Compute Required to train baselines?

[seanbenhur]

Thanks for sharing the baseline code

Can you share the compute required to train these models on the entire dataset?

[franbvalero]

Hello @seanbenhur thanks for asking.

Each task contains the corresponding script for replicating the experiments with the results presented, therefore, for each task you have an specific script which uses the corresponding dataset of the task (check the the README).

If you want to train a model with different data, you should change the corresponding variables related with the input of the model in the corresponding script.

Ask me in more detail if I do not answer your question properly.

[seanbenhur]

Thanks for the details @franbvalero

When I mean compute, I mean the GPUs required for fine-tuning the models, what configuration of GPUs(P100/V100) were used to train the models>, How long did it take to train on the entire dataset?

[seanbenhur]

Hi @franbvalero will you be able to share the info?

[franbvalero]

I can share you an approximation. For all the experiments we used one V100.

The training for the task 1 was close to 2 hours and for tasks 2 and 3 close to 10 hours (each one).

[seanbenhur]

Thank you for sharing the info!