Skip to content

Validators

Annotated Validators

API Documentation

pydantic.functional_validators.WrapValidator
pydantic.functional_validators.PlainValidator
pydantic.functional_validators.BeforeValidator
pydantic.functional_validators.AfterValidator

Pydantic provides a way to apply validators via use of Annotated. You should use this whenever you want to bind validation to a type instead of model or field.

from typing import Any, List from typing_extensions import Annotated from pydantic import BaseModel, ValidationError from pydantic.functional_validators import AfterValidator def check_squares(v: int) -> int: assert v**0.5 % 1 == 0, f'{v} is not a square number' return v def double(v: Any) -> Any: return v * 2 MyNumber = Annotated[int, AfterValidator(double), AfterValidator(check_squares)] class DemoModel(BaseModel): number: List[MyNumber] print(DemoModel(number=[2, 8])) #> number=[4, 16] try: DemoModel(number=[2, 4]) except ValidationError as e: print(e)  """  1 validation error for DemoModel  number.1  Assertion failed, 8 is not a square number  assert ((8 ** 0.5) % 1) == 0 [type=assertion_error, input_value=4, input_type=int]  """

In this example we used some type aliases (MyNumber = Annotated[...]). While this can help with legibility of the code, it is not required, you can use Annotated directly in a model field type hint. These type aliases are also not actual types but you can use a similar approach with TypeAliasType to create actual types. See Custom Types for a more detailed explanation of custom types.

It is also worth noting that you can nest Annotated inside other types. In this example we used that to apply validation to the inner items of a list. The same approach can be used for dict keys, etc.

Before, After, Wrap and Plain validators

Pydantic provides multiple types of validator functions:

  • After validators run after Pydantic's internal parsing. They are generally more type safe and thus easier to implement.
  • Before validators run before Pydantic's internal parsing and validation (e.g. coercion of a str to an int). These are more flexible than After validators since they can modify the raw input, but they also have to deal with the raw input, which in theory could be any arbitrary object.
  • Plain validators are like a mode='before' validator but they terminate validation immediately, no further validators are called and Pydantic does not do any of its internal validation.
  • Wrap validators are the most flexible of all. You can run code before or after Pydantic and other validators do their thing or you can terminate validation immediately, both with a successful value or an error.

You can use multiple before, after, or mode='wrap' validators, but only one PlainValidator since a plain validator will not call any inner validators.

Here's an example of a mode='wrap' validator:

import json from typing import Any, List from typing_extensions import Annotated from pydantic import ( BaseModel, ValidationError, ValidationInfo, ValidatorFunctionWrapHandler, ) from pydantic.functional_validators import WrapValidator def maybe_strip_whitespace( v: Any, handler: ValidatorFunctionWrapHandler, info: ValidationInfo ) -> int: if info.mode == 'json': assert isinstance(v, str), 'In JSON mode the input must be a string!' # you can call the handler multiple times try: return handler(v) except ValidationError: return handler(v.strip()) assert info.mode == 'python' assert isinstance(v, int), 'In Python mode the input must be an int!' # do no further validation return v MyNumber = Annotated[int, WrapValidator(maybe_strip_whitespace)] class DemoModel(BaseModel): number: List[MyNumber] print(DemoModel(number=[2, 8])) #> number=[2, 8] print(DemoModel.model_validate_json(json.dumps({'number': [' 2 ', '8']}))) #> number=[2, 8] try: DemoModel(number=['2']) except ValidationError as e: print(e)  """  1 validation error for DemoModel  number.0  Assertion failed, In Python mode the input must be an int!  assert False  + where False = isinstance('2', int) [type=assertion_error, input_value='2', input_type=str]  """

The same "modes" apply to @field_validator, which is discussed in the next section.

Ordering of validators within Annotated

Order of validation metadata within Annotated matters. Validation goes from right to left and back. That is, it goes from right to left running all "before" validators (or calling into "wrap" validators), then left to right back out calling all "after" validators.

from typing import Any, Callable, List, cast from typing_extensions import Annotated, TypedDict from pydantic import ( AfterValidator, BaseModel, BeforeValidator, PlainValidator, ValidationInfo, ValidatorFunctionWrapHandler, WrapValidator, ) from pydantic.functional_validators import field_validator class Context(TypedDict): logs: List[str] def make_validator(label: str) -> Callable[[Any, ValidationInfo], Any]: def validator(v: Any, info: ValidationInfo) -> Any: context = cast(Context, info.context) context['logs'].append(label) return v return validator def make_wrap_validator( label: str, ) -> Callable[[Any, ValidatorFunctionWrapHandler, ValidationInfo], Any]: def validator( v: Any, handler: ValidatorFunctionWrapHandler, info: ValidationInfo ) -> Any: context = cast(Context, info.context) context['logs'].append(f'{label}: pre') result = handler(v) context['logs'].append(f'{label}: post') return result return validator class A(BaseModel): x: Annotated[ str, BeforeValidator(make_validator('before-1')), AfterValidator(make_validator('after-1')), WrapValidator(make_wrap_validator('wrap-1')), BeforeValidator(make_validator('before-2')), AfterValidator(make_validator('after-2')), WrapValidator(make_wrap_validator('wrap-2')), BeforeValidator(make_validator('before-3')), AfterValidator(make_validator('after-3')), WrapValidator(make_wrap_validator('wrap-3')), BeforeValidator(make_validator('before-4')), AfterValidator(make_validator('after-4')), WrapValidator(make_wrap_validator('wrap-4')), ] y: Annotated[ str, BeforeValidator(make_validator('before-1')), AfterValidator(make_validator('after-1')), WrapValidator(make_wrap_validator('wrap-1')), BeforeValidator(make_validator('before-2')), AfterValidator(make_validator('after-2')), WrapValidator(make_wrap_validator('wrap-2')), PlainValidator(make_validator('plain')), BeforeValidator(make_validator('before-3')), AfterValidator(make_validator('after-3')), WrapValidator(make_wrap_validator('wrap-3')), BeforeValidator(make_validator('before-4')), AfterValidator(make_validator('after-4')), WrapValidator(make_wrap_validator('wrap-4')), ] val_x_before = field_validator('x', mode='before')( make_validator('val_x before') ) val_x_after = field_validator('x', mode='after')( make_validator('val_x after') ) val_y_wrap = field_validator('y', mode='wrap')( make_wrap_validator('val_y wrap') ) context = Context(logs=[]) A.model_validate({'x': 'abc', 'y': 'def'}, context=context) print(context['logs']) """ [  'val_x before',  'wrap-4: pre',  'before-4',  'wrap-3: pre',  'before-3',  'wrap-2: pre',  'before-2',  'wrap-1: pre',  'before-1',  'after-1',  'wrap-1: post',  'after-2',  'wrap-2: post',  'after-3',  'wrap-3: post',  'after-4',  'wrap-4: post',  'val_x after',  'val_y wrap: pre',  'wrap-4: pre',  'before-4',  'wrap-3: pre',  'before-3',  'plain',  'after-3',  'wrap-3: post',  'after-4',  'wrap-4: post',  'val_y wrap: post', ] """

Validation of default values

Validators won't run when the default value is used. This applies both to @field_validator validators and Annotated validators. You can force them to run with Field(validate_default=True). Setting validate_default to True has the closest behavior to using always=True in validator in Pydantic v1. However, you are generally better off using a @model_validator(mode='before') where the function is called before the inner validator is called.

from typing_extensions import Annotated from pydantic import BaseModel, Field, field_validator class Model(BaseModel): x: str = 'abc' y: Annotated[str, Field(validate_default=True)] = 'xyz' @field_validator('x', 'y') @classmethod def double(cls, v: str) -> str: return v * 2 print(Model()) #> x='abc' y='xyzxyz' print(Model(x='foo')) #> x='foofoo' y='xyzxyz' print(Model(x='abc')) #> x='abcabc' y='xyzxyz' print(Model(x='foo', y='bar')) #> x='foofoo' y='barbar'

Field validators

API Documentation

pydantic.functional_validators.field_validator

If you want to attach a validator to a specific field of a model you can use the @field_validator decorator.

from pydantic import ( BaseModel, ValidationError, ValidationInfo, field_validator, ) class UserModel(BaseModel): name: str id: int @field_validator('name') @classmethod def name_must_contain_space(cls, v: str) -> str: if ' ' not in v: raise ValueError('must contain a space') return v.title() # you can select multiple fields, or use '*' to select all fields @field_validator('id', 'name') @classmethod def check_alphanumeric(cls, v: str, info: ValidationInfo) -> str: if isinstance(v, str): # info.field_name is the name of the field being validated is_alphanumeric = v.replace(' ', '').isalnum() assert is_alphanumeric, f'{info.field_name} must be alphanumeric' return v print(UserModel(name='John Doe', id=1)) #> name='John Doe' id=1 try: UserModel(name='samuel', id=1) except ValidationError as e: print(e)  """  1 validation error for UserModel  name  Value error, must contain a space [type=value_error, input_value='samuel', input_type=str]  """ try: UserModel(name='John Doe', id='abc') except ValidationError as e: print(e)  """  1 validation error for UserModel  id  Input should be a valid integer, unable to parse string as an integer [type=int_parsing, input_value='abc', input_type=str]  """ try: UserModel(name='John Doe!', id=1) except ValidationError as e: print(e)  """  1 validation error for UserModel  name  Assertion failed, name must be alphanumeric  assert False [type=assertion_error, input_value='John Doe!', input_type=str]  """

A few things to note on validators:

  • @field_validators are "class methods", so the first argument value they receive is the UserModel class, not an instance of UserModel. We recommend you use the @classmethod decorator on them below the @field_validator decorator to get proper type checking.
  • the second argument is the field value to validate; it can be named as you please
  • the third argument, if present, is an instance of pydantic.ValidationInfo
  • validators should either return the parsed value or raise a ValueError or AssertionError (assert statements may be used).
  • A single validator can be applied to multiple fields by passing it multiple field names.
  • A single validator can also be called on all fields by passing the special value '*'.

Warning

If you make use of assert statements, keep in mind that running Python with the -O optimization flag disables assert statements, and validators will stop working.

Note

FieldValidationInfo is deprecated in 2.4, use ValidationInfo instead.

If you want to access values from another field inside a @field_validator, this may be possible using ValidationInfo.data, which is a dict of field name to field value. Validation is done in the order fields are defined, so you have to be careful when using ValidationInfo.data to not access a field that has not yet been validated/populated — in the code above, for example, you would not be able to access info.data['id'] from within name_must_contain_space. However, in most cases where you want to perform validation using multiple field values, it is better to use @model_validator which is discussed in the section below.

Model validators

API Documentation

pydantic.functional_validators.model_validator

Validation can also be performed on the entire model's data using @model_validator.

from typing import Any from typing_extensions import Self from pydantic import BaseModel, ValidationError, model_validator class UserModel(BaseModel): username: str password1: str password2: str @model_validator(mode='before') @classmethod def check_card_number_omitted(cls, data: Any) -> Any: if isinstance(data, dict): assert ( 'card_number' not in data ), 'card_number should not be included' return data @model_validator(mode='after') def check_passwords_match(self) -> Self: pw1 = self.password1 pw2 = self.password2 if pw1 is not None and pw2 is not None and pw1 != pw2: raise ValueError('passwords do not match') return self print(UserModel(username='scolvin', password1='zxcvbn', password2='zxcvbn')) #> username='scolvin' password1='zxcvbn' password2='zxcvbn' try: UserModel(username='scolvin', password1='zxcvbn', password2='zxcvbn2') except ValidationError as e: print(e)  """  1 validation error for UserModel  Value error, passwords do not match [type=value_error, input_value={'username': 'scolvin', '... 'password2': 'zxcvbn2'}, input_type=dict]  """ try: UserModel( username='scolvin', password1='zxcvbn', password2='zxcvbn', card_number='1234', ) except ValidationError as e: print(e)  """  1 validation error for UserModel  Assertion failed, card_number should not be included  assert 'card_number' not in {'card_number': '1234', 'password1': 'zxcvbn', 'password2': 'zxcvbn', 'username': 'scolvin'} [type=assertion_error, input_value={'username': 'scolvin', '..., 'card_number': '1234'}, input_type=dict]  """

On return type checking

Methods decorated with @model_validator should return the self instance at the end of the method. For type checking purposes, you can use Self from either typing or the typing_extensions backport as the return type of the decorated method. In the context of the above example, you could also use def check_passwords_match(self: 'UserModel') -> 'UserModel' to indicate that the method returns an instance of the model.

On not returning self

If you fail to return self at the end of a @model_validator method (either, returning None or returning something other than self), you may encounter unexpected behavior.

Specifically, for nested models, if you return None (or equivalently, don't include a return statement), despite a potentially successful validation, the nested model will be None in the parent model.

Returning a value other than self causes unexpected behavior at the top level of validation when validating via __init__. In order to avoid this, we recommend one of: 1. Simply mutate and return self at the end of the method. 2. If you must return a value other than self, use a method like model_validate where you can directly fetch the return value.

Here's an example of the unexpected behavior, and the warning you'll receive:

from pydantic import BaseModel from pydantic.functional_validators import model_validator class Child(BaseModel): name: str @model_validator(mode='after') # type: ignore def validate_model(self) -> 'Child': return Child.model_construct(name='different!') print(repr(Child(name='foo'))) """ UserWarning: A custom validator is returning a value other than `self`. Returning anything other than `self` from a top level model validator isn't supported when validating via `__init__`. See the `model_validator` docs (https://docs.pydantic.dev/latest/concepts/validators/#model-validators) for more details. Child(name='foo') """

On Inheritance

A @model_validator defined in a base class will be called during the validation of a subclass instance.

Overriding a @model_validator in a subclass will override the base class' @model_validator, and thus only the subclass' version of said @model_validator will be called.

Model validators can be mode='before', mode='after' or mode='wrap'.

Before model validators are passed the raw input which is often a dict[str, Any] but could also be an instance of the model itself (e.g. if UserModel.model_validate(UserModel.construct(...)) is called) or anything else since you can pass arbitrary objects into model_validate. Because of this mode='before' validators are extremely flexible and powerful but can be cumbersome and error prone to implement. Before model validators should be class methods. The first argument should be cls (and we also recommend you use @classmethod below @model_validator for proper type checking), the second argument will be the input (you should generally type it as Any and use isinstance to narrow the type) and the third argument (if present) will be a pydantic.ValidationInfo.

mode='after' validators are instance methods and always receive an instance of the model as the first argument. Be sure to return the instance at the end of your validator. You should not use (cls, ModelType) as the signature, instead just use (self) and let type checkers infer the type of self for you. Since these are fully type safe they are often easier to implement than mode='before' validators. If any field fails to validate, mode='after' validators for that field will not be called.

Handling errors in validators

As mentioned in the previous sections you can raise either a ValueError or AssertionError (including ones generated by assert ... statements) within a validator to indicate validation failed. You can also raise a PydanticCustomError which is a bit more verbose but gives you extra flexibility. Any other errors (including TypeError) are bubbled up and not wrapped in a ValidationError.

from pydantic_core import PydanticCustomError from pydantic import BaseModel, ValidationError, field_validator class Model(BaseModel): x: int @field_validator('x') @classmethod def validate_x(cls, v: int) -> int: if v % 42 == 0: raise PydanticCustomError( 'the_answer_error', '{number} is the answer!', {'number': v}, ) return v try: Model(x=42 * 2) except ValidationError as e: print(e)  """  1 validation error for Model  x  84 is the answer! [type=the_answer_error, input_value=84, input_type=int]  """

Special Types

Pydantic provides a few special types that can be used to customize validation.

  • InstanceOf is a type that can be used to validate that a value is an instance of a given class.
from typing import List from pydantic import BaseModel, InstanceOf, ValidationError class Fruit: def __repr__(self): return self.__class__.__name__ class Banana(Fruit): ... class Apple(Fruit): ... class Basket(BaseModel): fruits: List[InstanceOf[Fruit]] print(Basket(fruits=[Banana(), Apple()])) #> fruits=[Banana, Apple] try: Basket(fruits=[Banana(), 'Apple']) except ValidationError as e: print(e)  """  1 validation error for Basket  fruits.1  Input should be an instance of Fruit [type=is_instance_of, input_value='Apple', input_type=str]  """
  • SkipValidation is a type that can be used to skip validation on a field.
from typing import List from pydantic import BaseModel, SkipValidation class Model(BaseModel): names: List[SkipValidation[str]] m = Model(names=['foo', 'bar']) print(m) #> names=['foo', 'bar'] m = Model(names=['foo', 123]) # (1)! print(m) #> names=['foo', 123]
  1. Note that the validation of the second item is skipped. If it has the wrong type it will emit a warning during serialization.

Field checks

During class creation, validators are checked to confirm that the fields they specify actually exist on the model.

This may be undesirable if, for example, you want to define a validator to validate fields that will only be present on subclasses of the model where the validator is defined.

If you want to disable these checks during class creation, you can pass check_fields=False as a keyword argument to the validator.

Dataclass validators

Validators also work with Pydantic dataclasses.

from pydantic import field_validator from pydantic.dataclasses import dataclass @dataclass class DemoDataclass: product_id: str # should be a five-digit string, may have leading zeros @field_validator('product_id', mode='before') @classmethod def convert_int_serial(cls, v): if isinstance(v, int): v = str(v).zfill(5) return v print(DemoDataclass(product_id='01234')) #> DemoDataclass(product_id='01234') print(DemoDataclass(product_id=2468)) #> DemoDataclass(product_id='02468')

Validation Context

You can pass a context object to the validation methods which can be accessed from the info argument to decorated validator functions:

from pydantic import BaseModel, ValidationInfo, field_validator class Model(BaseModel): text: str @field_validator('text') @classmethod def remove_stopwords(cls, v: str, info: ValidationInfo): context = info.context if context: stopwords = context.get('stopwords', set()) v = ' '.join(w for w in v.split() if w.lower() not in stopwords) return v data = {'text': 'This is an example document'} print(Model.model_validate(data)) # no context #> text='This is an example document' print(Model.model_validate(data, context={'stopwords': ['this', 'is', 'an']})) #> text='example document' print(Model.model_validate(data, context={'stopwords': ['document']})) #> text='This is an example'

This is useful when you need to dynamically update the validation behavior during runtime. For example, if you wanted a field to have a dynamically controllable set of allowed values, this could be done by passing the allowed values by context, and having a separate mechanism for updating what is allowed:

from typing import Any, Dict, List from pydantic import ( BaseModel, ValidationError, ValidationInfo, field_validator, ) _allowed_choices = ['a', 'b', 'c'] def set_allowed_choices(allowed_choices: List[str]) -> None: global _allowed_choices _allowed_choices = allowed_choices def get_context() -> Dict[str, Any]: return {'allowed_choices': _allowed_choices} class Model(BaseModel): choice: str @field_validator('choice') @classmethod def validate_choice(cls, v: str, info: ValidationInfo): allowed_choices = info.context.get('allowed_choices') if allowed_choices and v not in allowed_choices: raise ValueError(f'choice must be one of {allowed_choices}') return v print(Model.model_validate({'choice': 'a'}, context=get_context())) #> choice='a' try: print(Model.model_validate({'choice': 'd'}, context=get_context())) except ValidationError as exc: print(exc)  """  1 validation error for Model  choice  Value error, choice must be one of ['a', 'b', 'c'] [type=value_error, input_value='d', input_type=str]  """ set_allowed_choices(['b', 'c']) try: print(Model.model_validate({'choice': 'a'}, context=get_context())) except ValidationError as exc: print(exc)  """  1 validation error for Model  choice  Value error, choice must be one of ['b', 'c'] [type=value_error, input_value='a', input_type=str]  """

Similarly, you can use a context for serialization.

Using validation context with BaseModel initialization

Although there is no way to specify a context in the standard BaseModel initializer, you can work around this through the use of contextvars.ContextVar and a custom __init__ method:

from contextlib import contextmanager from contextvars import ContextVar from typing import Any, Dict, Iterator from pydantic import BaseModel, ValidationInfo, field_validator _init_context_var = ContextVar('_init_context_var', default=None) @contextmanager def init_context(value: Dict[str, Any]) -> Iterator[None]: token = _init_context_var.set(value) try: yield finally: _init_context_var.reset(token) class Model(BaseModel): my_number: int def __init__(self, /, **data: Any) -> None: self.__pydantic_validator__.validate_python( data, self_instance=self, context=_init_context_var.get(), ) @field_validator('my_number') @classmethod def multiply_with_context(cls, value: int, info: ValidationInfo) -> int: if info.context: multiplier = info.context.get('multiplier', 1) value = value * multiplier return value print(Model(my_number=2)) #> my_number=2 with init_context({'multiplier': 3}): print(Model(my_number=2)) #> my_number=6 print(Model(my_number=2)) #> my_number=2

Reusing Validators

Occasionally, you will want to use the same validator on multiple fields/models (e.g. to normalize some input data). The "naive" approach would be to write a separate function, then call it from multiple decorators. Obviously, this entails a lot of repetition and boiler plate code. The following approach demonstrates how you can reuse a validator so that redundancy is minimized and the models become again almost declarative.

from pydantic import BaseModel, field_validator def normalize(name: str) -> str: return ' '.join((word.capitalize()) for word in name.split(' ')) class Producer(BaseModel): name: str _normalize_name = field_validator('name')(normalize) class Consumer(BaseModel): name: str _normalize_name = field_validator('name')(normalize) jane_doe = Producer(name='JaNe DOE') print(repr(jane_doe)) #> Producer(name='Jane Doe') john_doe = Consumer(name='joHN dOe') print(repr(john_doe)) #> Consumer(name='John Doe')