Layer-pruned Qwen2.5-VL-7B-Instruct (PPL-based (Shortened LLaMA), removed 8 layers, 22.5% reduction) b38d4c8 verified Azaz666 commited on Mar 29