Skip to main content
The Run guides cover each interface on its own. Tutorials go the other way: one goal, start to finish, on a specific class of Snapdragon hardware — the model pull, the flags, the numbers you should see, and what to do when they don’t show up. Each tutorial states the platform it was verified on at the top. Most techniques carry across families, but the model sizes and compute units that make sense do not — a 26B target that fits comfortably on a Copilot+ PC will not fit an IoT board.

​
Compute — Windows ARM64 / Copilot+ PC

Snapdragon X-series laptops and desktops, where memory headroom allows large targets and multi-model setups.

Speculative decoding with MTP

Accelerate Gemma-4-26B decoding with a Multi-Token Prediction draft model, in both geniex infer and the local server.

​
Before you start

Every tutorial assumes you have:
  • The CLI installed and on your PATH — see Install.
  • A supported Snapdragon device, or a remote session on Qualcomm Developer Cloud — see Platforms & runtimes.
  • Enough free disk for the models involved. Tutorials list sizes up front; geniex list shows what you already have cached.
New to GenieX entirely? Start with the Quickstart and come back here once you have a model answering prompts.