On free tier, Colab runs out of RAM during the dummy benchmark: https://colab.research.google.com/drive/15YLOWHB_NkIIqKLO0ik784fsK2xJD08l?usp=sharing
A task.unload() function would delete task.df, and call gc.collect(), i.e.:
import gc
...
task.load()
...
del task.df
gc.collect()
...
This gets it a bit further (completes mp_e_form which it didn't before, but crashes at mp_gap). What other large objects could be safely deleted while preserving the benchmark results?