"""Configuration for Accentik's parallel Russian vowel-ranking model."""
from copy import deepcopy

from transformers import PreTrainedConfig


class AccentikConfig(PreTrainedConfig):
    model_type = "accentik"

    def __init__(self, encoder_config=None, context_window=512,
                 batch_token_budget=8192, release="audio-v3", yo_threshold=0.8, **kwargs):
        super().__init__(**kwargs)
        self.encoder_config = deepcopy(encoder_config or {})
        self.context_window = int(context_window)
        self.batch_token_budget = int(batch_token_budget)
        self.release = release
        self.yo_threshold = float(yo_threshold)
        self.restore_yo_default = False
        if not 0.0 <= self.yo_threshold <= 1.0:
            raise ValueError("yo_threshold must be between zero and one")
        if self.context_window < 8 or self.context_window % 8:
            raise ValueError("context_window must be a positive multiple of eight")
        if self.batch_token_budget < self.context_window:
            raise ValueError("batch_token_budget must fit one complete context window")