GPU Memory issues during evaluation

[RibinMathewChalumattu]

Hey Chris,

I tried to run the evaluation.py script with the following model facebook/mbart-large-cc25 (fine-tuned on a german dataset for the summarization task). However, during evaluation, I ran into GPU memory issues (on an NVIDIA GeForce RTX 2070 SUPER with 8 GB) with a beam size of 3 and a maximum of 3 articles per cluster. Do you have some suggestions on reducing the GPU memory footprint? Could you also share the resources you used to evaluate the DynE decoder?

Best,

Ribin