"""GLAP (Generalized Language Audio Pretraining) configuration."""

from transformers import PretrainedConfig


class GlapConfig(PretrainedConfig):
    model_type = "glap"

    def __init__(
        self,
        # Audio encoder (Dasheng)
        audio_embed_dim: int = 768,
        audio_depth: int = 12,
        audio_num_heads: int = 12,
        patch_size: list = None,
        patch_stride: list = None,
        target_length: int = 1008,
        sample_rate: int = 16000,
        # Text encoder (SONAR)
        text_vocab_size: int = 256206,
        text_model_dim: int = 1024,
        text_num_layers: int = 24,
        text_num_heads: int = 16,
        text_ffn_inner_dim: int = 8192,
        text_max_seq_len: int = 514,
        text_pad_idx: int = 0,
        text_dropout_p: float = 0.1,
        # Projection
        embed_size: int = 1024,
        **kwargs,
    ):
        super().__init__(**kwargs)
        self.audio_embed_dim = audio_embed_dim
        self.audio_depth = audio_depth
        self.audio_num_heads = audio_num_heads
        self.patch_size = patch_size or [64, 4]
        self.patch_stride = patch_stride or [64, 4]
        self.target_length = target_length
        self.sample_rate = sample_rate
        self.text_vocab_size = text_vocab_size
        self.text_model_dim = text_model_dim
        self.text_num_layers = text_num_layers
        self.text_num_heads = text_num_heads
        self.text_ffn_inner_dim = text_ffn_inner_dim
        self.text_max_seq_len = text_max_seq_len
        self.text_pad_idx = text_pad_idx
        self.text_dropout_p = text_dropout_p
        self.embed_size = embed_size