#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
#           This file was automatically generated from src/transformers/models/qwen3_asr/modular_qwen3_asr.py.
#               Do NOT edit this file manually as any edits will be overwritten by the generation of
#             the file from the modular. If any change should be done, please apply the change to the
#                          modular_qwen3_asr.py file directly. One of our CI enforces this.
#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
# Copyright 2026 The HuggingFace Inc. team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from huggingface_hub.dataclasses import strict

from ...configuration_utils import PreTrainedConfig
from ...utils import auto_docstring
from ..auto import CONFIG_MAPPING, AutoConfig


@auto_docstring(checkpoint="Qwen/Qwen3-ASR-1.7B-hf")
@strict
class Qwen3ASREncoderConfig(PreTrainedConfig):
    r"""
    n_window (`int`, *optional*, defaults to 50):
        Half the number of mel frames in one encoder chunk. Each chunk processed by the conv stack has
        ``2 * n_window`` mel frames (1 second of audio at 16 kHz with a 10 ms hop).
    output_dim (`int`, *optional*, defaults to 3584):
        Dimensionality of the output.
    n_window_infer (`int`, *optional*, defaults to 800):
        Number of mel frames worth of audio over which each attention window spans. Must be a multiple
        of ``n_window * 2`` so attention windows align with encoder chunks.
    downsample_hidden_size (`int`, *optional*, defaults to 480):
        Hidden size of the convolutional downsampling stack.
    """

    model_type = "qwen3_asr_encoder"
    attribute_map = {
        "num_hidden_layers": "encoder_layers",
        "hidden_size": "d_model",
        "num_attention_heads": "encoder_attention_heads",
        "intermediate_size": "encoder_ffn_dim",
    }

    num_mel_bins: int = 128
    encoder_layers: int = 24
    encoder_attention_heads: int = 16
    encoder_ffn_dim: int = 4096
    d_model: int = 1024
    dropout: float | int = 0.0
    attention_dropout: float | int = 0.0
    activation_function: str = "gelu"
    activation_dropout: float | int = 0.0
    scale_embedding: bool = False
    initializer_range: float = 0.02

    n_window: int = 50
    output_dim: int = 3584
    n_window_infer: int = 800
    downsample_hidden_size: int = 480
    max_position_embeddings: int = 13


@auto_docstring(checkpoint="Qwen/Qwen3-ASR-1.7B-hf")
@strict
class Qwen3ASRConfig(PreTrainedConfig):
    r"""
    audio_token_id (`int`, *optional*, defaults to 151676):
        The audio token id to encode the audio prompt.
    timestamp_token_id (`int`, *optional*, defaults to 151705):
        Token ID of the ``<timestamp>`` marker in the tokenizer vocabulary. These markers
        delimit word boundaries in the forced-alignment input sequence.
    token_classification_bias (`bool`, *optional*, defaults to False):
        Whether the token classification head for forced alignment should have a bias term.

    Example:

    ```python
    >>> from transformers import Qwen3ASRForConditionalGeneration, Qwen3ASRConfig

    >>> # Initializing a Qwen3ASR style configuration
    >>> configuration = Qwen3ASRConfig()

    >>> # Initializing a model from the configuration
    >>> model = Qwen3ASRForConditionalGeneration(configuration)

    >>> # Accessing the model configuration
    >>> configuration = model.config
    ```"""

    model_type = "qwen3_asr"
    sub_configs = {"audio_config": AutoConfig, "text_config": AutoConfig}

    audio_config: dict | PreTrainedConfig | None = None
    text_config: dict | PreTrainedConfig | None = None
    audio_token_id: int = 151676
    timestamp_token_id: int = 151705
    pad_token_id: int = 151645
    eos_token_id: list[int] | tuple[int, ...] | int = (151643, 151645)
    initializer_range: float = 0.02
    tie_word_embeddings: bool = True
    token_classification_bias: bool = False

    def __post_init__(self, **kwargs):
        if isinstance(self.audio_config, dict):
            self.audio_config["model_type"] = self.audio_config.get("model_type", "qwen3_asr_encoder")
            self.audio_config = CONFIG_MAPPING[self.audio_config["model_type"]](**self.audio_config)
        elif self.audio_config is None:
            self.audio_config = CONFIG_MAPPING["qwen3_asr_encoder"]()

        if isinstance(self.text_config, dict):
            self.text_config["model_type"] = self.text_config.get("model_type", "qwen3")
            self.text_config = CONFIG_MAPPING[self.text_config["model_type"]](**self.text_config)
        elif self.text_config is None:
            self.text_config = CONFIG_MAPPING["qwen3"](
                hidden_size=2048,
                intermediate_size=6144,
                num_hidden_layers=28,
                num_attention_heads=16,
                num_key_value_heads=8,
                head_dim=128,
                max_position_embeddings=65536,
                tie_word_embeddings=True,
            )

        super().__post_init__(**kwargs)


__all__ = ["Qwen3ASREncoderConfig", "Qwen3ASRConfig"]