Lightning-AI/pytorch-lightning · error · RuntimeError

Missing folder: {os.path.dirname(tags_csv)}.

Error message

Missing folder: {os.path.dirname(tags_csv)}.

What it means

RuntimeError from save_hparams_to_tags_csv when the directory portion of the tags_csv path does not exist on the target filesystem (checked via fsspec). The writer refuses to create the file if its parent folder is missing.

Source

Thrown at src/lightning/pytorch/core/saving.py:284

    >>> vars(hparams) == hparams_new
    True
    >>> os.remove(path_csv)

    """
    fs = get_filesystem(tags_csv)
    if not fs.exists(tags_csv):
        rank_zero_warn(f"Missing Tags: {tags_csv}.", category=RuntimeWarning)
        return {}

    with fs.open(tags_csv, "r", newline="") as fp:
        csv_reader = csv.reader(fp, delimiter=",")
        return {row[0]: convert(row[1]) for row in list(csv_reader)[1:]}


def save_hparams_to_tags_csv(tags_csv: _PATH, hparams: Union[dict, Namespace]) -> None:
    fs = get_filesystem(tags_csv)
    if not _is_dir(fs, os.path.dirname(tags_csv)):
        raise RuntimeError(f"Missing folder: {os.path.dirname(tags_csv)}.")

    if isinstance(hparams, Namespace):
        hparams = vars(hparams)

    with fs.open(tags_csv, "w", newline="") as fp:
        fieldnames = ["key", "value"]
        writer = csv.DictWriter(fp, fieldnames=fieldnames)
        writer.writerow({"key": "key", "value": "value"})
        for k, v in hparams.items():
            writer.writerow({"key": k, "value": v})


def load_hparams_from_yaml(config_yaml: _PATH, use_omegaconf: bool = True) -> dict[str, Any]:
    """Load hparams from a file.

        Args:
            config_yaml: Path to config yaml file
            use_omegaconf: If omegaconf is available and ``use_omegaconf=True``,

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Create the directory first: os.makedirs(os.path.dirname(path), exist_ok=True)
2. Fix a typo'd or missing parent path in the argument
3. For remote filesystems, verify the bucket/prefix exists and credentials are valid

Example fix

// before
save_hparams_to_tags_csv("logs/exp1/tags.csv", hparams)  # logs/exp1 missing
// after
os.makedirs("logs/exp1", exist_ok=True)
save_hparams_to_tags_csv("logs/exp1/tags.csv", hparams)

import os
os.makedirs(os.path.dirname(tags_csv) or ".", exist_ok=True)

When it happens

Trigger: Calling save_hparams_to_tags_csv('newdir/tags.csv', hparams) when 'newdir' doesn't exist; also triggered indirectly by trainer save / checkpointing flows that write a tags.csv into a not-yet-created folder.

Common situations: Custom checkpoint callbacks writing to nested paths, or fsspec remote paths (s3://, gs://) where the directory check fails due to permissions or a wrong bucket/prefix.

Related errors

• Missing folder: {os.path.dirname(config_yaml)}.
• .csv, .yml or .yaml is required for `hparams_file`
• hparams must be dictionary
• Error while merging hparams: the keys {inconsistent_keys} ar
• Received multiple values for {', '.join(duplicated_plugin_ke

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/e19fb560b7ec24b9. Report an issue: GitHub.