List your company
ResearchFormerly WizWand

Multi-modal Evaluation on MME (Cognition and Perception Scores)

1,731MME Perception ScoreVanilla
141.568554.209966.851,379.491Apr 14, 2026Apr 26, 2026May 8, 2026May 20, 2026Jun 1, 2026Jun 13, 2026Jun 25, 2026
Updated 4mo ago

Evaluation results

MethodLinks
VanillaToken Retention Ratio=100%2026.051,731704
DARTToken Retention Ratio=70%2026.051,724702
SPprunerToken Retention Ratio=70%2026.051,712709
SPprunerToken Retention Ratio=50%2026.051,711689
BAGEL (Ours)Baseline=BAGEL, Params=7B active2026.061,701.7715.9
BLIP3o-8B (Ours)Baseline=BLIP3o-8B, Params=8B2026.061,698.4660.3
DARTToken Retention Ratio=50%2026.051,696664
VARGPT-v1.1 (Ours)Baseline=VARGPT-v1.1, Params=7B+2B2026.061,695.7606.4
BAGELBaseline=BAGEL, Params=7B active2026.061,687701
BLIP3o-8BBaseline=BLIP3o-8B, Params=8B2026.061,682.6647.1
VARGPT-v1.1Baseline=VARGPT-v1.1, Params=7B+2B2026.061,678.3592.9
FastVToken Retention Ratio=70%2026.051,674718
SPprunerToken Retention Ratio=30%2026.051,662657
UniCornBaseline=BAGEL, Params=7B active2026.061,660677
DARTToken Retention Ratio=30%2026.051,654625
FastVToken Retention Ratio=50%2026.051,628698
Full ModelBase Model=LLaVA-NeXT, Sparsity Ratio=0%2026.041,588.3376.8
Janus-Pro-7BBaseline=Janus-Pro-7B, Params=7B2026.061,567.1—
TokenFlowBaseline=TokenFlow2026.061,551.1371.1
FastVToken Retention Ratio=30%2026.051,523580
TopoVLMBase Model=LLaVA-NeXT, Sparsity Ratio=50%, Calibration Sets=3 (Average)2026.041,506.7352.4
FullFTTrainable=7B2026.061,500350
DPVR-PCTrainable=202M, evaluation=3-seed average2026.061,499326
LoRATrainable=20M, r=162026.061,495358
LoRATrainable=80M, r=642026.061,491346
DPVR-KVTrainable=202M2026.061,480322
DPVR-LFTrainable=202M, evaluation=3-seed average2026.061,468326
TAMPBase Model=LLaVA-NeXT, Sparsity Ratio=50%, Calibration Sets=3 (Average)2026.041,461.3337.2
Vanilla LLaVA-1.5-7BTrainable=02026.061,457318
SEED-XBaseline=SEED-X2026.061,457—
FastVTrainable=0, mode=inference-only2026.061,450334
SparseGPTBase Model=LLaVA-NeXT, Sparsity Ratio=50%, Calibration Sets=3 (Average)2026.041,447.9313
ILLUMEBaseline=Vicuna-7B, Params=7B2026.061,445.3—
ECoFLaPBase Model=LLaVA-NeXT, Sparsity Ratio=50%, Calibration Sets=3 (Average)2026.041,439.9273
LLM-StreamlineBase Model=LLaVA-NeXT, Sparsity Ratio=50%, Calibration Sets=3 (Average)2026.041,428.5321.6
VILA-UBaseline=VILA-U2026.061,401.8—
LLM-PrunerBase Model=LLaVA-NeXT, Sparsity Ratio=50%, Calibration Sets=3 (Average)2026.041,372.1298.4
WandaBase Model=LLaVA-NeXT, Sparsity Ratio=50%, Calibration Sets=3 (Average)2026.041,348.3286.7
JanusBaseline=Janus2026.061,338—
OWLBase Model=LLaVA-NeXT, Sparsity Ratio=50%, Calibration Sets=3 (Average)2026.041,335.6271.9
Emu3Baseline=Emu32026.061,243.8266.1
Show-oBaseline=Show-o2026.061,232.9—
ChameleonBaseline=Chameleon2026.06202.7—