Run SmolLM2 on Android

Copy the model package to application-private storage

The application loads each model from filesDir/models/<model-id>/. For development, stage the downloaded package under /data/local/tmp, then use Android’s run-as command to copy it into the debuggable application’s private directory.

Run the following commands from the starter application repository:

    

        
        

export ANDROID_PACKAGE="com.arm.learningpath.texttotext"
export MODEL_ID="smollm2-360m-instruct-8da4w-xnnpack-executorch"

adb shell "mkdir -p /data/local/tmp/text-to-text-models/$MODEL_ID"
adb push ../model/. "/data/local/tmp/text-to-text-models/$MODEL_ID/"

adb shell "run-as $ANDROID_PACKAGE mkdir -p files/models/$MODEL_ID"
adb shell "run-as $ANDROID_PACKAGE cp -r /data/local/tmp/text-to-text-models/$MODEL_ID/. files/models/$MODEL_ID/"
adb shell "run-as $ANDROID_PACKAGE find files/models/$MODEL_ID -maxdepth 4 -type f"
  

    

    

        
        

$ANDROID_PACKAGE = "com.arm.learningpath.texttotext"
$MODEL_ID = "smollm2-360m-instruct-8da4w-xnnpack-executorch"

adb shell "mkdir -p /data/local/tmp/text-to-text-models/$MODEL_ID"
adb push ..\model\. "/data/local/tmp/text-to-text-models/$MODEL_ID/"

adb shell "run-as $ANDROID_PACKAGE mkdir -p files/models/$MODEL_ID"
adb shell "run-as $ANDROID_PACKAGE cp -r /data/local/tmp/text-to-text-models/$MODEL_ID/. files/models/$MODEL_ID/"
adb shell "run-as $ANDROID_PACKAGE find files/models/$MODEL_ID -maxdepth 4 -type f"
  

    

The final command lists the .pte program, config.yaml, and tokenizer files under the model ID directory. That directory name must match the id field in model_catalog.json.

Generate text

Start or return to the application:

    

        
        
adb shell am start -n com.arm.learningpath.texttotext/.ui.MainActivity

    

To run the model:

1. Select SmolLM2 360M Instruct 8da4w.

2. Select Load and wait for the status to report the model load time.

3. Enter the following prompt:

       
   
           
           
   Write one sentence explaining why on-device AI is more private than cloud AI.
   
       
   

4. Select Run.

The result area displays a generated completion together with model load and run times. The wording can vary because this is a generative model, but it should answer the prompt and shouldn’t contain chat-template markers or tokenizer control tokens.

The timing is illustrative rather than a benchmark. Measure the model on your target phone before making deployment choices.

What you’ve accomplished and what’s next

You’ve now copied the complete SmolLM2 package into application-private storage and generated text locally with the packaged ExecuTorch runtime.

Next, you’ll compare the validated text-generation adapters.