Port of Batch Engine code

[ralph-msft]

A port of the bare minimum batch engine code from the Promptflow repo. The focus was on expediency, rather than elegance and some items are still stubs or don't do much currently. A follow-up PR will integrate this new code so that it can be invoked as part of an evaluate call.

Things that still need to be done:

• Handle errors properly for OpenAI requests
• Handle "target" functions being passed
• Capture token usage from OpenAI requests (though this does not currently work right now)
• Handle logging in a cleaner way
• Determine what tracing (if any) is needed and either implement that code or remove it entirely
• Determine if we still need the saving to a local folder structure, whether this should be changed to save to a single file (optionally?), or remove that code outright

[Copilot]

Pull Request Overview

This PR ports the minimal Batch Engine functionality from the Promptflow repository into the Azure Evaluation SDK. Key changes include implementing core components for batch execution (engine, run, submitter, storage, logging, and exceptions), integration of OpenAI API injection stubs, and updating proxy and client interfaces for batch runs.

Reviewed Changes

Copilot reviewed 17 out of 17 changed files in this pull request and generated 1 comment.

File	Description
legacy/_batch_engine/_run_submitter.py	Introduces the RunSubmitter class that submits batch runs and integrates tracing and storage calls.
_engine.py	Implements the BatchEngine that processes batch data via asynchronous tasks using semaphores and aggregates results.
_evaluate/_batch_run/init.py	Updates all exports to include RunSubmitterClient and refactors import ordering.
Other files	New and updated modules for configuration, status handling, logging, exceptions, and result aggregation to support the batch execution functionality.

Comments suppressed due to low confidence (1)

sdk/evaluation/azure-ai-evaluation/azure/ai/evaluation/legacy/_batch_engine/_engine.py:140

• [nitpick] The variable name 'input' shadows the built-in function 'input'. Consider renaming it (e.g. to 'record' or 'item') to improve clarity.

for input in data:

[azure-sdk]

API change check

APIView has identified API level changes in this PR and created following API reviews.

azure-ai-evaluation