Efficient LLaMA-3.2-Vision by Trimming Cross-attended Visual Features Paper • 2504.00557 • Published Apr 1 • 15
