Getting Models

Every model in the RunMyAI Selection tab was downloaded, loaded and run by the RunMyAI team before it was added. Each card says which platform and which RunMyAI version it was tested on.

Models are grouped in three tiers; Tier 1 holds the top recommendations. Use the filter at the top of the tab to show one tier or all of them.

Each quantization lists its file size, the RAM it needs and an estimated speed. Models that may not fit in your memory carry a Large tag, and a quantization too big for your machine says so before you download it.

Pick a quantization and download it. The format (GGUF or MLX) and the engine that runs it are chosen for you from the file itself.

Every file of a RunMyAI Selection model comes from one place: the RunMyAI model mirror, which serves the exact copy the team tested. There is no second source to fall back to, because a second source could serve different bytes.

The catalog inside the app pins the SHA-256 digest of every file. The download is hashed while it streams, and a file whose digest does not match is never installed -- the download fails instead.

An interrupted download keeps what it already received and resumes from there. A failed download shows a Retry button in the download list.

A model in your local list tagged Unverified is a download that could not be checked against its publisher's SHA-256 digests. Only structural format checks were applied to it.

Never on a RunMyAI Selection model: those are always checked against the pinned digests. It can appear on a model you downloaded from the Hugging Face browser when the repository publishes no digest for a weight file.

The model still loads and runs. If you want a checked copy, download the same model from the RunMyAI Selection, or from a repository that publishes digests for its files.