configuration_motif.py

from transformers.configuration_utils import PretrainedConfig
from transformers.modeling_rope_utils import rope_config_validation
from transformers.utils import logging
from typing import Optional
import math

logger = logging.get_logger(__name__)


class MotifConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`MotifModel`]. It is used to instantiate a
    Motif model according to the specified arguments, defining the model architecture. Instantiating a configuration
    with the defaults will yield a similar configuration to that of
    Motif-102B [moreh/Motif-102B](https://huggingface.co/moreh/Motif-102B).

    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 Motif model. Defines the number of different tokens that can be represented by the
            `inputs_ids` passed when calling [`MotifModel`]
        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 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-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
        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 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.

    ```python
    >>> from transformers import MotifModel, MotifConfig

    >>> # Initializing a Motif style configuration
    >>> configuration = MotifConfig()

    >>> # Initializing a model from the Motif-102B style configuration
    >>> model = MotifModel(configuration)

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

    model_type = "Motif"
    keys_to_ignore_at_inference = ["past_key_values"]

    def __init__(
        self,
        vocab_size=151936,
        hidden_size=4096,
        intermediate_size=22016,
        num_hidden_layers=32,
        num_attention_heads=32,
        num_key_value_heads=32,
        hidden_act="silu",
        max_position_embeddings=32768,
        initializer_range=0.02,
        rms_norm_eps=1e-6,
        use_cache=True,
        tie_word_embeddings=False,
        rope_theta=10000.0,
        rope_scaling=None,
        use_sliding_window=False,
        sliding_window=4096,
        max_window_layers=28,
        attention_dropout=0.0,
        multi_token_heads: Optional[int] = None,
        **kwargs,
    ):
        """
        Arguments:
            multi_token_heads: If not None, use multi-token heads as in the paper https://arxiv.org/pdf/2404.19737
        """

        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.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.rope_scaling = rope_scaling
        self.attention_dropout = attention_dropout

        ###kwargs

        # some scale factors

        self.scale_emb = getattr(kwargs, "scale_emb", 1)
        self.init_scale_o = getattr(kwargs, "init_scale_o", 1)

        # muparam
        self.hidden_states_shrink = 1 / math.sqrt(num_hidden_layers)
        self.dim_model_base = hidden_size
        self.dim_model_base_attn  = (hidden_size // num_attention_heads)
        self.dim_model_base_init = hidden_size
        self.dim_model_base_lr = getattr(kwargs, "dim_model_base_lr", hidden_size//8)
        self.dim_model_base_lmh = 1
        self.dim_model_base_logits = hidden_size

        self.muP = getattr(kwargs, "muP", False)
        # proxy hidden size ( following YuLan-Mini )
        # reparameterization(wesar_weights)
        logger.info(kwargs)
        self.wesar_weights = getattr(kwargs, "wesar_weights", False)
        logger.info(f'initial wesar reparameterization : {self.wesar_weights}')

        # alpha (scale factor)
        self.embed_tokens_alpha = getattr(kwargs, "embed_tokens_alpha", None)
        self.q_proj_alpha = getattr(kwargs, "q_proj_alpha", None)
        self.k_proj_alpha = getattr(kwargs, "k_proj_alpha", None)
        self.v_proj_alpha = getattr(kwargs, "v_proj_alpha", None)
        self.o_proj_alpha = getattr(kwargs, "o_proj_alpha", None)
        self.down_proj_alpha =  getattr(kwargs, "down_proj_alpha", None)
        self.gate_up_proj_alpha =  getattr(kwargs, "gate_up_proj_alpha", None)
        self.input_layernorm_alpha = getattr(kwargs, "input_layernorm_alpha", None)
        self.post_attention_layernorm_alpha =  getattr(kwargs, "post_attention_layernorm_alpha", None)
        self.norm_alpha = getattr(kwargs, "norm_alpha", None)
        self.lm_head_alpha =  getattr(kwargs, "lm_head_alpha", None)
        self.use_norm_alpha =  getattr(kwargs, "use_norm_alpha", False)
        self.use_emb_alpha =  getattr(kwargs, "use_emb_alpha", False)

        # 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.multi_token_heads = multi_token_heads
        self.multi_token_config_validation()



        # moe
        self.topk_method = getattr(kwargs, "topk_method", None)
        self.scoring_func = getattr(kwargs, "scoring_func", None)
        self.routed_scaling_factor = getattr(kwargs, "routed_scaling_factor", None)
        self.norm_topk_prob = getattr(kwargs, "norm_topk_prob", None)
        self.seq_aux = getattr(kwargs, "seq_aux", None)
        self.hidden_act_moe =  getattr(kwargs, "hidden_act_moe", None)


        self.n_group =  getattr(kwargs, "n_group", None)
        self.n_routed_experts = getattr(kwargs, "n_routed_experts", None)
        self.moe_intermediate_size =  getattr(kwargs, "moe_intermediate_size", None)
        self.topk_group = getattr(kwargs, "topk_group", None)


        self.use_fused_mlp =  getattr(kwargs, "use_fused_mlp", None)
        self.use_moreh_moe =  getattr(kwargs, "use_moreh_moe", False)
        self.continual_training =  getattr(kwargs, "continual_training", False)

        # external
        self.first_expansion =  getattr(kwargs, "first_expansion", False)
        self.moe_layer =  getattr(kwargs, "moe_layer", False)



        super().__init__(
            tie_word_embeddings=tie_word_embeddings,
            **kwargs,
        )
        logger.info(f' kwargs : {kwargs}')
        logger.info(f'after wesar reparameterization : {self.wesar_weights}')

    def multi_token_config_validation(self):
        if self.multi_token_heads is not None:
            assert isinstance(self.multi_token_heads, int) and self.multi_token_heads >= 1