Update 3.mdx

[Karthika-ps]

This change fixes a small but important technical inconsistency in Chapter 2, Section 3 regarding tensor creation from tokenized sequences.

The current example shows two encoded sequences of different lengths being directly converted into a PyTorch tensor and describes them as “already of rectangular shape.” In practice, PyTorch tensors require rectangular (uniform-length) inputs, and attempting to convert ragged sequences without padding raises a ValueError.

This update clarifies that padding is required before conversion and aligns the example with actual PyTorch behavior and the tokenizer-based workflow introduced earlier in the section. The revised explanation helps avoid confusion for readers who try to run the code verbatim and reinforces the correct mental model: tokenization can produce variable-length sequences, but padding is necessary when creating tensors or passing inputs to a model.

[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.

[Karthika-ps]

Fix line wrapping to satisfy quality check