Visual Question Answering on GQA (BD-Metric)
61.9BD MetricVanilla
Evaluation Results
| Method | Links | |
|---|---|---|
| VanillaBackbone=LLaVA-1.5-7B, Retained Tokens=576 (Upper Bound)2025.08 | 61.9 | |
| Prune2DriveBackbone=LLaVA-1.5-7B, Retained Tokens=1282025.08 | 59.2 | |
| Prune2DriveBackbone=LLaVA-1.5-7B, Retained Tokens=642025.08 | 57.6 | |
| SparseVLMBackbone=LLaVA-1.5-7B, Retained Tokens=1282025.08 | 57.3 | |
| FastVBackbone=LLaVA-1.5-7B, Retained Tokens=1282025.08 | 54 | |
| SparseVLMBackbone=LLaVA-1.5-7B, Retained Tokens=642025.08 | 52 | |
| FastVBackbone=LLaVA-1.5-7B, Retained Tokens=642025.08 | 46 | |
| VLM adaptorCodec=JPEG, Model=QwenVL2.5-3B2025.12 | 11.63 | |
| VLM adaptorCodec=ELIC, Model=QwenVL2.5-3B2025.12 | 3.88 | |
| VLM adaptorCodec=ILLM, Model=QwenVL2.5-3B2025.12 | 2.38 |