Add user-provided entity types to prompt tuning

[YingzuoLiu]

Description

Adds support for user-provided entity types during prompt tuning.

Prompt tuning previously supported automatic entity type discovery via --discover-entity-types, but there was no direct way for users to seed the prompt tuning flow with entity types they already care about. This PR adds an --entity-types CLI option and passes those values through the prompt tuning API.

When entity discovery is enabled, user-provided entity types are merged with discovered entity types while preserving user-provided order and deduplicating repeated values. When discovery is disabled, the provided entity types are used directly.

Related Issues

Fixes #1802

Proposed Changes

• Add an entity_types parameter to generate_indexing_prompts().
• Add --entity-types to the prompt-tune CLI.
• Parse comma-separated entity types from the CLI.
• Normalize and deduplicate user-provided and generated entity types.
• Add unit tests for parsing, normalization, and merge behavior.
• Add a semversioner minor change entry.

Validation

Ran locally:

uv run ruff check \
  packages/graphrag/graphrag/api/prompt_tune.py \
  packages/graphrag/graphrag/cli/prompt_tune.py \
  packages/graphrag/graphrag/cli/main.py \
  tests/unit/graphrag/prompt_tune/test_entity_types.py \
  tests/unit/graphrag/prompt_tune/__init__.py

uv run pytest tests/unit/graphrag/prompt_tune/test_entity_types.py

Result:

All checks passed
6 passed

Checklist

• I have tested these changes locally.
• I have reviewed the code changes.
• I have updated the documentation if necessary. CLI help text was updated for the new option.
• I have added appropriate unit tests.

[YingzuoLiu]

@microsoft-github-policy-service agree