[coreml] MLModelAsset: load failed with error

[aloveric]

Import your CoreML and run iOS code, it has this issue:

[aloveric]

BTW, Simulation works fine, error on real iPhone

[backnotprop]

I’ll give it a check, what is the iOS version on the phone?

…

On Sun, Jan 21, 2018 at 11:13 AM Eric Woo ***@***.***> wrote: BTW, Simulation works fine, error on real iPhone — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#1 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AG6KHTZKNg3Q4mDjM3lI8OB4_ziHRF7rks5tM2I1gaJpZM4Rl1n0> .

[aloveric]

@mdramos Thanks for your quick feedback. It is iPhone 6 and iOS 11.1.2. Also check with iPad Air 2 on iOS 11.1.2 with same issue.

[backnotprop]

related: https://forums.developer.apple.com/thread/82700

[aloveric]

Also found this but there is no solution to fix it ...

[jst000000]

i run the code on my iphoneX but its not work well but good with iphone7P and Simulation

[backnotprop]

I'm getting better results with lazy loading the models separately, still experiencing some memory issues, so I will continue to look for a better solution.

The first step in the right direction definitely seems to be lazy loading.

[backnotprop]

actually, a single model in memory seems to work best without issues. Just nil-ing out old models and keeping one in memory.

Ill push this fix in a bit, but will hopefully find a better solution for having multiple models loaded in memory.

[jst000000]

yes, most of phone work well with two models but break with three

[aloveric]

Guys, I left only one model and it will get whole black screen with loading then show the lock screen via iPhone 6 and iOS 11.1.2

[mithileshggn]

Running on simulator only but not in real device? please suggest why? Even though device is running with iOS 11.0

[bbthitenp]

Way not run in Real Device?

[backnotprop]

Hey all, I see folks stay actively commenting here.

The models this produce are most likely too large for devices. A new architecture needs to be slimmed down and there have been many advancements/examples to do this.

Im going to archive the repo, as I can not maintain and this should be used as experimentation only.