# coding=utf-8 # Copyright 2026 The Qwen team, Alibaba Group and 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 transformers.configuration_utils import PretrainedConfig, layer_type_validation from transformers.modeling_rope_utils import rope_config_validation from transformers.utils import logging logger = logging.get_logger(__name__) class Qwen3TTSSpeakerEncoderConfig(PretrainedConfig): r""" This is the configuration class to store the configuration of a [`Qwen3TTSSpeakerEncoder`]. It is used to instantiate a Qwen3TTS speaker encoder model according to the specified arguments, defining the model architecture. The architecture is based on the ECAPA-TDNN model. Args: mel_dim (`int`, *optional*, defaults to 128): The dimension of the input mel-spectrogram. enc_dim (`int`, *optional*, defaults to 192): The dimension of the final speaker embedding. enc_channels (`list[int]`, *optional*, defaults to `[512, 512, 512, 512, 1536]`): A list of output channels for each TDNN/SERes2Net layer in the encoder. The first channel size is for the initial TDNN layer, the intermediate ones for the `SqueezeExcitationRes2NetBlock` layers, and the last one for the multi-layer feature aggregation. enc_kernel_sizes (`list[int]`, *optional*, defaults to `[5, 3, 3, 3, 1]`): A list of kernel sizes for each layer in the encoder, corresponding to `enc_channels`. enc_dilations (`list[int]`, *optional*, defaults to `[1, 2, 3, 4, 1]`): A list of dilations for each layer in the encoder, corresponding to `enc_channels`. enc_attention_channels (`int`, *optional*, defaults to 128): The number of attention channels in the `AttentiveStatisticsPooling` layer. enc_res2net_scale (`int`, *optional*,defaults to 8): The scale of the `Res2NetBlock` in the encoder. enc_se_channels (`int`, *optional*, defaults to 128): The number of channels in the squeeze part of the `SqueezeExcitationBlock`. """ def __init__( self, mel_dim=128, enc_dim=1024, enc_channels=[512, 512, 512, 512, 1536], enc_kernel_sizes=[5, 3, 3, 3, 1], enc_dilations=[1, 2, 3, 4, 1], enc_attention_channels=128, enc_res2net_scale=8, enc_se_channels=128, sample_rate=24000, ): self.mel_dim = mel_dim self.enc_dim = enc_dim self.enc_channels = enc_channels self.enc_kernel_sizes = enc_kernel_sizes self.enc_dilations = enc_dilations self.enc_attention_channels = enc_attention_channels self.enc_res2net_scale = enc_res2net_scale self.enc_se_channels = enc_se_channels self.sample_rate = sample_rate class Qwen3TTSTalkerCodePredictorConfig(PretrainedConfig): r""" This is the configuration class to store the configuration of a [`Qwen3TTSTalkerCodePredictorModel`]. It is used to instantiate a Qwen3TTSTalkerCodePredictor model according to the specified arguments, defining the model architecture. Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the documentation from [`PretrainedConfig`] for more information. Args: vocab_size (`int`, *optional*, defaults to 151936): Vocabulary size of the Qwen3TTSTalkerCodePredictor model. Defines the number of different tokens that can be represented by the `inputs_ids` passed when calling [`Qwen3TTSTalkerCodePredictorModel`] hidden_size (`int`, *optional*, defaults to 4096): Dimension of the hidden representations. intermediate_size (`int`, *optional*, defaults to 22016): Dimension of the MLP representations. num_hidden_layers (`int`, *optional*, defaults to 32): Number of hidden layers in the Transformer encoder. num_attention_heads (`int`, *optional*, defaults to 32): Number of attention heads for each attention layer in the Transformer encoder. num_key_value_heads (`int`, *optional*, defaults to 32): This is the number of key_value heads that should be used to implement Grouped Query Attention. If `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed by meanpooling all the original heads within that group. For more details, check out [this paper](https://huggingface.co/papers/2305.13245). If it is not specified, will default to `32`. head_dim (`int`, *optional*, defaults to 128): The attention head dimension. hidden_act (`str` or `function`, *optional*, defaults to `"silu"`): The non-linear activation function (function or string) in the decoder. max_position_embeddings (`int`, *optional*, defaults to 32768): The maximum sequence length that this model might ever be used with. initializer_range (`float`, *optional*, defaults to 0.02): The standard deviation of the truncated_normal_initializer for initializing all weight matrices. rms_norm_eps (`float`, *optional*, defaults to 1e-06): The epsilon used by the rms normalization layers. use_cache (`bool`, *optional*, defaults to `True`): Whether or not the model should return the last key/values attentions (not used by all models). Only relevant if `config.is_decoder=True`. tie_word_embeddings (`bool`, *optional*, defaults to `False`): Whether the model's input and output word embeddings should be tied. rope_theta (`float`, *optional*, defaults to 10000.0): The base period of the RoPE embeddings. rope_scaling (`Dict`, *optional*): Dictionary containing the scaling configuration for the RoPE embeddings. NOTE: if you apply new rope type and you expect the model to work on longer `max_position_embeddings`, we recommend you to update this value accordingly. Expected contents: `rope_type` (`str`): The sub-variant of RoPE to use. Can be one of ['default', 'linear', 'dynamic', 'yarn', 'longrope', 'llama3'], with 'default' being the original RoPE implementation. `factor` (`float`, *optional*): Used with all rope types except 'default'. The scaling factor to apply to the RoPE embeddings. In most scaling types, a `factor` of x will enable the model to handle sequences of length x * original maximum pre-trained length. `original_max_position_embeddings` (`int`, *optional*): Used with 'dynamic', 'longrope' and 'llama3'. The original max position embeddings used during pretraining. `attention_factor` (`float`, *optional*): Used with 'yarn' and 'longrope'. The scaling factor to be applied on the attention computation. If unspecified, it defaults to value recommended by the implementation, using the `factor` field to infer the suggested value. `beta_fast` (`float`, *optional*): Only used with 'yarn'. Parameter to set the boundary for extrapolation (only) in the linear ramp function. If unspecified, it defaults to 32. `beta_slow` (`float`, *optional*): Only used with 'yarn'. Parameter to set the boundary for interpolation (only) in the linear ramp function. If unspecified, it defaults to 1. `short_factor` (`list[float]`, *optional*): Only used with 'longrope'. The scaling factor to be applied to short contexts (< `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden size divided by the number of attention heads divided by 2 `long_factor` (`list[float]`, *optional*): Only used with 'longrope'. The scaling factor to be applied to long contexts (< `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden size divided by the number of attention heads divided by 2 `low_freq_factor` (`float`, *optional*): Only used with 'llama3'. Scaling factor applied to low frequency components of the RoPE `high_freq_factor` (`float`, *optional*): Only used with 'llama3'. Scaling factor applied to high frequency components of the RoPE attention_bias (`bool`, defaults to `False`, *optional*, defaults to `False`): Whether to use a bias in the query, key, value and output projection layers during self-attention. use_sliding_window (`bool`, *optional*, defaults to `False`): Whether to use sliding window attention. sliding_window (`int`, *optional*, defaults to 4096): Sliding window attention (SWA) window size. If not specified, will default to `4096`. max_window_layers (`int`, *optional*, defaults to 28): The number of layers using full attention. The first `max_window_layers` layers will use full attention, while any additional layer afterwards will use SWA (Sliding Window Attention). layer_types (`list`, *optional*): Attention pattern for each layer. attention_dropout (`float`, *optional*, defaults to 0.0): The dropout ratio for the attention probabilities. """ model_type = "qwen3_tts_talker_code_predictor" keys_to_ignore_at_inference = ["past_key_values"] # Default tensor parallel plan for base model `Qwen3TTSTalkerCodePredictor` base_model_tp_plan = { "layers.*.self_attn.q_proj": "colwise", "layers.*.self_attn.k_proj": "colwise", "layers.*.self_attn.v_proj": "colwise", "layers.*.self_attn.o_proj": "rowwise", "layers.*.mlp.gate_proj": "colwise", "layers.*.mlp.up_proj": "colwise", "layers.*.mlp.down_proj": "rowwise", } base_model_pp_plan = { "embed_tokens": (["input_ids"], ["inputs_embeds"]), "layers": (["hidden_states", "attention_mask"], ["hidden_states"]), "norm": (["hidden_states"], ["hidden_states"]), } def __init__( self, vocab_size=2048, hidden_size=1024, intermediate_size=3072, num_hidden_layers=5, num_attention_heads=16, num_key_value_heads=8, head_dim=128, hidden_act="silu", max_position_embeddings=32768, initializer_range=0.02, rms_norm_eps=0.000001, use_cache=True, tie_word_embeddings=False, rope_theta=10000, rope_scaling=None, attention_bias=False, use_sliding_window=False, sliding_window=4096, max_window_layers=28, layer_types=None, attention_dropout=0, num_code_groups=32, **kwargs, ): super().__init__( tie_word_embeddings=tie_word_embeddings, **kwargs, ) self.vocab_size = vocab_size self.max_position_embeddings = max_position_embeddings self.hidden_size = hidden_size self.intermediate_size = intermediate_size self.num_hidden_layers = num_hidden_layers self.num_attention_heads = num_attention_heads self.use_sliding_window = use_sliding_window self.sliding_window = sliding_window if self.use_sliding_window else None self.max_window_layers = max_window_layers # for backward compatibility if num_key_value_heads is None: num_key_value_heads = num_attention_heads self.num_key_value_heads = num_key_value_heads self.head_dim = head_dim self.hidden_act = hidden_act self.initializer_range = initializer_range self.rms_norm_eps = rms_norm_eps self.use_cache = use_cache self.rope_theta = rope_theta self.rope_scaling = rope_scaling self.attention_bias = attention_bias self.attention_dropout = attention_dropout # Validate the correctness of rotary position embeddings parameters # BC: if there is a 'type' field, move it to 'rope_type'. if self.rope_scaling is not None and "type" in self.rope_scaling: self.rope_scaling["rope_type"] = self.rope_scaling["type"] rope_config_validation(self) self.layer_types = layer_types if self.layer_types is None: self.layer_types = [ "sliding_attention" if self.sliding_window is not None and i >= self.max_window_layers else "full_attention" for i in range(self.num_hidden_layers) ] layer_type_validation(self.layer_types) self.num_code_groups = num_code_groups class Qwen3TTSTalkerConfig(PretrainedConfig): r""" This is the configuration class to store the configuration of a [`Qwen3TTSTalkerModel`]. It is used to instantiate a Qwen3TTSTalker model according to the specified arguments, defining the model architecture. Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the documentation from [`PretrainedConfig`] for more information. Args: vocab_size (`int`, *optional*, defaults to 151936): Vocabulary size of the Qwen3TTSTalker model. Defines the number of different tokens that can be represented by the `inputs_ids` passed when calling [`Qwen3TTSTalkerModel`] hidden_size (`int`, *optional*, defaults to 2048): Dimension of the hidden representations. intermediate_size (`int`, *optional*, defaults to 6144): Dimension of the MLP representations. num_hidden_layers (`int`, *optional*, defaults to 24): Number of hidden layers in the Transformer encoder. num_attention_heads (`int`, *optional*, defaults to 32): Number of attention heads for each attention layer in the Transformer encoder. num_key_value_heads (`int`, *optional*, defaults to 4): This is the number of key_value heads that should be used to implement Grouped Query Attention. If `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed by meanpooling all the original heads within that group. For more details, check out [this paper](https://huggingface.co/papers/2305.13245). If it is not specified, will default to `32`. hidden_act (`str` or `function`, *optional*, defaults to `"silu"`): The non-linear activation function (function or string) in the decoder. max_position_embeddings (`int`, *optional*, defaults to 32768): The maximum sequence length that this model might ever be used with. initializer_range (`float`, *optional*, defaults to 0.02): The standard deviation of the truncated_normal_initializer for initializing all weight matrices. rms_norm_eps (`float`, *optional*, defaults to 1e-06): The epsilon used by the rms normalization layers. use_cache (`bool`, *optional*, defaults to `True`): Whether or not the model should return the last key/values attentions (not used by all models). Only relevant if `config.is_decoder=True`. tie_word_embeddings (`bool`, *optional*, defaults to `False`): Whether the model's input and output word embeddings should be tied. rope_theta (`float`, *optional*, defaults to 10000.0): The base period of the RoPE embeddings. rope_scaling (`Dict`, *optional*): Dictionary containing the scaling configuration for the RoPE embeddings. NOTE: if you apply new rope type and you expect the model to work on longer `max_position_embeddings`, we recommend you to update this value accordingly. Expected contents: `rope_type` (`str`): The sub-variant of RoPE to use. Can be one of ['default', 'linear', 'dynamic', 'yarn', 'longrope', 'llama3'], with 'default' being the original RoPE implementation. `factor` (`float`, *optional*): Used with all rope types except 'default'. The scaling factor to apply to the RoPE embeddings. In most scaling types, a `factor` of x will enable the model to handle sequences of length x * original maximum pre-trained length. `original_max_position_embeddings` (`int`, *optional*): Used with 'dynamic', 'longrope' and 'llama3'. The original max position embeddings used during pretraining. `attention_factor` (`float`, *optional*): Used with 'yarn' and 'longrope'. The scaling factor to be applied on the attention computation. If unspecified, it defaults to value recommended by the implementation, using the `factor` field to infer the suggested value. `beta_fast` (`float`, *optional*): Only used with 'yarn'. Parameter to set the boundary for extrapolation (only) in the linear ramp function. If unspecified, it defaults to 32. `beta_slow` (`float`, *optional*): Only used with 'yarn'. Parameter to set the boundary for interpolation (only) in the linear ramp function. If unspecified, it defaults to 1. `short_factor` (`list[float]`, *optional*): Only used with 'longrope'. The scaling factor to be applied to short contexts (< `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden size divided by the number of attention heads divided by 2 `long_factor` (`list[float]`, *optional*): Only used with 'longrope'. The scaling factor to be applied to long contexts (< `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden size divided by the number of attention heads divided by 2 `low_freq_factor` (`float`, *optional*): Only used with 'llama3'. Scaling factor applied to low frequency components of the RoPE `high_freq_factor` (`float`, *optional*): Only used with 'llama3'. Scaling factor applied to high frequency components of the RoPE attention_bias (`bool`, defaults to `False`, *optional*, defaults to `False`): Whether to use a bias in the query, key, value and output projection layers during self-attention. use_sliding_window (`bool`, *optional*, defaults to `False`): Whether to use sliding window attention. sliding_window (`int`, *optional*, defaults to 4096): Sliding window attention (SWA) window size. If not specified, will default to `4096`. attention_dropout (`float`, *optional*, defaults to 0.0): The dropout ratio for the attention probabilities. """ model_type = "qwen3_tts_talker" keys_to_ignore_at_inference = ["past_key_values"] # Default tensor parallel plan for base model `Qwen3TTSTalker` base_model_tp_plan = { "layers.*.self_attn.q_proj": "colwise", "layers.*.self_attn.k_proj": "colwise", "layers.*.self_attn.v_proj": "colwise", "layers.*.self_attn.o_proj": "rowwise", "layers.*.mlp.gate_proj": "colwise", "layers.*.mlp.up_proj": "colwise", "layers.*.mlp.down_proj": "rowwise", } base_model_pp_plan = { "embed_tokens": (["input_ids"], ["inputs_embeds"]), "layers": (["hidden_states", "attention_mask"], ["hidden_states"]), "norm": (["hidden_states"], ["hidden_states"]), } sub_configs = {"code_predictor_config": Qwen3TTSTalkerCodePredictorConfig} def __init__( self, code_predictor_config=None, vocab_size=3072, hidden_size=1024, intermediate_size=2048, num_hidden_layers=20, num_attention_heads=16, num_key_value_heads=2, hidden_act="silu", max_position_embeddings=32768, initializer_range=0.02, rms_norm_eps=0.000001, use_cache=True, tie_word_embeddings=False, rope_theta=10000, rope_scaling=None, attention_bias=False, use_sliding_window=False, sliding_window=4096, attention_dropout=0, num_code_groups=32, text_hidden_size=2048, codec_eos_token_id=4198, codec_think_id=4202, codec_nothink_id=4203, codec_think_bos_id=4204, codec_think_eos_id=4205, codec_pad_id=4196, codec_bos_id=4197, spk_id=None, spk_is_dialect=None, codec_language_id=None, **kwargs, ): super().__init__( tie_word_embeddings=tie_word_embeddings, **kwargs, ) self.vocab_size = vocab_size self.max_position_embeddings = max_position_embeddings self.hidden_size = hidden_size self.intermediate_size = intermediate_size self.num_hidden_layers = num_hidden_layers self.num_attention_heads = num_attention_heads self.use_sliding_window = use_sliding_window self.sliding_window = sliding_window if use_sliding_window else None self.num_key_value_heads = num_key_value_heads self.hidden_act = hidden_act self.initializer_range = initializer_range self.rms_norm_eps = rms_norm_eps self.use_cache = use_cache self.rope_theta = rope_theta self.rope_scaling = rope_scaling self.attention_bias = attention_bias self.attention_dropout = attention_dropout # Validate the correctness of rotary position embeddings parameters # BC: if there is a 'type' field, move it to 'rope_type'. if self.rope_scaling is not None and "type" in self.rope_scaling: self.rope_scaling["rope_type"] = self.rope_scaling["type"] if code_predictor_config is None: code_predictor_config = {} self.code_predictor_config = Qwen3TTSTalkerCodePredictorConfig() logger.info("code_predictor_config is None. Initializing code_predictor model with default values") elif isinstance(code_predictor_config, Qwen3TTSTalkerCodePredictorConfig): self.code_predictor_config = code_predictor_config else: self.code_predictor_config = Qwen3TTSTalkerCodePredictorConfig(**code_predictor_config) self.num_code_groups = num_code_groups self.text_hidden_size = text_hidden_size self.codec_eos_token_id = codec_eos_token_id self.codec_think_id = codec_think_id self.codec_language_id = codec_language_id self.codec_nothink_id = codec_nothink_id self.codec_think_bos_id = codec_think_bos_id self.codec_think_eos_id = codec_think_eos_id self.codec_pad_id = codec_pad_id self.codec_bos_id = codec_bos_id self.spk_id = spk_id self.spk_is_dialect = spk_is_dialect class Qwen3TTSConfig(PretrainedConfig): """ This is the configuration class to store the configuration of a [`Qwen3TTSForConditionalGeneration`]. """ model_type = "qwen3_tts" sub_configs = { "talker_config": Qwen3TTSTalkerConfig, "speaker_encoder_config": Qwen3TTSSpeakerEncoderConfig, } def __init__( self, talker_config=None, speaker_encoder_config=None, tokenizer_type=None, tts_model_size=None, tts_model_type=None, im_start_token_id=151644, im_end_token_id=151645, tts_pad_token_id=151671, tts_bos_token_id=151672, tts_eos_token_id=151673, **kwargs, ): super().__init__(**kwargs) if talker_config is None: talker_config = {} logger.info("talker_config is None. Initializing talker model with default values") if speaker_encoder_config is None: speaker_encoder_config = {} logger.info("speaker_encoder_config is None. Initializing talker model with default values") self.talker_config = Qwen3TTSTalkerConfig(**talker_config) self.speaker_encoder_config = Qwen3TTSSpeakerEncoderConfig(**speaker_encoder_config) self.tokenizer_type = tokenizer_type self.tts_model_size = tts_model_size self.tts_model_type = tts_model_type self.im_start_token_id = im_start_token_id self.im_end_token_id = im_end_token_id self.tts_pad_token_id = tts_pad_token_id self.tts_bos_token_id = tts_bos_token_id self.tts_eos_token_id = tts_eos_token_id __all__ = ["Qwen3TTSConfig", "Qwen3TTSTalkerConfig", "Qwen3TTSSpeakerEncoderConfig"]