search

Unbabel / OpenKiwi

Open-Source Machine Translation Quality Estimation in PyTorch
https://unbabel.github.io/OpenKiwi/
GNU Affero General Public License v3.0
229 stars 48 forks source link

Add model parameters log to mlflow #50

Closed francoishernandez closed 4 years ago

francoishernandez commented 4 years ago

Seems useful to have the number of parameters in there. Feel free to close if you don't feel the same.

captainvera commented 4 years ago

We had never considered this. Why do you think it'd be useful to have the parameters? So you could (if needed) re-construct the model from an old mlflow run? I'm not sure this is a practical necessity and might just be adding noise to the mlflow logs, but I might be missing something here!

What are your motivations?

francoishernandez commented 4 years ago

I mean number of parameters, not the parameters themselves. Not sure that's adding that much noise. It's quite useful when tuning model architectures to have a rough idea of the variation of the size of the model I reckon. Screenshot from 2019-12-11 17-41-33

captainvera commented 4 years ago

Oh! I'm really sorry, I absolutely misread your original message. Yes, I feel this shouldn't add a lot of noise to the mlflow logs. Looks good to me :)

Again thank you for your contribution!