Project Setup guide

Step 1: Load project in Android Studio

Step 2: Download the onnx model: model_q4f16.onnx from here: https://huggingface.co/HuggingFaceTB/SmolLM2-135M-Instruct/tree/main/onnx

Step 3: Place the onnx models here: mylibrary/src/main/assets

Project developer journal

This guide explains the steps to set up a project with a basic views activity and integrate ONNX for machine learning inference.

---

Steps

Step 1: Set Up a Basic Views Activity

• Task: Initialize a new project.
• Action: Create a basic activity to serve as the main UI.

---

Step 2: Create a Library Module

• Task: Add a new library module.
• Action: Name the module mylibrary to allow for modularization and reusability.

---

Step 3: Add Required Permission

• File: AndroidManifest.xml (in the app module).

• Action: Add the following permission for audio recording:

  <uses-permission android:name="android.permission.RECORD_AUDIO" />
  
  

Step 4: Add ONNX Dependency

• File: build.gradle (app module).

• Task: Integrate the ONNX Runtime library to enable inference using ONNX models.

• Action: Add the following dependency to the dependencies block:

  implementation("com.microsoft.onnxruntime:onnxruntime-android:latest.release")
  

Step 5: Download the ONNX Model

• Source: Download the model.onnx file from HuggingFace.
• Task: Place the model file in the assets folder of the app module.
• Action:
  1. Locate or create the assets directory under src/main.
  2. Copy the model.onnx file into this directory.

---

Step 6: Create OnnxReader Class

• Location: mylibrary module.
• Task: Implement a utility class for handling ONNX model operations.
• Action:
  1. Create a new Java/Kotlin class named OnnxReader.
  2. Write methods to:
     • Load the model.onnx file.
     • Execute inference using the ONNX Runtime library.

---

Step 7: Implement Inference in MyActivity

• Location: MyActivity in the app module.
• Task: Add logic for performing inference using the ONNX model.
• Action:
  1. Use the OnnxReader class to load the model.
  2. Process input data through the model.
  3. Display or utilize the output as needed.

Step 8: Implement the Tokenizer Class Using JSON File

• Location: mylibrary module.
• Task: Create a tokenizer class that processes input text into input_ids and attention_mask.