huggingface/transformers · error · ValueError

key {k} isn't in the original config dict

Error message

key {k} isn't in the original config dict

What it means

ValueError from PreTrainedConfig.update_from_string: the CLI-style update string contains a key that does not already exist as an attribute on the config object. update_from_str only mutates existing attributes; it never creates new ones.

Source

Thrown at src/transformers/configuration_utils.py:1187

    def update_from_string(self, update_str: str):
        """
        Updates attributes of this class with attributes from `update_str`.

        The expected format is ints, floats and strings as is, and for booleans use `true` or `false`. For example:
        "n_embd=10,resid_pdrop=0.2,scale_attn_weights=false,summary_type=cls_index"

        The keys to change have to already exist in the config object.

        Args:
            update_str (`str`): String with attributes that should be updated for this class.

        """

        d = dict(x.split("=") for x in update_str.split(","))
        for k, v in d.items():
            if not hasattr(self, k):
                raise ValueError(f"key {k} isn't in the original config dict")

            old_v = getattr(self, k)
            if isinstance(old_v, bool):
                if v.lower() in ["true", "1", "y", "yes"]:
                    v = True
                elif v.lower() in ["false", "0", "n", "no"]:
                    v = False
                else:
                    raise ValueError(f"can't derive true or false from {v} (key {k})")
            elif isinstance(old_v, int):
                v = int(v)
            elif isinstance(old_v, float):
                v = float(v)
            elif not isinstance(old_v, str):
                raise TypeError(
                    f"You can only update int, float, bool or string values in the config, got {v} for key {k}"
                )

View on GitHub (pinned to a597f97485)

Solutions

1. Check hasattr(config, key) for the failing key and fix the typo or use the correct attribute name.
2. For nested configs, target the sub-config: config.text_config.update_from_string(...).
3. If you truly need a new attribute, set it directly (config.my_key = value) instead of via update_from_string.

Example fix

// before
config.update_from_string("n_embdd=10")  # ValueError

// after
config.update_from_string("n_embd=10")

updates = {"n_embd": "10", "scale_attn_weights": "false"}
missing = [k for k in updates if not hasattr(config, k)]
if missing:
    raise KeyError(f"Unknown config keys: {missing}")
config.update_from_string(",".join(f"{k}={v}" for k, v in updates.items()))

def can_update(config, key: str) -> bool:
    return hasattr(config, key)

try:
    config.update_from_string(s)
except ValueError as e:
    if "isn't in the original config dict" in str(e):
        raise KeyError(f"bad update string {s!r}: {e}")

When it happens

Trigger: config.update_from_string("n_embdd=10") (typo), or updating a key that belongs to a different model family's config (e.g. num_attention_heads on a config class that does not define it, or a parameter only present on the nested text_config).

Common situations: Scripts that build the update string dynamically from a dict of hyperparameters; copy-pasting example strings from a different model's docs; trying to set sub-config attributes at the top level of a composite (vision-language) config.

Related errors

• 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_
• out_indices must be in the same order as stage_names, expect
• out_features and out_indices should have the same length if

AI-assisted analysis of huggingface/transformers@a597f97485 (2026-08-14). Data as JSON: /api/errors/fe613b8f27c8b616. Report an issue: GitHub.