docs(zh-TW): fix tokenizer input type in chapter3/2.mdx

[jieyao-MilestoneHub]

Description

Fix tokenizer input in the MRPC preprocessing example.
In recent versions, raw_datasets["train"]["sentence1"] and ["sentence2"] are datasets.arrow_dataset.Column objects, which must be converted to Python lists before tokenization.

---

Change

# Before
tokenized_dataset = tokenizer(
    raw_datasets["train"]["sentence1"],
    raw_datasets["train"]["sentence2"],
    padding=True,
    truncation=True,
)

# After
tokenized_dataset = tokenizer(
    list(raw_datasets["train"]["sentence1"]),
    list(raw_datasets["train"]["sentence2"]),
    padding=True,
    truncation=True,
)
# or
tokenized_dataset = tokenizer(
    raw_datasets["train"].data["sentence1"].to_pylist(),
    raw_datasets["train"].data["sentence2"].to_pylist(),
    padding=True,
    truncation=True,
)

---

Impact

Prevents

ValueError: text input must be of type `str`, `list[str]`, or `list[list[str]]`

and ensures compatibility with transformers ≥4.56 and datasets ≥3.0.

[jieyao-MilestoneHub]

Hi @thliang01 — I noticed this tokenizer input issue in the zh-TW course file.
Could you please help review and confirm if the fix looks correct? Thank you! 🙏

[HuggingFaceDocBuilderDev]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.