Visual Question Answering on VQA v2 (Score & Performance Retention)
78.9ScoreLLaVA-1.5-7B
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| LLaVA-1.5-7BToken=576, RL Status=Upper Bound, Backbone=LLaVA-1.5-7B2026.03 | 78.9 | 100 | |
| TPRLToken=192*, RL Status=Methods with RL, Backbone=LLaVA-1.5-7B2026.03 | 78.6 | 99.6 | |
| Upper BoundTokens=5762025.12 | 77.7 | 100 | |
| TPRLToken=144*, RL Status=Methods with RL, Backbone=LLaVA-1.5-7B2026.03 | 77.5 | 98.2 | |
| PruMerge+Token=144, RL Status=Methods without RL, Backbone=LLaVA-1.5-7B2026.03 | 76.8 | 97.3 | |
| TPRLToken=128*, RL Status=Methods with RL, Backbone=LLaVA-1.5-7B2026.03 | 76.8 | 97.3 | |
| VisprunerToken=192, RL Status=Methods without RL, Backbone=LLaVA-1.5-7B2026.03 | 76.5 | 97 | |
| VisprunerToken=128, RL Status=Methods without RL, Backbone=LLaVA-1.5-7B2026.03 | 75.8 | 96.1 | |
| SparseVLMToken=192, RL Status=Methods without RL, Backbone=LLaVA-1.5-7B2026.03 | 75.6 | 95.8 | |
| PruMerge+Token=128, RL Status=Methods without RL, Backbone=LLaVA-1.5-7B2026.03 | 74.7 | 94.7 | |
| SparseVLMToken=128, RL Status=Methods without RL, Backbone=LLaVA-1.5-7B2026.03 | 73.2 | 92.8 | |
| HTC-VLMTokens=1 (hybrid)2025.12 | 72.4 | 85.5 | |
| VoCo-LLaMATokens=12025.12 | 71.8 | 83.8 | |
| ToMeToken=192, RL Status=Methods without RL, Backbone=LLaVA-1.5-7B2026.03 | 67.9 | 86.1 | |
| FastVToken=192, RL Status=Methods without RL, Backbone=LLaVA-1.5-7B2026.03 | 67.2 | 85.2 | |
| Avg. PoolTokens=12025.12 | 65 | 79.6 | |
| Q-FormerTokens=12025.12 | 63.4 | 70.5 | |
| ToMeToken=128, RL Status=Methods without RL, Backbone=LLaVA-1.5-7B2026.03 | 63 | 79.8 | |
| FastVToken=128, RL Status=Methods without RL, Backbone=LLaVA-1.5-7B2026.03 | 61.8 | 78.3 | |
| Lower BoundTokens=12025.12 | 41.2 | 0 |