Less Is More, but Where? Dynamic Token Compression via LLM-Guided Keyframe Prior

Open in new window