base_eval.py: hellaswag 在小型模型上逐渐变慢并泄露内存 (Mac Studio)

I'm running `base_eval.py` on a very small model trained with `base_train.py`. Evaluation starts normally, but on the **hellaswag** task each example becomes slower and slower, memory usage grows continuously, and the script will take massive amount of time to finish. Other tasks finish in minutes, but hellaswag runs for hours.