PaddlePaddle/PaddleOCR · error · ValueError

File not found: '{file_path}'

Error message

File not found: '{file_path}'

What it means

ValueError from PaddleOCRVLLoader's lazy loading when a non-URL file_path does not exist on the local filesystem. The loader first checks _is_url; for local inputs it verifies Path(file_path).exists() before calling client.parse_document, raising with the exact path in the message.

Source

Thrown at langchain-paddleocr/langchain_paddleocr/document_loaders/paddleocr_vl.py:215

    def _process_file(self, file_path: str) -> tuple[str, dict[str, Any]]:
        """Process a single file through the SDK and return text + raw result."""
        with PaddleOCRClient(
            token=self._token,
            base_url=self._base_url,
            client_platform="langchain",
            poll_timeout=self._timeout,
        ) as client:
            parse_kwargs: dict[str, Any] = {"options": self._options}
            if self._model is not None:
                parse_kwargs["model"] = self._model
            if self._is_url(file_path):
                result = client.parse_document(file_url=file_path, **parse_kwargs)
            else:
                local_path = Path(file_path)
                if not local_path.exists():
                    msg = f"File not found: '{file_path}'"
                    raise ValueError(msg)
                result = client.parse_document(
                    file_path=str(local_path),
                    **parse_kwargs,
                )

        text_parts = [page.markdown_text for page in result.pages if page.markdown_text]
        text = _PAGES_DELIMITER.join(text_parts)

        raw_response = {
            "job_id": result.job_id,
            "data_info": result.data_info,
            "pages": [
                {
                    "markdown_text": page.markdown_text,
                    "markdown_images": page.markdown_images,
                    "output_images": page.output_images,
                    "pruned_result": page.pruned_result,
                    "input_image_url": page.input_image_url,

View on GitHub (pinned to 2661c7c0ef)

Solutions

1. Pass an absolute path: `str(Path(file_path).resolve())`.
2. Verify the file exists before constructing the load call (os.path.isfile).
3. For remote files, use an http(s) URL so _is_url routes it to parse_document(file_url=...), or download it locally first.

Example fix

// before
loader = PaddleOCRVLLoader("data/report.pdf")
docs = loader.load()  # ValueError if cwd != project root

// after
from pathlib import Path
loader = PaddleOCRVLLoader(str(Path("data/report.pdf").resolve()))
docs = loader.load()

from pathlib import Path
import urllib.parse as _u

def is_http_url(s: str) -> bool:
    p = _u.urlparse(s)
    return p.scheme in ("http", "https") and bool(p.netloc)

def loader_input_ok(file_path: str) -> bool:
    return is_http_url(file_path) or Path(file_path).is_file()

def is_loadable_source(src: object) -> bool:
    """True for http(s) URLs or existing local files."""
    if not isinstance(src, str):
        return False
    p = urlparse(src)
    if p.scheme in ("http", "https") and p.netloc:
        return True
    return Path(src).is_file()

try:
    docs = PaddleOCRVLLoader(path).load()
except ValueError as e:
    if "File not found" in str(e):
        path = str(Path(path).resolve())  # or download from remote storage first
        docs = PaddleOCRVLLoader(path).load()
    else:
        raise

When it happens

Trigger: Passing a relative path resolved from a different working directory; a typo'd or moved file; a remote-looking path (e.g. s3:// or ftp://) that fails the http(s) URL check and is then treated as local.

Common situations: Notebooks/agents running with a different cwd than expected; file already consumed/moved by a prior step; passing cloud-storage URIs unsupported by the loader.

Related errors

• Unsupported model: {model!r}
• {file_path}
• fileUrl and filePath are mutually exclusive.
• Environment Variable CUDA_VISIBLE_DEVICES is not set correct
• The input data is inconsistent with expectations.

AI-assisted analysis of PaddlePaddle/PaddleOCR@2661c7c0ef (2026-08-14). Data as JSON: /api/errors/930b2f12f7b548f0. Report an issue: GitHub.