[data] fix KeyError on invalid role tag in ranking dataset converters

[he-yufeng]

What does this PR do?

OpenAIDatasetConverter and SharegptDatasetConverter handle a pairwise (ranking) example whose chosen/rejected role tag is not an accepted tag by logging Invalid role tag and setting broken_data = True. But the if broken_data: skip guard is evaluated at the top of the if/elif chain, so flipping the flag inside the ranking branch is too late — the code then indexes tag_mapping[chosen[role_tag]] and raises KeyError, which crashes align_dataset() (dataset.map) instead of skipping the bad example as intended.

This skips such examples (prompt, response = []) — exactly what the top guard does for the cases it catches — so a misconfigured/edge ranking example is dropped with a warning rather than crashing data preparation. Both converters share the identical branch and are fixed together.

Before / after

# invalid role in a ranking example
example = {"conversations": [{"from": "human", "value": "..."}],
           "chosen": {"from": "not_a_role", "value": "..."},
           "rejected": {"from": "gpt", "value": "..."}}
# before: KeyError: 'not_a_role'   (crashes align_dataset)
# after:  {"_prompt": [], "_response": [], ...}   (skipped)

Verification

PYTHONPATH=src pytest tests/data/test_converter.py — added test_sharegpt_converter_ranking_skips_invalid_role; it fails with KeyError on the old code and passes after this change (3 passed total). ruff check and ruff format --check clean.

[gemini-code-assist]

Code Review

This pull request updates the dataset converter to safely handle invalid role tags in pairwise (ranking) datasets by skipping them (setting prompt and response to empty lists) instead of setting a flag and continuing, which previously caused crashes. A corresponding unit test has been added to verify this behavior. The reviewer pointed out a potential issue in OpenAIDatasetConverter where raw role tags are compared directly against standardized roles in accept_tags[-1], which will fail for custom role tags, and suggested using tag_mapping.get() to resolve this.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

In OpenAIDatasetConverter, accept_tags contains standardized roles (Role.ASSISTANT.value, Role.FUNCTION.value), whereas chosen and rejected contain raw role tags from the dataset (e.g., "gpt"). Comparing the raw role tags directly against accept_tags[-1] will always fail if custom role tags are used, causing valid examples to be incorrectly skipped as having an invalid role tag.

To fix this, we should look up the raw role tags in tag_mapping using .get() before comparing them to accept_tags[-1].

            if (
                tag_mapping.get(chosen[self.dataset_attr.role_tag]) not in accept_tags[-1]
                or tag_mapping.get(rejected[self.dataset_attr.role_tag]) not in accept_tags[-1]
            ):