def after_model_validations()

in src/hyperpod_nemo_adapter/conf/config_schemas.py [0:0]

• 13 lines of code
• 6 McCabe index (conditional complexity)

    def after_model_validations(self) -> "BaseTrainerConfig":
        if "LOCAL_WORLD_SIZE" in os.environ and "WORLD_SIZE" in os.environ:
            # read from torchrun environment variables
            actual_devices = int(os.environ["LOCAL_WORLD_SIZE"])
            actual_num_nodes = int(os.environ["WORLD_SIZE"]) // actual_devices
            if isinstance(self.devices, int) and self.devices != actual_devices:
                raise ValueError(
                    f"'devices' ({self.devices}) does not equal actual number of devices ({actual_devices})"
                )
            if self.num_nodes != actual_num_nodes:
                raise ValueError(
                    f"'num_nodes' ({self.num_nodes}) does not equal actual number of nodes ({actual_num_nodes})"
                )

        return self