91.1 | |
| SparseVLMRetained Tokens=320, Backbone=LLaVA-NeXT-7B2026.01 | 57.7 | 89.9 | |
| VisionTrimRetained Tokens=160, Backbone=LLaVA-NeXT-7B2026.01 | 57.2 | 89.1 | |
| PDropRetained Tokens=160, Backbone=LLaVA-NeXT-7B2026.01 | 56.1 | 87.4 | |
| VisionZipRetained Tokens=160, Backbone=LLaVA-NeXT-7B2026.01 | 55.5 | 86.4 | |
| SparseVLMRetained Tokens=160, Backbone=LLaVA-NeXT-7B2026.01 | 51.2 | 79.8 |