List your company
ResearchFormerly WizWand

Multimodal Evaluation on LLaVA-Bench-Wild (LLaVA-W)

97.3Overall ScoreMoE-LLaVA
35.73251.71667.783.684Dec 12, 2023May 16, 2024Oct 19, 2024Mar 24, 2025Aug 27, 2025Jan 30, 2026Jul 5, 2026
Updated 4mo ago

Evaluation results

MethodLinks
MoE-LLaVALLM=Phi2-2.7B, Size=3.9B, Res.=3842024.0297.3
MoE-LLaVALLM=Phi2-2.7B, Size=3.9B, Res.=3362024.0294.1
VILA-13B + ShareGPT4VLLM=Llama-2-13B, Resolution=336, Pre-training samples=50M, Instruction tuning samples=1M2023.1278.4
TinyLLaVA-share-Sig-PhiLLM=Phi2-2.7B, Size=3.1B, Res.=3842024.0275.8
VILA-13BLLM=Llama-2-13B, Resolution=336, Pre-training samples=50M, Instruction tuning samples=1M2023.1273
LLaVA-1.5LLM=Vicuna-1.5-13B, Resolution=336, Pre-training samples=0.6M, Instruction tuning samples=0.7M2023.1270.7
VILA-7BLLM=Llama-2-7B, Resolution=336, Pre-training samples=50M, Instruction tuning samples=1M2023.1269.7
Full-FinetuneTarget Model=LLaVA-1.5-13B, Data Budget=100%2026.0569.5
Self-FilterTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.0569.1
MAGICTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.0569
PerplexityTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.0568.7
SemDedupTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.0568.7
RandomTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.0568.6
CLIP-ScoreTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.0568.1
COINCIDETarget Model=LLaVA-1.5-13B, Data Budget=20%2026.0567.4
TinyLLaVA-share-C-PhiLLM=Phi2-2.7B, Size=3.0B, Res.=3362024.0267.1
D2-PruningTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.0566.5
PADEintervention strength λ=0.12026.0265.92
EL2NTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.0565.8
V-ITI2026.0265.44
TrustLLaVALLM=Vicuna-7B, Resolution=3362026.0765.3
Self-SupTarget Model=LLaVA-1.5-13B, Data Budget=20%2026.0564.9
Vanilla2026.0264.8
OPERA2026.0264.31
LLaVA-1.5LLM=Vicuna-7B, Resolution=3362026.0764.2
HyperLLaVALLM=Vicuna-7B, Res.=336, PT=558K, IT=665K2024.0364
HyperLLaVA w/o EvLLM=Vicuna-7B, Res.=336, PT=558K, IT=665K2024.0363.7
LLaVA-1.5LLM=V-7B, Size=7B, Res.=3362024.0263.4
LLaVA-1.5LLM=Vicuna-7B, Res.=336, PT=558K, IT=665K2024.0363.4
LLaVA-1.5LLM=Vicuna-1.5-7B, Resolution=336, Pre-training samples=0.6M, Instruction tuning samples=0.7M2023.1263.4
LLaVA-1.5LLM=Vicuna-7B, Resolution=3362026.0763.4
Full-FinetuneDataset=Vision-Flan-186K, #Data=186k2026.0563.3
VCD2026.0263.21
MAGICDataset=Vision-Flan-186K, #Data=∼37k2026.0563.2
HyperLLaVA w/o ELLLM=Vicuna-7B, Res.=336, PT=558K, IT=665K2024.0362.6
INTER2026.0261.7
LLaVA-SPLLM=Vicuna-7B, Resolution=3362026.0761.7
InstructBLIPLLM=V-7B, Size=8.2B, Res.=2242024.0260.9
InstructBLIPLLM=Vicuna-7B, Resolution=224, Pre-training samples=129M, Instruction tuning samples=1.2M2023.1260.9
RandomDataset=Vision-Flan-186K, #Data=∼37k2026.0558.7
ICD2026.0256.9
TrustLLaVA (MLP)LLM=Vicuna-7B, Resolution=3362026.0751.5
BLIP-2LLM=Vicuna-13B, Resolution=224, Pre-training samples=129M, Instruction tuning samples=null2023.1238.1