lllyasviel/Fooocus · error · OSError

hash of {path} (url: {url}) failed to validate

Error message

hash of {path} (url: {url}) failed to validate

What it means

download_file() fetches a file over HTTP and, when a SHA-256 digest is supplied, recomputes the hash of the on-disk file and compares it to the expected value. On mismatch it raises OSError naming the path and URL. This catches both corrupted/interrupted downloads and upstream files that changed after the digest was pinned.

Source

Thrown at ldm_patched/k_diffusion/utils.py:47

    return expanded.detach().clone() if expanded.device.type == 'mps' else expanded


def n_params(module):
    """Returns the number of trainable parameters in a module."""
    return sum(p.numel() for p in module.parameters())


def download_file(path, url, digest=None):
    """Downloads a file if it does not exist, optionally checking its SHA-256 hash."""
    path = Path(path)
    path.parent.mkdir(parents=True, exist_ok=True)
    if not path.exists():
        with urllib.request.urlopen(url) as response, open(path, 'wb') as f:
            shutil.copyfileobj(response, f)
    if digest is not None:
        file_digest = hashlib.sha256(open(path, 'rb').read()).hexdigest()
        if digest != file_digest:
            raise OSError(f'hash of {path} (url: {url}) failed to validate')
    return path


@contextmanager
def train_mode(model, mode=True):
    """A context manager that places a model into training mode and restores
    the previous mode on exit."""
    modes = [module.training for module in model.modules()]
    try:
        yield model.train(mode)
    finally:
        for i, module in enumerate(model.modules()):
            module.training = modes[i]


def eval_mode(model):
    """A context manager that places a model into evaluation mode and restores
    the previous mode on exit."""

View on GitHub (pinned to ae05379cc9)

Solutions

1. Delete the cached file at `path` and retry so it is downloaded fresh.
2. Verify the remote file actually changed (recompute sha256 of a fresh copy); if it legitimately changed, update the pinned `digest`.
3. Check free disk space / network stability if the file is truncated (size much smaller than expected).
4. If the URL is dead or redirected, update the url argument along with the digest.

Example fix

# before
download_file('ckpt/vgg16.pth', URL, digest='oldhash...')  # OSError: hash failed

# after (re-download then re-pin)
!rm ckpt/vgg16.pth
download_file('ckpt/vgg16.pth', URL)  # inspect, then:
digest = hashlib.sha256(open('ckpt/vgg16.pth','rb').read()).hexdigest()
download_file('ckpt/vgg16.pth', URL, digest=digest)

def sha256_of(path):
    h = hashlib.sha256()
    with open(path, 'rb') as f:
        for chunk in iter(lambda: f.read(1 << 20), b''):
            h.update(chunk)
    return h.hexdigest()

if path.exists() and digest is not None and sha256_of(path) != digest:
    path.unlink()  # force clean re-download

for attempt in range(3):
    try:
        p = download_file(path, url, digest=digest)
        break
    except OSError as e:
        if 'failed to validate' not in str(e):
            raise
        Path(path).unlink(missing_ok=True)  # corrupt cache; retry fresh
else:
    raise RuntimeError(f'hash validation failed after retries: {path}')

When it happens

Trigger: A partial download caused by a dropped connection (the file exists so it is not re-fetched, then fails hashing); a previously downloaded file corrupted on disk; the remote asset at `url` was replaced so the pinned digest no longer matches.

Common situations: Pretrained-weight bootstrap in k-diffusion training scripts; shared caches where another process wrote to the same path; mirrors/CDNs serving different bytes; disk-full truncation during first download.

Related errors

• Order {order} too high for step {i}
• eta must be 0 for reverse sampling
• order should be 2 or 3
• sigma_min and sigma_max must not be 0
• solver_type must be 'heun' or 'midpoint'

AI-assisted analysis of lllyasviel/Fooocus@ae05379cc9 (2026-08-15). Data as JSON: /api/errors/935fa91a18f02899. Report an issue: GitHub.