huggingface/transformers · error · TypeError

Expected config to be a DynamoConfig or dict, got {type(conf

Error message

Expected config to be a DynamoConfig or dict, got {type(config)}

What it means

DynamoExporter.export accepts config only as a DynamoConfig instance or as a plain dict of its fields (which it converts via DynamoConfig(**config)). Any other type — an OnnxConfig/ExecutorchConfig, a string, None — raises this TypeError before any tracing starts.

Source

Thrown at src/transformers/exporters/exporter_dynamo.py:96

    >>> exported = exporter.export(model, inputs, config=DynamoConfig(dynamic=True))
    >>> outputs = exported.module()(**inputs)
    ```
    """

    required_packages = ["torch"]
    min_versions = {"torch": "2.11.0"}
    tested_versions = {"torch": "2.12.0"}

    def export(
        self,
        model: PreTrainedModel,
        sample_inputs: MutableMapping[str, Any],
        config: DynamoConfig | dict[str, Any],
    ) -> ExportedProgram:
        if isinstance(config, dict):
            config = DynamoConfig(**config)
        elif not isinstance(config, DynamoConfig):
            raise TypeError(f"Expected config to be a DynamoConfig or dict, got {type(config)}")

        model, sample_inputs, output_flags = prepare_for_export(model, sample_inputs)

        dynamic_shapes = config.dynamic_shapes
        if config.dynamic and dynamic_shapes is None:
            logger.warning_once(
                "`dynamic=True` with no explicit `dynamic_shapes` marks every input axis `Dim.AUTO`, so "
                "torch.export resolves symbolic shapes for all of them — including axes that are actually "
                "fixed (batch, a size-1 decode step, num_heads/head_dim). Passing explicit `dynamic_shapes` "
                "that mark only the axes which vary bypasses that symbolic-shape resolution and exports "
                "significantly faster."
            )
            dynamic_shapes = get_auto_dynamic_shapes(sample_inputs)

        register_cache_pytrees_for_model(model)

        with (
            apply_patches("dynamo"),

View on GitHub (pinned to a597f97485)

Solutions

1. Pass a DynamoConfig: DynamoExporter().export(model, inputs, config=DynamoConfig(dynamic=True)).
2. Or pass its fields as a dict: config={"dynamic": True, "dynamic_shapes": {...}}.
3. If dispatching by backend, build the matching config class per exporter (or use AutoHfExporter.from_config).

Example fix

# before
DynamoExporter().export(model, inputs, config=OnnxConfig())  # TypeError

# after
from transformers.exporters.exporter_dynamo import DynamoConfig
DynamoExporter().export(model, inputs, config=DynamoConfig(dynamic=True))

from transformers.exporters.exporter_dynamo import DynamoConfig

if not isinstance(config, DynamoConfig):
    config = DynamoConfig(**config)  # normalize dicts; anything else fails loudly here, not mid-export
DynamoExporter().export(model, inputs, config)

def is_dynamo_config_like(cfg) -> bool:
    from transformers.exporters.exporter_dynamo import DynamoConfig
    return isinstance(cfg, DynamoConfig) or isinstance(cfg, dict)

When it happens

Trigger: Passing OnnxConfig or ExecutorchConfig to DynamoExporter.export; passing config=None expecting defaults; passing a config dataclass from a custom backend; mixing up kwargs order so another object lands in the config slot.

Common situations: Copy-pasting an ONNX example into a dynamo export; a dispatch layer that forwards one config object to every exporter regardless of backend.

Related errors

• Expected config to be an ExecutorchConfig or dict, got {type
• Expected config to be an OnnxConfig or dict, got {type(confi
• out_indices must be a list, got {type(self._out_indices)}
• You can only update int, float, bool or string values in the
• Can only set a dictionary as `pp_plan`

AI-assisted analysis of huggingface/transformers@a597f97485 (2026-08-14). Data as JSON: /api/errors/d2f549ef0e5160b5. Report an issue: GitHub.