hiyouga/LlamaFactory · error · ValueError

Can not load dataset from {filepath}.

Error message

Can not load dataset from {filepath}.

What it means

The local data loader only handles two cases: an existing directory or an existing file. If the path is neither (nonexistent, a glob pattern, or a URL), it raises this ValueError. Note _get_builder_name may raise first for directories whose first file has a bad extension.

Source

Thrown at src/llamafactory/v1/plugins/data_plugins/loader.py:61

        Literal["arrow", "csv", "json", "parquet", "text"]: Dataset builder name.
    """
    filetype = os.path.splitext(path)[-1][1:]
    if filetype in ["arrow", "csv", "json", "jsonl", "parquet", "txt"]:
        return filetype.replace("jsonl", "json").replace("txt", "text")
    else:
        raise ValueError(f"Unknown dataset filetype: {filetype}.")


@DataLoaderPlugin("local").register()
def load_data_from_file(filepath: str, split: str, streaming: bool) -> HFDataset:
    if os.path.isdir(filepath):
        filetype = _get_builder_name(os.listdir(filepath)[0])
        dataset = load_dataset(filetype, data_dir=filepath, split=split)
    elif os.path.isfile(filepath):
        filetype = _get_builder_name(filepath)
        dataset = load_dataset(filetype, data_files=filepath, split=split)
    else:
        raise ValueError(f"Can not load dataset from {filepath}.")

    if streaming:  # faster when data is streamed from local files
        dataset = dataset.to_iterable_dataset()

    return dataset


def adjust_data_index(
    data_index: list[tuple[str, int]], size: int | None, weight: float | None
) -> list[tuple[str, int]]:
    """Adjust dataset index by size and weight.

    Args:
        data_index (list[tuple[str, int]]): List of (dataset_name, sample_index).
        size (Optional[int]): Desired dataset size.
        weight (Optional[float]): Desired dataset weight.

    Returns:

View on GitHub (pinned to f28afaf635)

Solutions

1. Check the path exists: correct typos, use absolute paths, or fix the working directory.
2. Expand globs yourself and list concrete files in the config.
3. For HuggingFace hub datasets, register them properly (dataset_info.json or the hub loader) instead of the local file loader.
4. Verify with `ls <path>` from the same shell that launches training.

Example fix

# before
dataset_dir: ./datset/train.json   # typo

# after
dataset_dir: ./dataset/train.json  # verified with ls

import os
assert os.path.isdir(dataset_dir) or os.path.isfile(dataset_dir), f'path not found: {dataset_dir}'

def is_loadable_path(p: str) -> bool:
    """True when p is an existing file or directory."""
    return os.path.exists(p)

try:
    ds = load_data_from_file(p, split, streaming)
except ValueError as e:
    if 'Can not load dataset' in str(e):
        raise SystemExit(f'check path: {p!r} (cwd={os.getcwd()})') from None
    raise

When it happens

Trigger: Passing a nonexistent path in dataset_dir; passing a glob like 'data/*.json'; passing an HF hub repo id (e.g. 'alpaca') where the local loader plugin is selected instead of the hub loader.

Common situations: Typo in the dataset path; using a relative path while the process cwd differs; expecting local-loader semantics for hub datasets because the dataset_info entry was misconfigured.

Related errors

• File {local_path} not found.
• The length of packed example should be identical to the cuto
• Streaming mode should have an integer val size.
• `max_samples` is incompatible with `streaming`.
• `mask_history` is incompatible with `train_on_prompt`.

AI-assisted analysis of hiyouga/LlamaFactory@f28afaf635 (2026-08-14). Data as JSON: /api/errors/d3e7c0a3ffe38c06. Report an issue: GitHub.