[Revert] Fix performance regression for GLM-4.7-GPTQ decode and MTP acceptance rate#33771

LucasWilkinson merged 2 commits intovllm-project:mainvllm-project/vllm:mainfrom

aabbccddwasd:perf-regression-fixaabbccddwasd/vllm:perf-regression-fixCopy head branch name to clipboard