Preprocessing text for NLP models often requires every token sequence to have the same length. Write a function that pads shorter sequences and truncates longer ones to a fixed max_len.

For a sequence $x$ of length $L$ and target length $M$, define $L'=\min(L, M)$ and produce an output $y$ of length $M$ by copying $L'$ tokens (from the start or end) and filling the rest with pad_value.

Requirements

Implement the function

Rules:

• Return a NumPy array of shape (batch_size, max_len).
• The input seqs is a standard Python list of lists.
• If a sequence is longer than max_len, truncate according to truncating ("pre" keeps the last max_len, "post" keeps the first max_len).
• If a sequence is shorter than max_len, pad using pad_value according to padding ("pre" pads on the left, "post" pads on the right).
• Do not modify the input seqs in-place.
• Use only NumPy and Python built-in libraries.

Example

Output: