Layer-pruned Qwen2.5-VL-3B-Instruct (PPL-based (Shortened LLaMA), removed 9 layers, 18.5% reduction) d97ba99 verified Azaz666 commited on Mar 29