Lightning-AI/pytorch-lightning · error · MisconfigurationException

Was unable to infer precision type, received {self.precision

Error message

Was unable to infer precision type, received {self.precision!r}.

What it means

The FSDPMixedPrecisionPlugin.mixed_precision_config property maps the precision string to FSDP MixedPrecision dtypes for 16-mixed/16-true/bf16-true/bf16-mixed/32-true; any other value that survived __init__ validation (or was set directly on the attribute afterwards) reaches the else branch and raises this MisconfigurationException.

Source

Thrown at src/lightning/pytorch/plugins/precision/fsdp.py:111

    @property
    def mixed_precision_config(self) -> "TorchMixedPrecision":
        from torch.distributed.fsdp.fully_sharded_data_parallel import MixedPrecision as TorchMixedPrecision

        if self.precision in ("16-true", "bf16-true"):
            rank_zero_warn(
                f"FSDP with `{self.precision}` enables computation in lower precision. "
                "FSDP will always retain a full-precision copy of the model parameters for sharding."
            )

        if self.precision in ("16-true", "16-mixed"):
            param_dtype = reduce_dtype = buffer_dtype = torch.float16
        elif self.precision in ("bf16-true", "bf16-mixed"):
            param_dtype = reduce_dtype = buffer_dtype = torch.bfloat16
        elif self.precision == "32-true":
            param_dtype = reduce_dtype = buffer_dtype = torch.float32
        else:
            raise MisconfigurationException(f"Was unable to infer precision type, received {self.precision!r}.")

        return TorchMixedPrecision(
            param_dtype=param_dtype,
            reduce_dtype=reduce_dtype,
            buffer_dtype=buffer_dtype,
        )

    @override
    def tensor_init_context(self) -> AbstractContextManager:
        return _DtypeContextManager(self._desired_input_dtype)

    @override
    def module_init_context(self) -> AbstractContextManager:
        return _DtypeContextManager(self._desired_input_dtype)

    @override
    def forward_context(self) -> AbstractContextManager:
        if "mixed" in self.precision:

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Only use the supported precision literals and set them at construction time, not by mutating the attribute
2. If you subclass, override mixed_precision_config (or the dtype map) to cover your custom precision
3. Upgrade lightning so the validated union matches the mapping

Example fix

# before
plugin = FSDPMixedPrecisionPlugin(precision='16-mixed')
plugin.precision = 'tf32'  # later mutation -> raises in mixed_precision_config

# after
plugin = FSDPMixedPrecisionPlugin(precision='32-true')  # choose a supported value up front

SUPPORTED = {'16-mixed','16-true','bf16-mixed','bf16-true','32-true'}

def make_fsdp_plugin(precision):
    assert precision in SUPPORTED, f'unsupported FSDP precision {precision!r}'
    return FSDPMixedPrecisionPlugin(precision=precision)

SUPPORTED = {'16-mixed','16-true','bf16-mixed','bf16-true','32-true'}

def has_dtype_mapping(precision: str) -> bool:
    return precision in SUPPORTED

When it happens

Trigger: Setting plugin.precision directly to an unmapped value after construction (bypassing __init__ validation), subclassing FSDPMixedPrecisionPlugin without extending the dtype map, or a precision alias not covered by the if/elif chain in a fork/older version.

Common situations: Monkey-patching or mutating plugin.precision in experiments; custom precision plugins inheriting from the FSDP plugin without overriding mixed_precision_config; version drift where the union and the mapping get out of sync.

Related errors

• `precision={precision!r})` is not supported in FSDP. `precis
• `precision={precision!r})` is not supported in FSDP. `precis
• `precision={precision!r}` does not use a scaler, found {scal
• The optimizer has references to the model's meta-device para
• `precision={precision!r})` is not supported in DeepSpeed. `p

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/f8db2aece56fc017. Report an issue: GitHub.