[static cache] fix device map per layer in VLMs#38488

zucchini-nlp merged 2 commits intohuggingface:mainhuggingface/transformers:mainfrom

zucchini-nlp:multi-gpu-vlmszucchini-nlp/transformers:multi-gpu-vlmsCopy head branch name to clipboard