[ML] Adding elser default endpoint for EIS

[jonathan-buttner]

This PR adds the elser v2 default inference endpoint for EIS.

model id: elser-v2
default endpoint id: .elser-v2-elastic

Testing

1. Run the model server locally
2. Run the gateway locally

Request using the default endpoint

curl --location 'http://localhost:9200/_inference/sparse_embedding/.elser-v2-elastic' \
--header 'Content-Type: application/json' \
--header 'Authorization: Basic <auth>' \
--data '{
    "input": "What is the weather in DC?"
}'

Example response

{
    "sparse_embedding": [
        {
            "is_truncated": false,
            "embedding": {
                "arizona": 0.004688698,
                "atmosphere": 0.44218868,
                "beautiful": 0.30036038,
                "best": 0.09538541,
                "climate": 1.3683774,
                "cold": 0.16555622,
                "continental": 0.0208058,
                "culture": 0.06171996,
                "d": 0.9625172,
                "dc": 2.7619488,
                "delaware": 0.081595294,
                "district": 1.6485157,
                "dr": 0.6397389,
                "forecast": 0.98528975,
                "geography": 0.22286704,
                "heat": 0.8082221,
                "humid": 0.9245074,
                "maryland": 0.97202545,
                "nearby": 0.03915935,
                "pollution": 0.013022495,
                "rain": 0.14599317,
                "season": 0.0049015363,
                "summer": 0.37453577,
                "te": 0.08253291,
                "temperature": 1.3041841,
                "tropical": 0.6053031,
                "usa": 0.5448736,
                "virginia": 0.91435903,
                "washington": 2.0816846,
                "weather": 2.705098,
                "winter": 0.6765218
            }
        }
    ]
}

[elasticsearchmachine]

Pinging @elastic/ml-core (Team:ML)

[jonathan-buttner]

Using a TreeSet to ensure sorting for tests

[jonathan-buttner]

The reason I'm using putDefaultIdIfAbsent is because we have some integration tests that use the same model registry to test revoking the authorization. addDefaultIds throws an IllegalStateException when you attempt to add the same default id which for tests we don't want to do.

In production this shouldn't be an issue because this will only be called when a node starts which will have a completely empty model registry anyway.

[elasticsearchmachine]

Hi @jonathan-buttner, I've created a changelog YAML for you.

[maxjakob]

The values look right 👍
For code changes I would defer to @davidkyle

[davidkyle]

LGTM

[elasticsearchmachine]

💔 Backport failed

Status	Branch	Result
❌	8.x	Commit could not be cherrypicked due to conflicts

You can use sqren/backport to manually backport by running backport --upstream elastic/elasticsearch --pr 122066

[jonathan-buttner]

💚 All backports created successfully

Status	Branch	Result
✅	8.x

Questions ?

Please refer to the Backport tool documentation