huggingface/pytorch-image-models · error · ValueError

Invalid epsilon value: {eps}

Error message

Invalid epsilon value: {eps}

What it means

NAdamW optimizer constructor validation: epsilon must satisfy 0.0 <= eps. Negative epsilon is rejected because it would break the sqrt(v)+eps denominator computation.

Source

Thrown at timm/optim/nadamw.py:58

    """

    def __init__(
            self,
            params: ParamsT,
            lr: float = 1e-3,
            betas: Tuple[float, float] = (0.9, 0.999),
            eps: float = 1e-8,
            weight_decay: float = 1e-2,
            caution: bool = False,
            corrected_weight_decay: bool = False,
            maximize: bool = False,
            foreach: Optional[bool] = None,
            capturable: bool = False,
    ):
        if not 0.0 <= lr:
            raise ValueError(f'Invalid learning rate: {lr}')
        if not 0.0 <= eps:
            raise ValueError(f'Invalid epsilon value: {eps}')
        if not 0.0 <= betas[0] < 1.0:
            raise ValueError(f'Invalid beta parameter at index 0: {betas[0]}')
        if not 0.0 <= betas[1] < 1.0:
            raise ValueError(f'Invalid beta parameter at index 1: {betas[1]}')
        if not 0.0 <= weight_decay:
            raise ValueError(f'Invalid weight_decay value: {weight_decay}')
        defaults = dict(
            lr=lr,
            betas=betas,
            eps=eps,
            weight_decay=weight_decay,
            caution=caution,
            corrected_weight_decay=corrected_weight_decay,
            foreach=foreach,
            maximize=maximize,
            capturable=capturable,
        )
        super().__init__(params, defaults)

View on GitHub (pinned to 9a5261e31b)

Solutions

  1. Use a positive eps like 1e-8 (the default)
  2. Audit config files for sign errors on numerical stability constants

Example fix

# before
opt = NAdamW(model.parameters(), eps=-1e-8)

# after
opt = NAdamW(model.parameters(), eps=1e-8)
Defensive patterns

Strategy: validation

Validate before calling

assert eps >= 0.0

Prevention

When it happens

Trigger: Calling timm.optim.NAdamW(params, eps=-1e-8) or any negative epsilon value.

Common situations: Config typo on eps, or eps accidentally set from another parameter's value during refactoring.

Related errors

AI-assisted analysis of huggingface/pytorch-image-models@9a5261e31b (2026-08-27). Data as JSON: /api/errors/acdc7aca2272fbea. Report an issue: GitHub.