Lightning-AI/pytorch-lightning · error · MisconfigurationException

The `avg_fn` should be callable.

Error message

The `avg_fn` should be callable.

What it means

StochasticWeightAveraging accepts an optional custom averaging function avg_fn (passed to torch's SWA utils), which must be callable. Passing anything non-callable (a lambda result, a string, None-check passes) raises this MisconfigurationException in __init__.

Source

Thrown at src/lightning/pytorch/callbacks/stochastic_weight_avg.py:113

                When None is provided, it will infer the `device` from ``pl_module``.
                (default: ``"cpu"``)

        """

        err_msg = "swa_epoch_start should be a >0 integer or a float between 0 and 1."
        if isinstance(swa_epoch_start, int) and swa_epoch_start < 1:
            raise MisconfigurationException(err_msg)
        if isinstance(swa_epoch_start, float) and not (0 <= swa_epoch_start <= 1):
            raise MisconfigurationException(err_msg)

        wrong_type = not isinstance(swa_lrs, (float, list))
        wrong_float = isinstance(swa_lrs, float) and swa_lrs <= 0
        wrong_list = isinstance(swa_lrs, list) and not all(lr > 0 and isinstance(lr, float) for lr in swa_lrs)
        if wrong_type or wrong_float or wrong_list:
            raise MisconfigurationException("The `swa_lrs` should a positive float, or a list of positive floats")

        if avg_fn is not None and not callable(avg_fn):
            raise MisconfigurationException("The `avg_fn` should be callable.")

        if device is not None and not isinstance(device, (torch.device, str)):
            raise MisconfigurationException(f"device is expected to be a torch.device or a str. Found {device}")

        self.n_averaged: Optional[Tensor] = None
        self._swa_epoch_start = swa_epoch_start
        self._swa_lrs = swa_lrs
        self._annealing_epochs = annealing_epochs
        self._annealing_strategy = annealing_strategy
        self._avg_fn = avg_fn or self.avg_fn
        self._device = device
        self._model_contains_batch_norm: Optional[bool] = None
        self._average_model: Optional[pl.LightningModule] = None
        self._initialized = False
        self._swa_scheduler: Optional[LRScheduler] = None
        self._scheduler_state: Optional[dict] = None
        self._init_n_averaged = 0
        self._latest_update_epoch = -1

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Pass a callable with signature avg_fn(averaged_model_parameter, model_parameter, num_averaged) -> tensor, e.g. torch.optim.swa_utils.get_ema_multi_avg_fn(0.9)
2. Or omit avg_fn to use the default equal averaging

Example fix

# before
swa = SWA(avg_fn=0.9)
# after
from torch.optim.swa_utils import get_ema_multi_avg_fn
swa = SWA(avg_fn=get_ema_multi_avg_fn(0.9))

import torch
avg_fn = None if cfg.ema is None else __import__('torch.optim.swa_utils', fromlist=['x']).get_ema_multi_avg_fn(cfg.ema)
assert avg_fn is None or callable(avg_fn)
swa = SWA(avg_fn=avg_fn)

def is_valid_avg_fn(fn) -> bool:
    return fn is None or callable(fn)

When it happens

Trigger: SWA(avg_fn=0.5) or avg_fn=some_object that isn't a function/callable class instance.

Common situations: Confusing avg_fn parameters with a float weight; passing an unpickled/serialized function reference that lost callability.

Related errors

• swa_epoch_start should be a >0 integer or a float between 0
• device is expected to be a torch.device or a str. Found {dev
• The `swa_lrs` should a positive float, or a list of positive
• SWA currently works with 1 `optimizer`.
• SWA currently not supported for more than 1 `lr_scheduler`.

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/1d5a92f7240eb98f. Report an issue: GitHub.