List your company
ResearchFormerly WizWand

Multimodal Understanding on MMMU (Accuracy, Rel. (%))

54.1AccuracyFull Model
7.40419.52731.6543.773Apr 14, 2026Apr 19, 2026Apr 25, 2026May 1, 2026May 6, 2026May 12, 2026May 18, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Full ModelBackbone=Qwen2-VL-7B, Sparsity Ratio=0%2026.0454.1100
TopoVLMBackbone=Qwen2-VL-7B, Sparsity Ratio=50%2026.0452.895.4
TAMPBackbone=Qwen2-VL-7B, Sparsity Ratio=50%2026.0451.493.2
ReGapModel=Qwen3-Omni-30B-A3B-Instruct2026.0550.6—
WandaBackbone=Qwen2-VL-7B, Sparsity Ratio=50%2026.0450.289.3
BaseModel=Qwen3-Omni-30B-A3B-Instruct2026.0550—
LLM-StreamlineBackbone=Qwen2-VL-7B, Sparsity Ratio=50%2026.0449.891.2
ECoFLaPBackbone=Qwen2-VL-7B, Sparsity Ratio=50%2026.0449.590.5
SparseGPTBackbone=Qwen2-VL-7B, Sparsity Ratio=50%2026.0448.990.1
OWLBackbone=Qwen2-VL-7B, Sparsity Ratio=50%2026.0448.388.4
LLM-PrunerBackbone=Qwen2-VL-7B, Sparsity Ratio=50%2026.0447.688.6
ReGapModel=MiniCPM-o-4.52026.0545.4—
BaseModel=MiniCPM-o-4.52026.0543.4—
DiM3Model Size=7B2026.0537.22—
LLaVA-CKDLLM=Qwen2.5-1.5B, #TS=1.2M2026.0536.9—
Task Arithmetic2026.0536.89—
NeuroMerging2026.0536.89—
LLaVA-v1.5Model Size=7B2026.0536.33—
TIES-Merging2026.0536.33—
DARE2026.0535.78—
LLaVA-MORELLM=Gemma-2-2B, #TS=1.2M2026.0533.4—
Breadcrumbs2026.0533.33—
ALLaVALLM=StableLM-2-1.6B, #TS=1.2M2026.0533.3—
PaloModel Size=7B2026.0533.11—
LLaVA-CKDLLM=Qwen2.5-0.5B, #TS=1.2M2026.0532.8—
MiniGeminiLLM=Gemma-2B, #TS=2.2M2026.0531.7—
EM-KDLLM=Qwen2.5-0.5B, #TS=1.3M2026.0531—
STF2026.0530.67—
LVIDALLM=Llama-3.2-1B, #TS=1.2M2026.0529.9—
Refusal SteeringModel=Qwen3-Omni-30B-A3B-Instruct2026.0525.6—
ReGapModel=Qwen2.5-Omni-3B2026.0525.4—
BaseModel=Qwen2.5-Omni-3B2026.0525—
Refusal SteeringModel=MiniCPM-o-4.52026.0516.2—
Refusal SteeringModel=Qwen2.5-Omni-3B2026.059.2—