zylon-ai/private-gpt · error · ValueError

Chunk size must be greater than 0.

Error message

Chunk size must be greater than 0.

What it means

Raised by MarkdownHelper.sanitize_markdown when the effective chunk_size is below 1. Note the coercion just above: chunk_size = chunk_size or len(markdown), so None/0 fall back to the document length (safe for non-empty input); the only way to reach the error is passing a negative chunk_size explicitly. The value matters because the markdown is split into line batches of chunk_size for parallel sanitizing.

Source

Thrown at private_gpt/components/markdown/markdown_helper.py:160

        index, lines = data

        processed = []
        for line in lines:
            processed.append(MarkdownHelper._safe_sanity_data(line))

        return index, processed

    @staticmethod
    def sanitize_markdown(
        markdown: str, chunk_size: int | None = 256, max_workers: int | None = None
    ) -> str:
        """Sanitize markdown content to fix common formatting issues."""
        if not markdown:
            return markdown

        chunk_size = chunk_size or len(markdown)
        if chunk_size < 1:
            raise ValueError("Chunk size must be greater than 0.")

        max_workers = max_workers or 1
        max_workers = min(max_workers, len(markdown) // chunk_size)
        max_workers = max(1, max_workers)

        # Split into chunks while trying to preserve line boundaries
        lines = markdown.splitlines(keepends=True)
        batches: list[list[str]] = list(iter_batch(lines, chunk_size))

        # Create pool of workers
        processed_chunks: list[tuple[int, list[str]]] = []
        if max_workers == 1:
            for index, batch in enumerate(batches):
                _, processed = MarkdownHelper._sanity_markdown_batches((index, batch))
                processed_chunks.append((index, processed))
        else:
            with ThreadPoolExecutor(max_workers=max_workers) as executor:
                for index, batch in executor.map(

View on GitHub (pinned to 4a030776a3)

Solutions

  1. Pass a positive chunk_size (typical 256) or None to let it default to the full document in one chunk.
  2. Clamp computed values: chunk_size = max(1, computed).
  3. Validate the chunk-size setting at load time (must be a positive int).

Example fix

# before
out = MarkdownHelper.sanitize_markdown(md, chunk_size=len(md) // num_workers)  # can be negative

# after
out = MarkdownHelper.sanitize_markdown(md, chunk_size=max(1, len(md) // num_workers))
Defensive patterns

Strategy: validation

Validate before calling

chunk_size = max(1, chunk_size) if chunk_size is not None else None
MarkdownHelper.sanitize_markdown(md, chunk_size=chunk_size)

Type guard

def is_valid_chunk_size(n: Any) -> bool:
    return n is None or (isinstance(n, int) and n >= 1)

Try / catch

try:
    out = MarkdownHelper.sanitize_markdown(md, chunk_size=cs)
except ValueError:
    out = MarkdownHelper.sanitize_markdown(md)  # default single chunk

Prevention

When it happens

Trigger: Calling sanitize_markdown(markdown, chunk_size=-5); computing chunk_size dynamically (e.g. len(text) // workers) and passing a negative result; configuration where a negative chunk size is read from settings/UI.

Common situations: Derived chunk sizes from division that can go negative; user-supplied settings validated only for presence, not sign; copy-paste of a size constant with the wrong sign.

Related errors

AI-assisted analysis of zylon-ai/private-gpt@4a030776a3 (2026-08-15). Data as JSON: /api/errors/5d5f14e022316d68. Report an issue: GitHub.