[Perf] Refactor tensor disposal logic to reduce memory usage#966

ganyi1996ppo merged 3 commits intovllm-project:mainvllm-project/vllm-ascend:mainfrom

ApsarasX:community-memory-optimizationApsarasX/vllm-ascend:community-memory-optimizationCopy head branch name to clipboard