[ai-projects] Use typed EvaluatorGenerationJob in rubric samples

[aprilk-ms]

The four rubric-evaluator-generation samples were passing flat dicts to create_generation_job. The service contract nests the job inputs under an inputs field on EvaluatorGenerationJob; the SDK was tolerating the flat form, but the rolling-out service change requires the nested form.

Changes (all under sdk/ai/azure-ai-projects/samples/evaluations/):

• sample_rubric_evaluator_generation_basic.py — convert single-prompt call site.
• sample_rubric_evaluator_generation_iterate.py — convert single-prompt call site.
• sample_rubric_evaluator_generation_lifecycle.py — convert shared job_body (used for create + idempotent replay).
• sample_rubric_evaluator_generation_all_sources.py — convert multi_job (prompt + agent + dataset) and traces_job (traces + agent companion). For the traces source, switch from int(time.time()) arithmetic to datetime + timedelta (the SDK model serializes start_time / end_time as format="unix-timestamp").

All samples now use the typed models from azure.ai.projects.models:

• EvaluatorGenerationJob(inputs=EvaluatorGenerationInputs(model=..., evaluator_name=..., evaluator_display_name=..., evaluator_description=..., sources=[...]))
• Sources: PromptEvaluatorGenerationJobSource, AgentEvaluatorGenerationJobSource, DatasetEvaluatorGenerationJobSource, TracesEvaluatorGenerationJobSource (discriminator is auto-set in __init__).

The stale top-level name field (which had no home in either EvaluatorGenerationJob or EvaluatorGenerationInputs — the SDK was silently dropping it) is removed.

Verified

• AST parses for all four files.
• Repo black config (eng/black-pyproject.toml) reports no changes needed.

Not yet verified

• Live run — the service contract change is still rolling out. Will live-test once the service requires the nested inputs form.

[Copilot]

Pull request overview

Updates the azure-ai-projects rubric evaluator generation samples to align with the service’s EvaluatorGenerationJob contract by nesting job parameters under inputs and using the typed model classes throughout, preventing failures as the service rolls out stricter request validation.

Changes:

• Updated rubric evaluator generation samples to pass EvaluatorGenerationJob(inputs=EvaluatorGenerationInputs(...)) instead of flat dict payloads.
• Converted all source payloads to typed *EvaluatorGenerationJobSource models (Prompt/Agent/Dataset/Traces), including switching the traces window calculation to datetime/timedelta.
• Documented the sample update in sdk/ai/azure-ai-projects/CHANGELOG.md.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
sdk/ai/azure-ai-projects/samples/evaluations/sample_rubric_evaluator_generation_basic.py	Uses typed EvaluatorGenerationJob / EvaluatorGenerationInputs and typed Prompt source for a single-prompt generation flow.
sdk/ai/azure-ai-projects/samples/evaluations/sample_rubric_evaluator_generation_iterate.py	Uses typed job + inputs + Prompt source for the initial generation step in the iteration workflow.
sdk/ai/azure-ai-projects/samples/evaluations/sample_rubric_evaluator_generation_lifecycle.py	Replaces the shared dict “job body” with a reusable typed EvaluatorGenerationJob request object for create + idempotent replay.
sdk/ai/azure-ai-projects/samples/evaluations/sample_rubric_evaluator_generation_all_sources.py	Uses typed sources for prompt/agent/dataset/traces and computes trace window with datetime/timedelta for unix-timestamp serialization.
sdk/ai/azure-ai-projects/CHANGELOG.md	Notes the rubric evaluator generation sample updates and the inputs nesting change.