langchain-ai/langchain · error · ValueError

Got input_variables={input_variables}, but based on prefix/s

Error message

Got input_variables={input_variables}, but based on prefix/suffix expected {expected_input_variables}

What it means

Pydantic after-validator on FewShotPromptWithTemplates (only when validate_template=True) raised when the declared `input_variables` do not cover all variables referenced by the suffix (and prefix if set) after accounting for `partial_variables`. missing_vars are the variables the sub-templates need that input_variables lacks.

Source

Thrown at libs/core/langchain_core/prompts/few_shot_with_templates.py:95

        return values

    @model_validator(mode="after")
    def template_is_valid(self) -> Self:
        """Check that prefix, suffix, and input variables are consistent."""
        if self.validate_template:
            input_variables = self.input_variables
            expected_input_variables = set(self.suffix.input_variables)
            expected_input_variables |= set(self.partial_variables)
            if self.prefix is not None:
                expected_input_variables |= set(self.prefix.input_variables)
            missing_vars = expected_input_variables.difference(input_variables)
            if missing_vars:
                msg = (
                    f"Got input_variables={input_variables}, but based on "
                    f"prefix/suffix expected {expected_input_variables}"
                )
                raise ValueError(msg)
        else:
            self.input_variables = sorted(
                set(self.suffix.input_variables)
                | set(self.prefix.input_variables if self.prefix else [])
                - set(self.partial_variables)
            )
        return self

    model_config = ConfigDict(
        arbitrary_types_allowed=True,
        extra="forbid",
    )

    def _get_examples(self, **kwargs: Any) -> list[dict[str, Any]]:
        if self.examples is not None:
            return self.examples
        if self.example_selector is not None:
            return self.example_selector.select_examples(kwargs)

View on GitHub (pinned to e32fa9a52e)

Solutions

Add the missing variable(s) to input_variables.
Or add the variable to partial_variables if you intend to fill it once: partial_variables={"style": "concise"}.
With validate_template=False the class auto-computes input_variables from suffix/prefix, so dropping the manual list also avoids drift.

Example fix

# before
FewShotPromptWithTemplates(
    ...,
    input_variables=["question"],
    suffix=PromptTemplate(template="{question} {style}", input_variables=["question"]),
    validate_template=True,
)

# after
FewShotPromptWithTemplates(
    ...,
    input_variables=["question", "style"],
    suffix=PromptTemplate(template="{question} {style}", input_variables=["question", "style"]),
    validate_template=True,
)

Defensive patterns

Strategy: validation

Validate before calling

def check_input_vars(suffix, prefix, input_variables, partial_variables):
    expected = set(suffix.input_variables) | set(partial_variables)
    if prefix is not None:
        expected |= set(prefix.input_variables)
    return expected <= set(input_variables)

Prevention

Prefer validate_template=False and let input_variables be auto-derived from the templates.
Regenerate input_variables from template.get_variables() after every template text edit.
Cover template construction with a unit test that formats with all declared inputs.

When it happens

Trigger: Constructing with validate_template=True, suffix=PromptTemplate(template="{question} {style}"), but input_variables=["question"] (missing 'style') and no partial_variables covering 'style'. The prefix's variables are unioned in too.

Common situations: Hand-maintained input_variables lists drifting from edited template strings; adding a variable to the suffix text during iteration without updating the list; setting partial_variables after copy() so the validator's accounting no longer matches.

Related errors

AI-assisted analysis of langchain-ai/langchain@e32fa9a52e (2026-08-14). Data as JSON: /api/errors/9ff22fb66fa0d4ee. Report an issue: GitHub.