ELSER - Elastic Learned Sparse EncodeR model is unable using Trial License

[Rahul_Agarwal1 (Rahul Agarwal)]

I want to play with the new ELSER model released by Elasticsearch in new release 8.8. I tried installing Elasticsearch 8.8 in my local system and tried with the trial license but the model is not available under: Stack Management -> Machine Learning -> Trained Models

How can I try this in my local?

[stephenb (Stephen Brown)]

Hi @Rahul_Agarwal1

Welcome to the community and thanks for checking out ELSER

You are not looking in the right place

You are in Stack Management that is not correct

See the docs

You Need to Navigate to the Machine Learning in the Top Level Menu The Trained Models

[Rahul_Agarwal1 (Rahul Agarwal)]

Thanks for the quick response !!

Also elasticsearch.py client library of python doesn't support the "rank" param yet(it's required to check the RRF ranking feature). When are you guys planning to add it?

[stephenb (Stephen Brown)]

We do not publish release dates in advance.
Typically the Language Clients follow pretty quick.
You could use a RAW HTTP request
@dadoonet Any thoughts?

soo you can not just add it here, I see it is not in the docs, but did you try it?

resp = es.search(index="test-index", query={"match_all": {}}, rrf={....})

[Rahul_Agarwal1 (Rahul Agarwal)]

Yes I tried this. Its giving an error:
TypeError: Elasticsearch.search() got an unexpected keyword argument 'rrf'

[dadoonet (David Pilato)]

@stephenb

We do not publish release dates in advance.
Typically the Language Clients follow pretty quick.
You could use a RAW HTTP request
@dadoonet Any thoughts?

Nope but @sethmlarson might know

[sethmlarson (Seth Michael Larson)]

Hello! Sorry for the inconvenience here, we're happy that everyone is excited about vector search and wants to use it through the language clients. The issue was caused by a missing annotation in our API specification which is being fixed here. That fix will make its way into the next release of the clients, in the case of the Python Elasticsearch client that will be v8.8.1 in a few weeks (note that the expected added parameter will be rank={"rrf": ...} instead of rrf=...)

As a workaround, you can send an HTTP request body which will complain of a deprecation warning for now but will still allow you to send the request:

from elasticsearch import Elasticsearch

client = Elasticsearch(...)

client.search(index="...", body={"rank": {"rrf": {"window_size": 100}})

Thanks for your interest in Elasticsearch and the Python client!

[Rahul_Agarwal1 (Rahul Agarwal)]

Happy to know this !!
Thanks for the prompt response guys👍🏻

[Rahul_Agarwal1 (Rahul Agarwal)]

@sethmlarson Even the "body" param is not working.