Gemma 4 26B runs on an iPhone 17 Pro via model paging
Agreeable-Rest9162 · reddit · 2026-07-25
Noema founder says Gemma 4 26B A4B is running on an iPhone 17 Pro using model paging.
The setup keeps non-expert weights in RAM and reads expert weights from SSD, trading speed for the ability to run a much larger model on-device. The post reports a 699-token prompt, 34.4 tokens/s prefill, 20.34 seconds prefill time, and 3.5 tokens/s decode speed. The full answer took about 6 minutes, but the author argues that the approach could be useful when accuracy matters more than latency, including on low-RAM MacBooks.
More from Infra
- $5M Seed Round Can't Afford One DeepSeek Training Run: GPU-less AI Startup's Three Paths — FinanceYF5 · 2026-10-26
- HBM demand shifts: low-stack inference plans shelved, Rubin to go 12-high only — BenBajarin · 2026-10-26
- CircuitHub unveils The Grid, a factory-scale robot making PCBs in the US twice as fast as Shenzhen — garrytan · 2026-10-26
- Ubicloud publishes "Fixing Tool-Call Accuracy in vLLM" — petrusenko_max · 2026-10-26
- Meta approves nonvoting Class C shares as 2026 capex guidance hits $135B–$145B — thione · 2026-10-26
- Unsloth Ships MiniMax-M3.1-Flash Quants: 2-bit Retains 91.5% Accuracy, Runs in 150GB RAM — danielhanchen · 2026-10-26