Mobile AI Factory automates specialized SLM creation for on-device inference

·2026.11.16 22:44

Key point

The agentic system outperforms two five-shot frontier models on five of ten tasks by up to 0.34 while using 19x less GPU time than exhaustive training.

Details

  • Mobile AI Factory is an agentic system that automatically builds fine-tuned, quantized small language models (SLMs) tailored to specific tasks and mobile devices.
  • On an unseen testset, the generated on-device models outperform two five-shot frontier models on five tasks by up to 0.34 and match them on three more.
  • The system's search strategy achieves results within 0.03 of exhaustively training all 27 model variants on an entity extraction task while using 19x less GPU time.
  • Phone-specific measurements are critical; without them, Claude Opus 5.5 misses app requirements in 25% of easy/moderate cases and 60% of hard cases despite having full hardware specifications.
  • The orchestrator LLM selects between hyperparameter tuning, real data mining, or synthetic data generation, with fixed rules rolling back changes that reduce accuracy.
  • Evaluation covered ten tasks across four Android phones, demonstrating that the optimal model configuration varies by device.

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.