Lightning-AI/pytorch-lightning · error · ValueError

`Trainer(strategy='deepspeed', precision={precision!r})` is

Error message

`Trainer(strategy='deepspeed', precision={precision!r})` is not supported. `precision` must be one of: {supported_precision}.

What it means

DeepSpeedPrecision was constructed with a precision string that is not part of the _PRECISION_INPUT literal union (e.g. '64-true', a typo, or an unsupported value). The plugin validates up front because DeepSpeed can only map a fixed set of precision modes to internal dtypes (bf16-mixed, fp16 variants, fp32).

Source

Thrown at src/lightning/pytorch/plugins/precision/deepspeed.py:58

class DeepSpeedPrecision(Precision):
    """Precision plugin for DeepSpeed integration.

    .. warning::  This is an :ref:`experimental <versioning:Experimental API>` feature.

    Args:
        precision: Full precision (32-true), half precision (16-true, bf16-true) or
            mixed precision (16-mixed, bf16-mixed).

    Raises:
        ValueError:
            If unsupported ``precision`` is provided.

    """

    def __init__(self, precision: _PRECISION_INPUT) -> None:
        supported_precision = get_args(_PRECISION_INPUT)
        if precision not in supported_precision:
            raise ValueError(
                f"`Trainer(strategy='deepspeed', precision={precision!r})` is not supported."
                f" `precision` must be one of: {supported_precision}."
            )
        self.precision = precision
        precision_to_type = {
            "bf16-mixed": torch.bfloat16,
            "16-mixed": torch.float16,
            "bf16-true": torch.bfloat16,
            "16-true": torch.float16,
            "32-true": torch.float32,
        }
        self._desired_dtype = precision_to_type[self.precision]

    @override
    def convert_module(self, module: Module) -> Module:
        if "true" in self.precision:
            return module.to(dtype=self._desired_dtype)
        return module

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Use one of the supported precision strings, e.g. '16-mixed', 'bf16-mixed', or '32-true'
2. Check the printed supported_precision tuple in the error and match it exactly
3. If passing a dtype, convert to the corresponding precision string

Example fix

# before
Trainer(strategy='deepspeed', precision='fp16')

# after
Trainer(strategy='deepspeed', precision='16-mixed')

from typing import get_args
from lightning.pytorch.plugins.precision.deepspeed import _PRECISION_INPUT

def valid_deepspeed_precision(p: str) -> bool:
    return p in get_args(_PRECISION_INPUT)

assert valid_deepspeed_precision(precision)

from typing import get_args
from lightning.pytorch.plugins.precision.deepspeed import _PRECISION_INPUT
SUPPORTED = set(get_args(_PRECISION_INPUT))

def is_supported_precision(p: str) -> bool:
    return p in SUPPORTED

When it happens

Trigger: DeepSpeedPrecision(precision='something-else') or Trainer(strategy='deepspeed', precision=<invalid value>); any value outside get_args(_PRECISION_INPUT) such as 'tf32' or a misspelled '16-mixed'.

Common situations: Typos in Trainer precision strings; passing raw torch dtypes (torch.float16) instead of the string literal; using precision values valid for other strategies but not registered in the union.

Related errors

• `precision={precision!r})` is not supported in DeepSpeed. `p
• `precision={precision!r})` is not supported in FSDP. `precis
• `precision={precision!r})` is not supported in XLA. `precisi
• `precision={precision!r})` is not supported in FSDP. `precis
• No precision set

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/d4f3de338d9570f4. Report an issue: GitHub.