felix-wang

we are using allowed/blocklisted candidates, and check membership during graph search.

For your question, actually, all of the graph nodes are considered valid during graph traversal. Only the nodes which match with the filter will be considered as the result candidates....

@tommykoctur The subindex has been released. https://docarray.jina.ai/fundamentals/documentarray/subindex/

You are talking about the training of quantizer?

What's the API design in annlite? Is this what we are expected: ``` def save_model(self, name: Union[str, PathLike]): if isinstance(name, str): # save to hubble with the given name `name`...

@myntoralex Currently, persistence is already supported via **local-disk kv-store** `lmdb` (the data is located at `workspace` folder defaultly). Is this local-disk persistence is what you expected? Or do you want...

I see, I think you want external persistence supports actually. And there maybe two branches we can proceed: 1. supporting external database e.g., ElasticeSearch, Postgres etc., 2. supporting object storage...

@mirzakhalov Thanks for pointing it out. Regarding your concern, there is no trivial approach to match your case in AnnLite. Only the types of scalar data are supported since the...

@ZiniuYu Please polish the PR's title. `4 new clip models` is meaningless. Please highlights the contribution in the title

It's hard to offer the exact timeline for supporting ONNX M-CLIP models. However, It's hopefully we can release it by end of this month.