Evaluation error

I was trained on a Bloomz model, and when I tried to run trainer.evaluate(), I encountered a CUDA out of memory error and did not receive any results.

I have already written the code to compute metrics for the GLUE score. Can someone please provide suggestions for a text generation model?