Consider the following yaml config file.

# config.yaml

key1:
  value1: x
level1:
  level2_1:
    key1: y
    level3:
      key1: z
      key2: zz
  level2_2:
    key1: a
    level3: null

I am looking for a pydantic model that loads the configuration from the config.yaml file. I am especially interested in using the dot-notation for all levels of this nested configuration. For example, I would like to use the settings as follows (assuming I assign the model to a variable called settings):

settings.key1.value1 # x
settings.level1.level2_1.key1 # y
settings.level1.level2_2.level3.key2 # zz

What I did so far:

from pathlib import Path

import yaml
from pydantic import BaseModel, ConfigDict


class Key1Config(BaseModel):
    value1: str


class Level3Config(BaseModel):
    model_config = ConfigDict(extra="allow")

    key1: str


class Level2Config(BaseModel):
    key1: str
    level3: Level3Config | None


class Config(BaseModel):
    key1: Key1Config
    level1: dict[str, Level2Config]

    @classmethod
    def from_yaml(cls, config_file: Path):
        with config_file.open() as f:
            data = yaml.safe_load(f)
        return cls.model_validate(data)


settings = Config.from_yaml(Path(__file__).parent.joinpath("./config.yaml"))

I am struggling in defining level1, because now it is a simply dict. I would like it to be another model so that I can access every nested level via dot-notation.

For example, my current solution requires me to do something like settings.level1["level2_1"].level3.key2.

But what I want is settings.level1.level2_1.level3.key2.

Further information:

level2_* can be any arbitrary number, i.e. level2_3, level2_4, or any arbitrary_name.

level3 can also have an arbitrary number of lower level entries OR null.