[Feature Request]: timeout parameter OllamaEmbeddingFunction

[Davidyz]

Describe the problem

When embedding large documents with a not-so-fast Ollama server over a poor network, sometimes the operation times out.

Describe the proposed solution

Add a float type parameter timeout for OllamaEmbeddingFunction, which controls the timeout for the httpx.Client object here.

ref: https://www.python-httpx.org/advanced/timeouts/

Alternatives considered

I was going to submit an PR, but then I saw #2946 which refactors OllamaEmbeddingFunction to use the Ollama python library. I'm not sure when (or whether) this will be merged, but if it ever gets merged, the proposed solution will no longer be appropriate.

Importance

would make my life easier

Additional Information

No response

[tazarov]

@Davidyz, thanks for raising this. I'll review and rebase #2946 and merge it soon.

[tazarov]

@Davidyz, rabased on latest main and added timeout kwarg to pass to ollama client.

[Davidyz]

Hi, thanks for the update. I tested it and it works, but this needs to be marked as a breaking change because Ollama API doesn't accept 'url' (like http://127.0.0.1:11434/api/embeddings), but rather server host (like http://127.0.0.1:11434). The original configuration simply wouldn't work.

With that said, may I ask what's the estimated time of this being merged into main? (presumably in 0.7?)

EDIT: to be clear, the breaking was due to the shift to Ollama python library, not the addition of timeout.

[tazarov]

@Davidyz, thanks for testing it out. Indeed, the URL part is a breaking change, and I'll update the docs to reflect this. That said, I've added support for backward compatibility with legacy URLs (e.g. ending with /api/embeddings).

As far as the release goes, possibly in 0.6.4, but once merged, you can always grab the release from main.

[Davidyz]

@Davidyz, thanks for testing it out. Indeed, the URL part is a breaking change, and I'll update the docs to reflect this. That said, I've added support for backward compatibility with legacy URLs (e.g. ending with /api/embeddings).

As far as the release goes, possibly in 0.6.4, but once merged, you can always grab the release from main.

I see. Thanks for the updates. I think I'll still wait for the release so that it's easier to write the version constraints in my pyproject.toml.