huggingface/transformers · error · ValueError

out_features and out_indices should have the same length if

Error message

out_features and out_indices should have the same length if both are set

What it means

Raised when both out_features and out_indices are set on a backbone config but have different lengths. The two attributes are two views of the same selection (feature names vs positional indices of stage_names); a length mismatch means they cannot possibly describe the same stages, so verification fails fast before any model code runs.

Source

Thrown at src/transformers/backbone_utils.py:120

            if any(idx for idx in positive_indices if idx not in range(len(self.stage_names))):
                raise ValueError(
                    f"out_indices must be valid indices for stage_names {self.stage_names}, got {self._out_indices}"
                )
            if len(positive_indices) != len(set(positive_indices)):
                msg = f"out_indices must not contain any duplicates, got {self._out_indices}"
                msg += f"(equivalent to {positive_indices}))" if positive_indices != self._out_indices else ""
                raise ValueError(msg)
            if positive_indices != tuple(sorted(positive_indices)):
                sorted_negative = [
                    idx for _, idx in sorted(zip(positive_indices, self._out_indices), key=lambda x: x[0])
                ]
                raise ValueError(
                    f"out_indices must be in the same order as stage_names, expected {sorted_negative} got {self._out_indices}"
                )

        if self._out_features is not None and self._out_indices is not None:
            if len(self._out_features) != len(self._out_indices):
                raise ValueError("out_features and out_indices should have the same length if both are set")
            if self._out_features != [self.stage_names[idx] for idx in self._out_indices]:
                raise ValueError("out_features and out_indices should correspond to the same stages if both are set")

    @property
    def out_features(self):
        return self._out_features

    @out_features.setter
    def out_features(self, out_features: list[str]):
        """
        Set the out_features attribute. This will also update the out_indices attribute to match the new out_features.
        """
        self.set_output_features_output_indices(out_features=out_features, out_indices=None)

    @property
    def out_indices(self):
        return self._out_indices

View on GitHub (pinned to a597f97485)

Solutions

Set only one of the two (out_features recommended); the setter keeps the other in sync automatically
If setting both, derive one from the other: out_indices = [stage_names.index(f) for f in out_features]
Set them together via config.set_output_features_output_indices(out_features, out_indices)

Example fix

# before
config.out_features = ['stage1', 'stage2', 'stage3']
config.out_indices = [1, 2]

# after
config.out_features = ['stage1', 'stage2', 'stage3']
config.out_indices = [1, 2, 3]  # or drop this line; the setter syncs it

Defensive patterns

Strategy: validation

Validate before calling

if config.out_features is not None and config.out_indices is not None:
    assert len(config.out_features) == len(config.out_indices), "length mismatch"

Prevention

Set only out_features; the config setter syncs out_indices automatically
When updating one attribute, update (or None out) the other in the same change
Use config.set_output_features_output_indices(f, i) to set both atomically

When it happens

Trigger: Setting config.out_features = ['stage1', 'stage2'] together with config.out_indices = [1] (or vice versa), then building the model or calling verify_out_features_out_indices(). Also fired by set_output_features_output_indices when callers pass mismatched pairs.

Common situations: Partially updating a fine-tuned config: user changes out_features but leaves stale out_indices from the base checkpoint; serialization/deserialization that drops one element; programmatic config surgery.

Related errors

AI-assisted analysis of huggingface/transformers@a597f97485 (2026-08-14). Data as JSON: /api/errors/49c78503025de750. Report an issue: GitHub.