Fine-Grained Preference Optimization Improves Spatial Reasoning in VLMs | Read Paper on Bytez