LensVLM: Compressing long context as images, expanding only relevant pages

11 points - today at 6:36 PM