Lightning-AI/pytorch-lightning · error · ValueError

`Passed `{type(self).__name__}(precision={precision!r})`. Pr

Error message

`Passed `{type(self).__name__}(precision={precision!r})`. Precision must be '16-mixed' or 'bf16-mixed'.

What it means

Raised by the MixedPrecisionPlugin constructor when the precision argument is anything other than the literals '16-mixed' or 'bf16-mixed'. The AMP plugin only wraps these two mixed-precision modes; full 32-bit or true 16-bit training use different precision plugin classes.

Source

Thrown at src/lightning/pytorch/plugins/precision/amp.py:66

class MixedPrecision(Precision):
    """Plugin for Automatic Mixed Precision (AMP) training with ``torch.autocast``.

    Args:
        precision: Whether to use ``torch.float16`` (``'16-mixed'``) or ``torch.bfloat16`` (``'bf16-mixed'``).
        device: The device for ``torch.autocast``.
        scaler: An optional :class:`torch.cuda.amp.GradScaler` to use.

    """

    def __init__(
        self,
        precision: Literal["16-mixed", "bf16-mixed"],
        device: str,
        scaler: Optional["torch.amp.GradScaler"] = None,
    ) -> None:
        if precision not in ("16-mixed", "bf16-mixed"):
            raise ValueError(
                f"`Passed `{type(self).__name__}(precision={precision!r})`."
                f" Precision must be '16-mixed' or 'bf16-mixed'."
            )

        self.precision = precision
        if scaler is None and self.precision == "16-mixed":
            scaler = torch.amp.GradScaler(device=device)
        if scaler is not None and self.precision == "bf16-mixed":
            raise MisconfigurationException(f"`precision='bf16-mixed'` does not use a scaler, found {scaler}.")
        self.device = device
        self.scaler = scaler

    @override
    def pre_backward(self, tensor: Tensor, module: "pl.LightningModule") -> Tensor:  # type: ignore[override]
        if self.scaler is not None:
            tensor = self.scaler.scale(tensor)
        return super().pre_backward(tensor, module)

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Use the new string literals: '16-mixed' or 'bf16-mixed'
2. Update Trainer calls: `Trainer(precision='16-mixed')` instead of `precision=16`
3. If you need a different precision scheme, use the corresponding plugin class (e.g. DoublePrecisionPlugin, Precision)

Example fix

# before (Lightning 1.x style)
trainer = pl.Trainer(precision=16)
# or
plugin = MixedPrecisionPlugin(precision=16, device='cuda')

# after
trainer = pl.Trainer(precision='16-mixed')
plugin = MixedPrecisionPlugin(precision='16-mixed', device='cuda')

VALID = ('16-mixed', 'bf16-mixed')
precision = trainer_config.get('precision', '32-true')
if precision in VALID:
    plugin = MixedPrecisionPlugin(precision=precision, device='cuda')
# else use default Precision plugin for '32-true' etc.

def is_mixed_precision_literal(p) -> bool:
    return p in ('16-mixed', 'bf16-mixed')

When it happens

Trigger: Instantiating amp plugins with legacy values like precision=16 or precision='bf16' (pre-2.0 naming); passing '32-true', 16, or 'fp16' to MixedPrecisionPlugin; configs migrated from Lightning 1.x using integer precision.

Common situations: Upgrading from Lightning 1.x where `Trainer(precision=16)` was valid; writing custom plugins that hardcode old precision strings; YAML configs with stale precision values.

Related errors

• Passed `{type(self).__name__}(precision={precision!r})`. Pre
• `precision='bf16-mixed'` does not use a scaler, found {scale
• `precision='bf16-mixed'` does not use a scaler, found {scale
• AMP and the LBFGS optimizer are not compatible.
• Gradient clipping is not implemented for optimizers handling

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/a4dc517a6e9a1c1c. Report an issue: GitHub.