It is better to know than assume, so let's hear from them :)

That is my main fear. 😅

15 models won't be mobile friendly, so we would rather hear from them than assume

Wait, have you been training specialized models?

Fair enough

🤔

Or an ensemble of 15 models. Other than the upcoming minor shake-up, the decision to this question will separate cash prize winners from the other winners.

Even with a heavy gpu it takes 10s to 2 mins if you are infering. I should have seen this from the start. That changes everything to be honest😭

Yes @crossentropy

True @nymfree . It would be so bad training specialized models , then during evaluation you get disqualified due to the mobile friendliness. We would rather ask than assume.

Wow!

Never even thought of that until i saw this discussion

Thank You for this

would be great if you could post it as a new message so that those not following this thread are also informed. It is highly likely that top scorers are using ensembles.

@meganomaly I think this is something that would have helped to clarify much earlier. I (and many others) read the rules and description, and since this wasn’t mentioned, it seemed fair to assume there were no restrictions.

Sharing this so close to the end is likely (and rightfully going) to create some tension.

We hear the concerns. We want to be fair and allow what will give the best solution - on a mobile app. While 1 may be too restrictive, 15 is also not feasible.

As a compromise we would suggest aiming for 3 or 4 models aligned with the variable characteristics.

@meganomaly, I think it will be best to post this recent update on a new thread just as @nymfree suggested.

Will do that as soon as I can 👍