Pretrained models

[Peter Fackeldey]

Hi @erikbuh @WilliamKorcari ,

So far the the repository supports the model reference model in terms of code.
I suppose some models have huge training requirements and long runtimes. What do you think of providing pretrained models for the reference models? Maybe they can be hosted in the same way as the datasets and loaded on demand, if one does not has the resources or time to retrain from start.
This is just an idea which came to my mind earlier, let me know what you think :)

(Another thing: Due to non-deterministic initialisation the training result will be slightly different all the time. Therefore, when quoting the performance of a model publicly, it would be safer and more correct to have pretrained models, which behave always the same.)

Best, Peter

[Erik Buhmann]

Hi @pfackeldey,

I agree that would a very nice addition to the repo. We could indeed train all the models once and upload the corresponding weights to our desy cloud.
At the moment I don't really have time to implement this however. If you find the time, feel free to give it a go. Otherwise we can leave the issue open for now and come back to it later.

(And I agree, when quoting the models performance we should maybe train each model a couple of times to be confident in its performance / be able to quote an uncertainty. We should definitely keep this in mind when we have all the models ready and are doing our final trainings.)

Thanks & cheers,
Erik

[Peter Fackeldey]

Thanks for your reply @erikbuh !
I can have a look and see if I find a suitable implementation in the next days :)