huggingface/pytorch-image-models · error · TypeError

Coefficients must be a preset name (str), a 3-sequence (a,b,

Error message

Coefficients must be a preset name (str), a 3-sequence (a,b,c), or a sequence of 3-sequences.

What it means

Thrown by resolve_ns_coefficients in timm's Muon optimizer when the ns_coefficients argument is neither a string preset name nor a sequence. The API accepts only a preset name, a single 3-sequence (a,b,c), or a sequence of 3-sequences.

Source

Thrown at timm/optim/muon.py:1039

    is_real = lambda x: isinstance(x, numbers.Real) and not isinstance(x, bool)

    def as_coeff(x: Sequence[float]) -> Tuple[float, float, float]:
        if not is_seq(x) or len(x) != 3 or not all(is_real(v) for v in x):
            raise ValueError(f"Coefficient must be length-3 of real numbers, got: {x!r}")
        a, b, c = x  # type: ignore[misc]
        return float(a), float(b), float(c)

    if isinstance(value, str):
        if value not in presets:
            valid = ", ".join(sorted(presets.keys()))
            raise ValueError(f"Unknown coefficients preset '{value}'. Valid options: {valid}")
        seq = presets[value]
        if not is_seq(seq) or len(seq) == 0:
            raise ValueError(f"Preset '{value}' is empty or invalid")
        return [as_coeff(item) for item in seq]  # validate & cast

    if not is_seq(value):
        raise TypeError(
            "Coefficients must be a preset name (str), a 3-sequence (a,b,c), "
            "or a sequence of 3-sequences."
        )

    # Decide single triple vs list-of-triples by structure
    if len(value) == 3 and all(is_real(v) for v in value):  # type: ignore[index]
        return [as_coeff(value)]  # single triple -> wrap

    # Otherwise treat as list/tuple of triples
    out = []
    for i, item in enumerate(value):  # type: ignore[assignment]
        if not is_seq(item):
            raise TypeError(f"Item {i} is not a sequence: {item!r}")
        out.append(as_coeff(item))
    if not out:
        raise ValueError("Coefficient list cannot be empty")
    return out

View on GitHub (pinned to 9a5261e31b)

Solutions

1. Pass a preset name string, e.g. ns_coefficients='nesterov'
2. Pass a single triple: ns_coefficients=(0.9, 0.999, 1e-8) style (a,b,c)
3. Pass a list of triples for per-tensor coefficients

Example fix

# before
Muon(params, ns_coefficients=0.9)

# after
Muon(params, ns_coefficients=(0.9, 0.95, 0.5))

from collections.abc import Sequence
assert isinstance(ns_coefficients, (str, Sequence)), 'ns_coefficients must be str preset or sequence'

def is_valid_coeffs(v) -> bool:
    return isinstance(v, str) or (isinstance(v, (list, tuple)) and not isinstance(v, str))

When it happens

Trigger: Passing a scalar (ns_coefficients=0.5), None, or any non-iterable to timm.optim.Muon/AdamMuon's ns_coefficients parameter.

Common situations: Assuming ns_coefficients takes a single float, passing an uninitialized None default, or passing a dict of coefficients.

Related errors

• Item {i} is not a sequence: {item!r}
• Preset '{value}' is empty or invalid
• Coefficient list cannot be empty
• Please provide `hook_fns` for each `hook_fn_locs`, their len
• Input image must have positive dimensions, got H={height}, W

AI-assisted analysis of huggingface/pytorch-image-models@9a5261e31b (2026-08-27). Data as JSON: /api/errors/7585f52949a43730. Report an issue: GitHub.