Search for a command to run...
Fine-Grained Preference Optimization Improves Spatial Reasoning in VLMs