HalLoc: Token-level Localization of Hallucinations for Vision Language Models | Read Paper on Bytez