# coding=utf-8 # Copyright 2024 Microsoft Research & University of Wisconsin-Madison 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. """Qwen2Audio model configuration""" from ...configuration_utils import PretrainedConfig from ...utils import logging from ..auto import CONFIG_MAPPING, AutoConfig logger = logging.get_logger(__name__) class Qwen2AudioEncoderConfig(PretrainedConfig): r""" This is the configuration class to store the configuration of a [`Qwen2AudioEncoder`]. It is used to instantiate a Qwen2-Audio audio encoder according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the audio encoder of the Qwen2-Audio architecture. e.g. [Qwen/Qwen2-Audio-7B](https://huggingface.co/Qwen/Qwen2-Audio-7B) Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the documentation from [`PretrainedConfig`] for more information. Args: num_mel_bins (`int`, *optional*, defaults to 128): Number of mel features used per input features. Should correspond to the value used in the `Qwen2AudioProcessor` class. encoder_layers (`int`, *optional*, defaults to 32): Number of encoder layers. encoder_attention_heads (`int`, *optional*, defaults to 20): Number of attention heads for each attention layer in the Transformer encoder. encoder_ffn_dim (`int`, *optional*, defaults to 5120): Dimensionality of the "intermediate" (often named feed-forward) layer in encoder. encoder_layerdrop (`float`, *optional*, defaults to 0.0): The LayerDrop probability for the encoder. See the [LayerDrop paper](see https://arxiv.org/abs/1909.11556) for more details. d_model (`int`, *optional*, defaults to 1280): Dimensionality of the layers. dropout (`float`, *optional*, defaults to 0.0): The dropout probability for all fully connected layers in the embeddings, encoder, and pooler. attention_dropout (`float`, *optional*, defaults to 0.0): The dropout ratio for the attention probabilities. activation_function (`str`, *optional*, defaults to `"gelu"`): The non-linear activation function (function or string) in the encoder and pooler. If string, `"gelu"`, `"relu"`, `"silu"` and `"gelu_new"` are supported. activation_dropout (`float`, *optional*, defaults to 0.0): The dropout ratio for activations inside the fully connected layer. scale_embedding (`bool`, *optional*, defaults to `False`): Scale embeddings by diving by sqrt(d_model). init_std (`float`, *optional*, defaults to 0.02): The standard deviation of the truncated_normal_initializer for initializing all weight matrices. max_source_positions (`int`, *optional*, defaults to 1500): The maximum sequence length of log-mel filter-bank features that this model might ever be used with. Example: ```python >>> from transformers import Qwen2AudioEncoderConfig, Qwen2AudioEncoder >>> # Initializing a Qwen2AudioEncoderConfig >>> configuration = Qwen2AudioEncoderConfig() >>> # Initializing a Qwen2AudioEncoder (with random weights) >>> model = Qwen2AudioEncoder(configuration) >>> # Accessing the model configuration >>> configuration = model.config ```""" model_type = "qwen2_audio_encoder" def __init__( self, num_mel_bins=128, encoder_layers=32, encoder_attention_heads=20, encoder_ffn_dim=5120, encoder_layerdrop=0.0, d_model=1280, dropout=0.0, attention_dropout=0.0, activation_function="gelu", activation_dropout=0.0, scale_embedding=False, init_std=0.02, max_source_positions=1500, **kwargs, ): super().__init__(**kwargs) self.num_mel_bins = num_mel_bins self.d_model = d_model self.encoder_layers = encoder_layers self.encoder_attention_heads = encoder_attention_heads self.encoder_ffn_dim = encoder_ffn_dim self.dropout = dropout self.attention_dropout = attention_dropout self.activation_function = activation_function self.activation_dropout = activation_dropout self.encoder_layerdrop = encoder_layerdrop self.num_hidden_layers = encoder_layers self.init_std = init_std self.scale_embedding = scale_embedding # scale factor will be sqrt(d_model) if True self.max_source_positions = max_source_positions class Qwen2AudioConfig(PretrainedConfig): r""" This is the configuration class to store the configuration of a [`Qwen2AudioForConditionalGeneration`]. It is used to instantiate an Qwen2-Audio model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the Qwen2-Audio. e.g. [Qwen/Qwen2-Audio-7B](https://huggingface.co/Qwen/Qwen2-Audio-7B) Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the documentation from [`PretrainedConfig`] for more information. Args: audio_config (`Union[AutoConfig, dict]`, *optional*, defaults to `CLIPVisionConfig`): The config object or dictionary of the audio backbone. text_config (`Union[AutoConfig, dict]`, *optional*, defaults to `LlamaConfig`): The config object or dictionary of the text backbone. audio_token_index (`int`, *optional*, defaults to 151646): The image token index to encode the image prompt. Example: ```python >>> from transformers import Qwen2AudioForConditionalGeneration, Qwen2AudioConfig, Qwen2AudioEncoderConfig, Qwen2Config >>> # Initializing a Qwen2AudioEncoder config >>> audio_config = Qwen2AudioEncoderConfig() >>> # Initializing a Qwen2 config >>> text_config = Qwen2Config() >>> # Initializing a Qwen2Audio configuration >>> configuration = Qwen2AudioConfig(audio_config, text_config) >>> # Initializing a model from the qwen2-audio style configuration >>> model = Qwen2AudioForConditionalGeneration(configuration) >>> # Accessing the model configuration >>> configuration = model.config ```""" model_type = "qwen2_audio" sub_configs = {"text_config": AutoConfig, "audio_config": AutoConfig} def __init__( self, audio_config=None, text_config=None, audio_token_index=151646, **kwargs, ): self.audio_token_index = audio_token_index if isinstance(audio_config, dict): audio_config["model_type"] = ( audio_config["model_type"] if "model_type" in audio_config else "qwen2_audio_encoder" ) audio_config = CONFIG_MAPPING[audio_config["model_type"]](**audio_config) elif audio_config is None: audio_config = CONFIG_MAPPING["qwen2_audio_encoder"]( d_model=1280, encoder_attention_heads=20, encoder_ffn_dim=5120, encoder_layerdrop=0.0, encoder_layers=32, num_mel_bins=128, max_source_positions=1500, scale_embedding=False, activation_function="gelu", ) self.audio_config = audio_config if isinstance(text_config, dict): text_config["model_type"] = text_config["model_type"] if "model_type" in text_config else "qwen2" text_config = CONFIG_MAPPING[text_config["model_type"]](**text_config) elif text_config is None: text_config = CONFIG_MAPPING["qwen2"]() self.text_config = text_config super().__init__(**kwargs) __all__ = ["Qwen2AudioConfig", "Qwen2AudioEncoderConfig"]