ML model doesn't load on background thread

[bhavin250495]

🐞Describe the bug

• The inference runs fine when the app is in foreground but when it goes background, and tries to run inference gets crashed or takes long time to load model MLModel.init()

Trace

2020-01-31 12:17:10.135653+0530 ********** [3505:875445] [coreml] Error in adding network -1.
2020-01-31 12:17:10.139437+0530 ********** [3505:875445] [coreml] MLModelAsset: load failed with error Error Domain=com.apple.CoreML Code=0 "Error in declaring network." UserInfo={NSLocalizedDescription=Error in declaring network.}
2020-01-31 12:17:10.140146+0530 ********** [3505:875445] [coreml] MLModelAsset: modelWithError: load failed with error Error Domain=com.apple.CoreML Code=0 "Error in declaring network." UserInfo={NSLocalizedDescription=Error in declaring network.}
Fatal error: 'try!' expression unexpectedly raised an error: Error Domain=com.apple.CoreML Code=0 "Error in declaring network." UserInfo={NSLocalizedDescription=Error in declaring network.}: file ML.swift, line 105

To Reproduce

• Just load the model when app in background

Default generated file to run ML model
/// Construct a model that automatically loads the model from the app's bundle
    convenience init() {
        
        try! self.init(contentsOf: model.urlOfModelInThisBundle)
    }

demoModel.mlmodel.zip

System environment (please complete the following information):

• coremltools version (e.g., 3.0b5):

• OS: MAC OS

• macOS version (if applicable): 14

• XCode version (if applicable): 11.3.1]

• How you install python (anaconda, virtualenv, system): system

• python version (e.g. 3.7): 3.7

• any other relevant information: 2.3

  • keras version: 2.3

[anilkatti]

@bhavin250495 could you share the code that you are using to run the prediction in the background? One thing to keep in mind is background mode is enabled only on CPU. You can do this by setting MLModelConfiguration.computeUnits.

[bhavin250495]

@anilkatti Thanks for replying, I tried adding MLPredcitionOptions but i don't have any problem
with prediction, the issue is with loading of ML model from local URL

1. I trigger a silent notification from the server to start inference on app
2. When app gets silent notification, it does some preprocessing and creates input for MLmodel
3. Now when i am trying to run prediction on MLmodel when the app is in background it gets crashed or just take forever to load model

 func predict(input:MLInput) throws -> Bool{
        print("Loading model")
//Model does't load here when in background
        if let model = SmartML(){
            let options = MLPredictionOptions.init()
            options.usesCPUOnly = true
            let outputLabel = try nudgeModel.prediction(input: input:MLInput,options: options).classLabel
            
            let status = (outputLabel == "yes") ? true : false
            return status
        }
        return false
    }

[anilkatti]

My suggestion was to use SmartML(configuration:) method to load the model but, I did not realize that you were running this on MacOS. Based on the error, it seem like the app is having issue reading the model files from disk in the background mode. Could you share a repro case so, I can debug the root-cause? The model itself looks good.

[bhavin250495]

I am using this model on iOS , and if i share the repo to reproduce the error you will need to setup silent notifications . Could you run the inference using local timer in background to reproduce the error

[hfnvbh]

Hi all,
Is there any updates about this issue?
I met same problem, when coreml model loaded in background mode after silent notification in my ios app, I have same message "Error in declaring network."

When I inited model in foreground, or in BGProcessing mode, it's ok.

[anilkatti]

Silent notifications are not meant as a way to keep your app awake in the background beyond quick refresh operations, nor are they meant for high priority updates.

I am looking for clear guidance in this scenario. I will update this tread.