Expand description
inillucent setup-embeddings: install the embedder, on any of three
platforms, in one command.
Invariant: a component this reports as installed is verified, complete and
reachable without an environment variable. Every archive and every weights
file is checked against a digest pinned in this file before anything is
written where the engine looks, the manifest is sealed from the files that
actually landed rather than from what was expected, and the directories are
the ones inillucent_core::install computes - so embed(TEXT) works
immediately afterwards on a shell that has never exported anything.
§What is pinned, and why by digest rather than only by version
A version pins what was asked for. A digest pins what arrived. The two differ whenever a release asset is replaced, a content network serves a truncated body, or something between here and the origin rewrites it - and the failure mode of the first two is a shared library that loads and misbehaves rather than one that refuses. So the digests are here, in the source, and a mismatch deletes what it fetched and names both digests.
A version this build has no digest for is still installable, because pinning
every future release is not possible and refusing them would make
--onnxruntime-version useless. It is fetched, it is reported as
unverified in the output and in the install state, and the state file is
where somebody later finds out that this one was taken on trust.
§Why 1.22.0
It is the version this repository’s embedding numbers were taken on, and it
is the last release Microsoft publishes a universal2 macOS archive for.
After it, macOS is two archives and an installer that picks the wrong one is
a support question rather than an error.
Enums§
- Accelerator
- Downloads and installs the ONNX Runtime shared library.
- Reinstall
- Whether an install replaces a runtime that is already there.
Constants§
- DEFAULT_
RUNTIME - The ONNX Runtime version installed when nothing else is asked for.
Functions§
- setup_
embeddings setup-embeddings: download and install the embedder.