Lightning-AI/pytorch-lightning · error · TypeError

`gradient_clip_val` should be an int or a float. Got {gradie

Error message

`gradient_clip_val` should be an int or a float. Got {gradient_clip_val}.

What it means

clip_gradients validates that gradient_clip_val is an int or float before use; anything else (string, None after no default, tensor, etc.) raises TypeError. Note the Trainer path defaults a missing value to 0.0, so this fires only for genuinely wrong types passed explicitly.

Source

Thrown at src/lightning/pytorch/core/module.py:1284

                " Please use only one of them."
            )

        if gradient_clip_algorithm is None:
            gradient_clip_algorithm = self.trainer.gradient_clip_algorithm or "norm"
        else:
            gradient_clip_algorithm = gradient_clip_algorithm.lower()
            if (
                self.trainer.gradient_clip_algorithm is not None
                and self.trainer.gradient_clip_algorithm != gradient_clip_algorithm
            ):
                raise MisconfigurationException(
                    f"You have set `Trainer(gradient_clip_algorithm={self.trainer.gradient_clip_algorithm.value!r})`"
                    f" and have passed `clip_gradients(gradient_clip_algorithm={gradient_clip_algorithm!r})"
                    " Please use only one of them."
                )

        if not isinstance(gradient_clip_val, (int, float)):
            raise TypeError(f"`gradient_clip_val` should be an int or a float. Got {gradient_clip_val}.")

        if not GradClipAlgorithmType.supported_type(gradient_clip_algorithm.lower()):
            raise MisconfigurationException(
                f"`gradient_clip_algorithm` {gradient_clip_algorithm} is invalid."
                f" Allowed algorithms: {GradClipAlgorithmType.supported_types()}."
            )

        gradient_clip_algorithm = GradClipAlgorithmType(gradient_clip_algorithm)
        self.trainer.precision_plugin.clip_gradients(optimizer, gradient_clip_val, gradient_clip_algorithm)

    def configure_gradient_clipping(
        self,
        optimizer: Optimizer,
        gradient_clip_val: Optional[Union[int, float]] = None,
        gradient_clip_algorithm: Optional[str] = None,
    ) -> None:
        """Perform gradient clipping for the optimizer parameters. Called before :meth:`optimizer_step`.

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Cast to float: gradient_clip_val=float(cfg.clip)
2. Validate config types before training (argparse type=float)
3. Ensure the value is not None when passed explicitly

Example fix

# before
self.clip_gradients(optimizer, gradient_clip_val=cfg['clip'])  # cfg['clip'] == '1.0'

# after
self.clip_gradients(optimizer, gradient_clip_val=float(cfg['clip']))

assert isinstance(gradient_clip_val, (int, float)) and not isinstance(gradient_clip_val, bool), gradient_clip_val
self.clip_gradients(optimizer, gradient_clip_val=float(gradient_clip_val))

def valid_clip_val(v) -> bool:
    return isinstance(v, (int, float)) and not isinstance(v, bool)

When it happens

Trigger: Calling self.clip_gradients(optimizer, gradient_clip_val='1.0') or with a torch.Tensor/None from a config system that didn't cast.

Common situations: Value comes from argparse/hydra as a string ('--clip 1.0'); config parsing produced None or an object type; programmatic schedules pass a tensor.

Related errors

• f"`gradient_clip_val` should be an int or a float. Got {grad
• Device should be CUDA, got {device} instead.
• You requested to find {num_devices} devices but there are no
• `setup_optimizers` requires at least one optimizer as input.
• `setup_dataloaders` requires at least one dataloader as inpu

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/c57cbf8261dcbec3. Report an issue: GitHub.