[Core] Optimize request output tokens putting back implementation to reduce overhead

[Core] Optimize request output tokens putting back implementation to reduce overhead #20