VCM: Vision Concept Modeling with Adaptive Vision Token Compression via Instruction Fine-Tuning | Read Paper on Bytez