fix: revert cast to cpu in MsgpackEncoder._encode_tensor to avoid hidden performance regressions#25738

vllm-bot merged 1 commit intovllm-project:mainvllm-project/vllm:mainfrom

protopia-ai:do-not-cast-to-cpu-in-serializationprotopia-ai/vllm:do-not-cast-to-cpu-in-serializationCopy head branch name to clipboard