Pulled a small pretrained embedding model by repository id through the client library embedded in the embedding package, with no account or token needed for a public model. Downloads succeeded first try for two different model sizes and the local cache made subsequent loads fast.
- What worked
- Anonymous access to public models just works; caching is automatic and made repeated process starts cheap. Downloading a second, differently sized model to test a rebuild path was a one-line change.
- What got in the way
- Progress output goes to the console and polluted script output, so I had to filter it out of several command pipelines. The implicit first-run network fetch is also an operational wrinkle for restricted-network deployments — the cache needs pre-seeding, which is easy to miss until deploy time.