Bug: documentation for server.cpp --embeddings is confusing/incorrect

[okigan]

What happened?

Following up on discussion in #8420, the documentation for server.cpp --embeddings is confusing/incorrect.

Perhaps the original usage of the --embedding flag has morphed overtime but documentation does not reflect it.

With that proposing to change the documentation to following

--embedding(s)   restrict to only support embedding use case (default: %s). Use only with dedicated embedding models.

Note the llama-server already refuses non-embedding endpoints if the above flag is turned on:

• chat
• chat completion
• infil

Name and Version

./llama-cli --version
version: 3488 (75af08c)
built with Apple clang version 15.0.0 (clang-1500.3.9.4) for arm64-apple-darwin23.5.0

What operating system are you seeing the problem on?

Linux

Relevant log output

No response

[iamlemec]

@okigan Yup, this looks good to me! The usage has indeed evolved. Now that most any model can be coaxed into providing embeddings, it's mostly to prevent embedding-only models from crashing out on a completion request.

Would be great if you could do a PR.

[okigan]

@iamlemec here is the PR, pending your review. Thanks!