configuration_hithinkomni.py

import os
from typing import Union

from transformers.configuration_utils import PretrainedConfig
from transformers.models.auto import CONFIG_MAPPING
from transformers.modeling_rope_utils import rope_config_validation
from transformers.utils import logging


logger = logging.get_logger(__name__)


class HithinkOmniVisionConfig(PretrainedConfig):
    model_type = "hithink_omni"
    base_config_key = "vision_config"

    def __init__(
        self,
        depth=32,
        hidden_size=3584,
        hidden_act="silu",
        intermediate_size=3420,
        num_heads=16,
        in_channels=3,
        patch_size=14,
        spatial_merge_size=2,
        temporal_patch_size=2,
        tokens_per_second=4,
        window_size=112,
        out_hidden_size=3584,
        fullatt_block_indexes=[7, 15, 23, 31],
        **kwargs,
    ):
        super().__init__(**kwargs)

        self.depth = depth
        self.hidden_size = hidden_size
        self.hidden_act = hidden_act
        self.intermediate_size = intermediate_size
        self.num_heads = num_heads
        self.in_channels = in_channels
        self.patch_size = patch_size
        self.spatial_merge_size = spatial_merge_size
        self.temporal_patch_size = temporal_patch_size
        self.tokens_per_second = tokens_per_second
        self.window_size = window_size
        self.fullatt_block_indexes = fullatt_block_indexes
        self.out_hidden_size = out_hidden_size


class HithinkAudioEncoderConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`HithinkAudioEncoder`]. It is used to instantiate a
    HithinkAudio 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 HithinkAudio
    architecture.

    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
            `HithinkOmniProcessor` 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 HithinkAudioEncoderConfig, HithinkAudioEncoder

    >>> # Initializing a HithinkAudioEncoderConfig
    >>> configuration = HithinkAudioEncoderConfig()

    >>> # Initializing a HithinkAudioEncoder (with random weights)
    >>> model = HithinkAudioEncoder(configuration)

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

    model_type = "hithink_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 HithinkAudioDecoderConfig(PretrainedConfig):

    model_type = "hithink_omni_audio_decoder"

    def __init__(
        self,
        num_hidden_layers=6,
        codebook_size=1024,
        num_codebooks=8,
        **kwargs,
    ):
        super().__init__(**kwargs)
        self.num_hidden_layers = num_hidden_layers
        self.codebook_size = codebook_size
        self.num_codebooks = num_codebooks


class HithinkOmniConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`HithinkOmniModel`]. It is used to instantiate a
    HithinkOmni 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 152064):
            Vocabulary size of the HithinkOmni model. Defines the number of different tokens that can be represented by the
            `inputs_ids` passed when calling [`HithinkOmniModel`]
        hidden_size (`int`, *optional*, defaults to 8192):
            Dimension of the hidden representations.
        intermediate_size (`int`, *optional*, defaults to 29568):
            Dimension of the MLP representations.
        num_hidden_layers (`int`, *optional*, defaults to 80):
            Number of hidden layers in the Transformer encoder.
        num_attention_heads (`int`, *optional*, defaults to 64):
            Number of attention heads for each attention layer in the Transformer encoder.
        num_key_value_heads (`int`, *optional*, defaults to 8):
            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 checkout [this
            paper](https://arxiv.org/pdf/2305.13245.pdf). 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-05):
            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 1000000.0):
            The base period of the RoPE embeddings.
        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 80):
            The number of layers that use SWA (Sliding Window Attention). The bottom layers use SWA while the top use full attention.
        attention_dropout (`float`, *optional*, defaults to 0.0):
            The dropout ratio for the attention probabilities.
        vision_config (`Dict`, *optional*):
            The config for the visual encoder initialization.
        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

    ```python
    >>> from transformers import HithinkOmniForConditionalGeneration, HithinkOmniConfig

    >>> # Initializing a HithinkOmni style configuration
    >>> configuration = HithinkOmniConfig()

    >>> # Initializing a model from the HithinkOmni-7B style configuration
    >>> model = HithinkOmniForConditionalGeneration(configuration)

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

    model_type = "hithink_omni"
    sub_configs = {"vision_config": HithinkOmniVisionConfig}
    keys_to_ignore_at_inference = ["past_key_values"]
    # Default tensor parallel plan for base model `HithinkOmni`
    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",
    }

    def __init__(
        self,
        vocab_size=152064,
        vocab_size_ext=None,
        hidden_size=8192,
        intermediate_size=29568,
        num_hidden_layers=80,
        num_attention_heads=64,
        num_key_value_heads=8,
        hidden_act="silu",
        max_position_embeddings=32768,
        initializer_range=0.02,
        rms_norm_eps=1e-05,
        use_cache=True,
        tie_word_embeddings=False,
        rope_theta=1000000.0,
        use_sliding_window=False,
        sliding_window=4096,
        max_window_layers=80,
        attention_dropout=0.0,
        vision_config=None,
        rope_scaling=None,
        audio_config=None,
        audio_token_index=151665,
        audio_decoder_config=None,
        **kwargs,
    ):
        if isinstance(vision_config, dict):
            self.vision_config = self.sub_configs["vision_config"](**vision_config)
        elif vision_config is None:
            self.vision_config = self.sub_configs["vision_config"]()

        self.vocab_size = vocab_size
        self.vocab_size_ext = vocab_size_ext
        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
        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.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.attention_dropout = attention_dropout
        self.rope_scaling = rope_scaling

        # define audio config
        self.audio_token_index = audio_token_index
        self.ignore_index = -100
        if isinstance(audio_config, dict):
            audio_config = HithinkAudioEncoderConfig(**audio_config)
        elif audio_config is None:
            audio_config = HithinkAudioEncoderConfig(
                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(audio_decoder_config, dict):
            self.audio_decoder_config = HithinkAudioDecoderConfig(**audio_decoder_config)
        else:
            self.audio_decoder_config = None

        # Validate the correctness of rotary position embeddings parameters
        # BC: if there is a 'type' field, move it to 'rope_type'.
        # and change type from 'mrope' to 'default' because `mrope` does defeault RoPE calculations
        # one can set it to "linear"/"dynamic" etc. to have scaled RoPE
        # TODO: @raushan update config in the hub
        if self.rope_scaling is not None and "type" in self.rope_scaling:
            if self.rope_scaling["type"] == "mrope":
                self.rope_scaling["type"] = "default"
            self.rope_scaling["rope_type"] = self.rope_scaling["type"]
        rope_config_validation(self, ignore_keys={"mrope_section"})

        super().__init__(tie_word_embeddings=tie_word_embeddings, **kwargs)