Hi everyone,
I wanted to bring up something that's been affecting our group: on Colab Free, the T4 GPU is not always assigned when you start a session. Sometimes you get CPU-only, which makes running the models either very slow or completely infeasible.
In our case, CPU is not even an option, the models we're using require more memory, so we need the T4.
Are you just retrying until you get a T4, scheduling runs at off-peak hours, or are you using smaller models?
Also, would it be fine to run the notebook locally instead of on Colab?
Any tips welcome.


