Period.

2024 - 2025

• General model conversion successful
• Stateful model conversion successful
• On-device porting experiments in progress

Used Tools & Skills

Skills

• Core ML / MLCompute
• Swift
• On-device inference pipeline configuration

Motivation

• I wanted to build ML models optimized for Apple CPU, GPU, and NPU that could be used in mobile keyboards.
• Knowing that CoreML doesn't track memory in the limited mobile keyboard memory resources, I wanted to implement AI-powered autocomplete inference logic for keyboards.

Learned

• Learned PyTorch model conversion methods using coremltools.
• Studied the conversion logic to Stateful models using KV-cache introduced in iOS 18.