[BUG]: Difference between OPT example code

[waterhorse1]

🐛 Describe the bug

Hi guys, I am new to colossal-AI. Currently, I am trying the OPT-1.3B model with 8 RTX3090. I find two example codes, one in /tutorial/opt/opt and the other one in /language/opt. I can successfully run the finetuning on the code in /language/opt while I meet OOM when using the code in /tutorial/opt/opt. Basically, I change nothing but the model size and keep other variables the same. Does anyone know why?

Environment

No response