huggingface/transformers · error · ValueError

Per-component `config` dict is missing entries for: {sorted(

Error message

Per-component `config` dict is missing entries for: {sorted(missing)}. Expected one entry per component: {sorted(components)}.

What it means

HfExporter.export_for_generation decomposes the model into named components (e.g. prefill/decode stages) via decompose_for_generation. When you pass config as a dict, it must contain exactly one entry per component name; this ValueError fires when components exist that have no key in your dict, and the message lists both the missing names and the full expected key set.

Source

Thrown at src/transformers/exporters/base.py:187

                stays dynamic under a dynamic-shape export (`config.dynamic=True`).

        Returns:
            `dict[str, Any]`: `{component_name: backend_specific_artifact}` — same keys as
            [`~exporters.utils.decompose_for_generation`]. Values are whatever
            [`~HfExporter.export`] returns for the concrete backend (`ExportedProgram`,
            `ONNXProgram`, `ExecutorchProgramManager`).
        """
        components = decompose_for_generation(
            model,
            sample_inputs,
            generation_config=generation_config,
            multi_token_decode=multi_token_decode,
        )

        if isinstance(config, dict):
            missing = set(components) - set(config)
            if missing:
                raise ValueError(
                    f"Per-component `config` dict is missing entries for: {sorted(missing)}. "
                    f"Expected one entry per component: {sorted(components)}."
                )
            configs = config
        else:
            configs = dict.fromkeys(components, config)

        exported: dict[str, object] = {}
        for name, (submodel, subinputs) in components.items():
            try:
                exported[name] = self.export(submodel, subinputs, config=configs[name])
            except Exception as e:
                raise RuntimeError(
                    f"{type(self).__name__}.export failed on component '{name}' "
                    f"(submodel={type(submodel).__name__}, input keys={list(subinputs)})."
                ) from e

        return exported

View on GitHub (pinned to a597f97485)

Solutions

1. Add the missing keys listed in the message — the expected key set is printed verbatim.
2. If all stages should share one config, pass the config object itself (not a dict); it is broadcast to every component.
3. Log the components once (inspect the message of a deliberate run, or decompose_for_generation) and template your dict from that.

Example fix

# before
exporter.export_for_generation(model, inputs, config={"prefill": onnx_cfg})  # missing 'decode'

# after (per-component)
exporter.export_for_generation(model, inputs, config={"prefill": onnx_cfg, "decode": onnx_cfg})

# after (shared config — simplest)
exporter.export_for_generation(model, inputs, config=onnx_cfg)

# Components are model/config dependent; safest pre-check is a dry-run or sharing one config.
# If you must pass a dict, share the same config object for all stages instead:
exporter.export_for_generation(model, inputs, config=single_cfg)  # broadcast, cannot hit this error

try:
    out = exporter.export_for_generation(model, inputs, config=configs)
except ValueError as e:
    if "missing entries for" in str(e):
        missing = ast.literal_eval(str(e).split("missing entries for:")[1].split(".")[0])
        configs.update({name: default_cfg for name in missing})
        out = exporter.export_for_generation(model, inputs, config=configs)
    else:
        raise

When it happens

Trigger: Calling export_for_generation(model, inputs, config={"prefill": cfg}) when decomposition produces {"prefill", "decode"}; passing stage names that don't match (e.g. 'forward' vs 'prefill'); multi_token_decode=True yielding more components than your dict covers.

Common situations: Upgrading transformers where the component set for a model changed (new decode stages); enabling multi-token decode and reusing an old two-entry config dict; hand-writing per-stage configs from an outdated example.

Related errors

• export_config_dict must contain key 'export_format' set to e
• Unknown exporter type, got {name} - supported exporters are:
• out_indices must be a list, got {type(self._out_indices)}
• out_indices must be valid indices for stage_names {self.stag
• out_indices must not contain any duplicates, got {self._out_

AI-assisted analysis of huggingface/transformers@a597f97485 (2026-08-14). Data as JSON: /api/errors/4f5696a76c63dfa2. Report an issue: GitHub.