List your company
ResearchFormerly WizWand

Vision-Language Understanding on MMBench cn

60.6AccuracyVanilla
39.030444.630250.2355.8298Mar 11, 2025May 17, 2025Jul 23, 2025Sep 28, 2025Dec 4, 2025Feb 9, 2026Apr 18, 2026
Updated 4mo ago

Evaluation results

MethodLinks
VanillaBackbone=LLaVA-NeXT-7B, Token Budget=2880 Tokens, Venue=-2026.0260.6
VanillaBase Model=LLaVA-Next-7B, Token Reduction Rate=0.0%2025.0560.6
VanillaBackbone=LLaVA-NeXT-7B, Retained Tokens=2880, Compression Ratio=100%2026.0460.6
PIO-FVLMBackbone=LLaVA-NeXT-7B, Token Budget=320 Tokens, Venue=Ours2026.0259.5
MoBBase Model=LLaVA-Next-7B, Pruning budget K=320, η-prior=true2025.0558.9
DARTBackbone=LLaVA-NeXT-7B, Token Budget=320 Tokens, Venue=EMNLP’252026.0258.2
VanillaBase Model=LLaVA-1.5-7B, Token Reduction Rate=0.0%2025.0558.1
MoBBase Model=LLaVA-1.5-7B, Pruning budget K=192, η-prior=true2025.0557.8
HoloVBackbone=LLaVA-NeXT-7B, Token Budget=320 Tokens, Venue=NeurIPS’252026.0257.5
MoBBase Model=LLaVA-1.5-7B, Pruning budget K=128, η-prior=true2025.0557.5
EvoCompBackbone=LLaVA-NeXT-7B, Retained Tokens=160, Compression Ratio=94.4%, l=02026.0455.8
CDPrunerBackbone=LLaVA-NeXT-7B, Token Budget=320 Tokens, Venue=NeurIPS’252026.0255.7
VisionZipBackbone=LLaVA-NeXT-7B, Token Budget=320 Tokens, Venue=CVPR’252026.0255.6
SparseVLMBackbone=LLaVA-NeXT-7B, Token Budget=320 Tokens, Venue=ICML’252026.0254.5
MoBBase Model=LLaVA-1.5-7B, Pruning budget K=64, η-prior=true2025.0554.5
SparseVLMBackbone=LLaVA-NeXT-7B, Retained Tokens=160, Compression Ratio=94.4%2026.0453.8
MustDropBackbone=LLaVA-NeXT-7B, Retained Tokens=160, Compression Ratio=94.4%2026.0453.5
PyramidDropBackbone=LLaVA-NeXT-7B, Retained Tokens=160, Compression Ratio=94.4%2026.0453.4
FastVBackbone=LLaVA-NeXT-7B, Retained Tokens=160, Compression Ratio=94.4%2026.0453.1
DARTBackbone=LLaVA-NeXT-7B, Retained Tokens=160, Compression Ratio=94.4%2026.0452.6
VisionZipBackbone=LLaVA-NeXT-7B, Retained Tokens=160, Compression Ratio=94.4%2026.0452.1
FastVBackbone=LLaVA-NeXT-7B, Token Budget=320 Tokens, Venue=ECCV’242026.0251.9
GlobalCom^2Backbone=LLaVA-NeXT-7B, Retained Tokens=160, Compression Ratio=94.4%2026.0450.9
DARTBackbone=LLaVA-NeXT-7B, Retained Tokens=160, Compression Ratio=94.4%, l=02026.0447.8
GKL-KDCLIP Model=GKL-KD2025.0343.38
KL-KDCLIP Model=KL-KD [27]2025.0342.01
BaselineCLIP Model=Baseline2025.0339.86