Mobile AI Factory: agentic pipeline builds on-device SLMs that beat 5-shot frontier models
kalyan_kpl · x · 2026-11-11
A paper presents Mobile AI Factory, an agentic system that builds a fine-tuned, quantized SLM for a given task and phone.
How it works:
- An orchestrator LLM decides whether to tune hyperparameters, mine real data, or generate synthetic examples, while fixed rules roll back changes that hurt accuracy.
- It searches small-to-large across base models and quantizations, adapts its accuracy goal during search, and uses real on-device measurements of the target phone across inference engines and backends.
Results: on an unseen test set, its on-device models outperform two 5-shot frontier models on five tasks (by up to 0.34) and match them on three more.
More from Infra
- "Compute is about power": the AI race's real bottleneck is energy, not chips — _sholtodouglas · 2026-11-11
- CoreWeave CEO says he canceled his 10b5-1 plan, halting insider selling — firstadopter · 2026-11-11
- Nvidia RTX 5080 24GB reportedly cancelled, 16GB model remains — chemist_slime · 2026-11-11
- AMAT reports Thursday: DRAM hits 26% of tool sales, outlook is the real signal — tengyanAI · 2026-11-11
- Starlink urban US users grow 73%, double rural rate, as upload speeds jump 42% — XFreeze · 2026-11-11
- Cloudflare picks QuePaxa, an unproven consensus algorithm, for Meerkat service — sull · 2026-11-11