Function _normalize_row

benchmark/benchmark_chat_completion.py:208–232

(
    row: dict[str, Any],
    dataset: str,
    row_index: int,
    tokenizer=None,
)

def _normalize_row(
    row: dict[str, Any],
    dataset: str,
    row_index: int,
    tokenizer=None,
) -> BenchmarkRequest:
    request_id = str(row.get('id', f'{dataset}-{row_index}'))
    messages = _extract_messages(row)

    if tokenizer is not None:
        prompt_str = tokenizer.apply_chat_template(
            messages,
            tokenize=False,
            add_generation_prompt=True,
        )
        return BenchmarkRequest(
            dataset=dataset,
            id=request_id,
            input_ids=tokenizer.encode(prompt_str, add_special_tokens=False),
            image_data=row.get('image_data'),
        )

    if not messages:
        raise ValueError(f'row {row_index} in {dataset} has invalid messages')
    return BenchmarkRequest(dataset=dataset, id=request_id, messages=messages)


def _read_raw_rows(