[Inference] Implementing the completion task type on EIS.

[afoucret]

Draft PR to add support for the completion task type for Elastic Inference Service.

[elasticsearchmachine]

Hi @afoucret, I've created a changelog YAML for you.

[Copilot]

Pull Request Overview

This PR implements support for the completion task type in the Elastic Inference Service (EIS). The implementation adapts simple text inputs from the completion task into chat message format and routes them to the existing EIS chat endpoint. Key architectural changes include introducing a distinction between COMPLETION and CHAT_COMPLETION task types, where completion uses a simpler text-based input format while chat completion uses structured message objects.

• Introduces ElasticInferenceServiceCompletionModel for the completion task type
• Refactors the original completion model to ElasticInferenceServiceChatCompletionModel to distinguish between task types
• Adds request adapters to convert simple text inputs into chat message format

Reviewed Changes

Copilot reviewed 16 out of 16 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
ElasticInferenceServiceCompletionRequestTests.java	Tests for the completion request adapter that converts text inputs to chat messages
ElasticInferenceServiceCompletionRequestEntityTests.java	Tests for the XContent serialization of completion request entities
ElasticInferenceServiceCompletionModelTests.java	Tests for the completion model including URI creation and service settings
ElasticInferenceServiceChatCompletionModelTests.java	Tests for the chat completion model (renamed from original completion model)
ElasticInferenceServiceAuthorizationHandlerTests.java	Updates test references to use ElasticInferenceServiceChatCompletionModel
ElasticInferenceServiceTests.java	Updates references and adds COMPLETION to supported task types
ElasticInferenceServiceUnifiedChatCompletionRequest.java	Updates to use ElasticInferenceServiceChatCompletionModel
ElasticInferenceServiceCompletionRequestEntity.java	Entity class for serializing completion requests to JSON
ElasticInferenceServiceCompletionRequest.java	Request adapter for COMPLETION task type
ElasticInferenceServiceCompletionModel.java	Model class for COMPLETION task type that adapts inputs to chat format
ElasticInferenceServiceChatCompletionModel.java	Renamed model class for CHAT_COMPLETION task type
ElasticInferenceServiceActionVisitor.java	Adds visitor method for completion model
ElasticInferenceServiceActionCreator.java	Implements action creator for completion requests
ElasticInferenceServiceUnifiedCompletionRequestManager.java	Updates to use ElasticInferenceServiceChatCompletionModel
ElasticInferenceService.java	Adds COMPLETION to supported task types and creates model instances
137677.yaml	Changelog entry for the enhancement

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[jonathan-buttner]

Thanks for the PR! I left a few suggestions.

[jonathan-buttner]

If I remember correctly, this is the only method that makes this class different from the ElasticInferenceServiceCompletionModel. How about we just use ElasticInferenceServiceCompletionModel and add this method there?

[jonathan-buttner]

Instead of creating a new class for this, let's follow the OpenAI example here: https://github.com/elastic/elasticsearch/blob/main/x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/openai/action/OpenAiActionCreator.java#L75-L86

Basically we should be able to leverage the existing ElasticInferenceServiceUnifiedChatCompletionRequest and just construct a request with only a list of input.

[jonathan-buttner]

Once we switch to using a single model class we should be able to follow what OpenAI does here: and have case CHAT_COMPLETION, COMPLETION -> ...

[afoucret]

@jonathan-buttner: thank you for the careful review. I made the requested change and it looks better now.