Stability-AI/generative-models · error · ValueError

Cannot find {}

Error message

Cannot find {}

What it means

After passing the mutual-exclusion check, main.py verifies that the path given to -r/--resume actually exists on disk. If os.path.exists(opt.resume) is False it raises this ValueError before attempting to load anything, so a typo or moved checkpoint fails fast instead of deep inside checkpoint loading.

Source

Thrown at main.py:561

    # running as `python main.py`
    # (in particular `main.DataModuleFromConfig`)
    sys.path.append(os.getcwd())

    parser = get_parser()

    opt, unknown = parser.parse_known_args()

    if opt.name and opt.resume:
        raise ValueError(
            "-n/--name and -r/--resume cannot be specified both."
            "If you want to resume training in a new log folder, "
            "use -n/--name in combination with --resume_from_checkpoint"
        )
    melk_ckpt_name = None
    name = None
    if opt.resume:
        if not os.path.exists(opt.resume):
            raise ValueError("Cannot find {}".format(opt.resume))
        if os.path.isfile(opt.resume):
            paths = opt.resume.split("/")
            # idx = len(paths)-paths[::-1].index("logs")+1
            # logdir = "/".join(paths[:idx])
            logdir = "/".join(paths[:-2])
            ckpt = opt.resume
            _, melk_ckpt_name = get_checkpoint_name(logdir)
        else:
            assert os.path.isdir(opt.resume), opt.resume
            logdir = opt.resume.rstrip("/")
            ckpt, melk_ckpt_name = get_checkpoint_name(logdir)

        print("#" * 100)
        print(f'Resuming from checkpoint "{ckpt}"')
        print("#" * 100)

        opt.resume_from_checkpoint = ckpt
        base_configs = sorted(glob.glob(os.path.join(logdir, "configs/*.yaml")))

View on GitHub (pinned to e8cd657656)

Solutions

1. Run `ls <path>` to verify the checkpoint path exists and correct the -r/--resume argument.
2. Use an absolute path for --resume, or launch from the same working directory used originally.
3. In containers/cluster jobs, mount or copy the directory containing the checkpoint and point --resume at the mounted path.

Example fix

// before
python main.py --base cfg.yaml -r logs/run1/checkpoints/last.ckpt   # run1 was renamed
// after
python main.py --base cfg.yaml -r logs/run1_renamed/checkpoints/last.ckpt

import os, sys
resume = next((sys.argv[i+1] for i, a in enumerate(sys.argv) if a in ('-r', '--resume')), None)
if resume and not os.path.exists(resume):
    raise SystemExit(f"Checkpoint not found: {os.path.abspath(resume)}")

When it happens

Trigger: Running with `-r /path/to/ckpt` where that path does not exist: checkpoint deleted or moved, wrong relative path from the launch working directory, or a container/mount that does not include the logs directory.

Common situations: Resuming after logs were archived or renamed; running from a different working directory so a relative path no longer resolves; submitting a SLURM/K8s job without mounting the volume holding the checkpoint; trailing whitespace or shell variable left empty in the path.

Understand the failure class

Background: "File not found" and ENOENT errors: why libraries can't find a file that should exist — this error's family across 50 libraries.

Related errors

• -n/--name and -r/--resume cannot be specified both.If you wa

AI-assisted analysis of Stability-AI/generative-models@e8cd657656 (2026-08-29). Data as JSON: /api/errors/9e86205af7451485. Report an issue: GitHub.