felix-wang

@NOT-HAL9000 I'm glad to say now, ONNX M-CLIP models have been enabled in the latest version. Please play it around, and leave your comments.

That's a good idea. Actually, we are investigating whether bf16 would benefits (and how much) the HNSW search.

@OrangeSodahub FYI, this PR looks good to me. But I would like to hold it until https://github.com/jina-ai/annlite/pull/186 is merged. This PR is a good example to test the wheel building.

Actually, the document website is not yet ready. So, the document section in the README is just a placehoder.

Thanks for your suggestion. We will answer these questions in the FAQ section which will be included in the next release README.

> Each reload takes about 2 minutes, which is too slow We are working on refactoring reload function to allow restoring index from local backup. please track the PR https://github.com/jina-ai/annlite/pull/177

@geekchen007 FYI, in the latest annlite, the `dump` and `load` functions are supported. Please simply read the README https://github.com/jina-ai/annlite#dump-and-load

+1

tbh, i'm not sure I fully got your points. Anyway, we already integrated this embedding product into our Jina Cloud and make it accessible public https://cloud.jina.ai/user/inference. It works smoothly and...

ah, maybe this is the bug on Windows we did not observe before. Thanks for your report, we will look into it. BTW., what's the clip_server version are you using?...