Add BM25 options to inference config#6893
Add BM25 options to inference config#6893JojiiOfficial wants to merge 5 commits intouncouple_tokenizer_and_textindex_paramsfrom
Conversation
e8c4b13 to
345c165
Compare
| address: "http://localhost:2114/api/v1/infer" | ||
| timeout: 10 | ||
| token: "98eb568c-dea5-4347-a3a5-583478983bb9" | ||
| # Define custom models/vectorizer, which get handled directly in Qdrant. These do not require an `address` to be configured. |
There was a problem hiding this comment.
Maybe they can, but it is up to the specific model params. In the case of BM25 it is not needed bc it is computed in qdrant directly.
There was a problem hiding this comment.
What do you mean? custom_models is supposed to only handle 'models' that are handled directly in Qdrant, like BM25.
There was a problem hiding this comment.
I was thinking that maybe in the future users might be able to plug in their own embedding services via a custom config like this. In this case another default namespace might be more fitting like bm25/bm25-for-paragraphs, or native/bm25
generall
left a comment
There was a problem hiding this comment.
I am not convinced inference config is the right place for this settings.
Inference settings are global, while we would want to have it per-collection.
I propose to postpone this change and make it part of custom collection parameters instead.
|
Closing in favor of #6939 |
Depends on #6891
Adds configuration for BM25 vectorizer "models" in inference. One example model with the name
bm25added in development config too.Highlights: