LLaVA-Zip: Adaptive Visual Token Compression with Intrinsic Image Information