List your company
ResearchFormerly WizWand

Spatial Reasoning on SpatialEval Maze-Nav 20% (train)

86.2AccuracyLoRA-all
23.59239.84656.172.354Jun 28, 2026
Updated 4mo ago

Evaluation results

MethodLinks
LoRA-allBackbone=Qwen2.5-VL-7B, Trainable Parameters=2.97M2026.0686.2
(IA)3Backbone=Qwen2.5-VL-3B, Trainable Parameters=524.2K2026.0682.2
(IA)3Backbone=Qwen2.5-VL-7B, Trainable Parameters=668.5K2026.0680.9
LoRA-allBackbone=Qwen2.5-VL-3B, Trainable Parameters=2.32M2026.0679.9
LoRA-allBackbone=LLaVA-13B, Trainable Parameters=4.06M2026.0675.2
LoRA-allBackbone=LLaVA-7B, Trainable Parameters=2.65M2026.0671.6
(IA)3Backbone=LLaVA-7B, Trainable Parameters=663.6K2026.0670.8
(IA)3Backbone=LLaVA-13B, Trainable Parameters=1.01M2026.0670.5
ScAle (attn)Backbone=Qwen2.5-VL-3B, Trainable Parameters=5762026.0670.3
ScAle (attn)Backbone=LLaVA-13B, Trainable Parameters=1.6K2026.0669.8
ScAle (attn)Backbone=LLaVA-7B, Trainable Parameters=1.0K2026.0668.9
LoRA-lastBackbone=Qwen2.5-VL-7B, Trainable Parameters=90.1K2026.0666.5
ScAle (attn)Backbone=Qwen2.5-VL-7B, Trainable Parameters=7842026.0666.2
LoRA-lastBackbone=LLaVA-7B, Trainable Parameters=78.1K2026.0653.6
LoRA-lastBackbone=LLaVA-13B, Trainable Parameters=97.8K2026.0646.2
LoRA-lastBackbone=Qwen2.5-VL-3B, Trainable Parameters=52.0K2026.0641.8
FrozenBackbone=Qwen2.5-VL-7B, Trainable Parameters=02026.0638.3
FrozenBackbone=LLaVA-7B, Trainable Parameters=02026.0631.7
FrozenBackbone=LLaVA-13B, Trainable Parameters=02026.0626.7
FrozenBackbone=Qwen2.5-VL-3B, Trainable Parameters=02026.0626