configuration_bolmo.py

from dataclasses import asdict
from typing import Any

from transformers.configuration_utils import PretrainedConfig, layer_type_validation
from transformers.modeling_rope_utils import rope_config_validation
from .tokenization_bolmo import BolmoTokenizerConfig

class BolmoConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`Olmo3Model`]. It is used to instantiate an OLMo3
    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 [allenai/OLMo-3-0725-1B](https://huggingface.co/allenai/OLMo-3-0725-1B).

    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 50304):
            Vocabulary size of the Olmo3 model. Defines the number of different tokens that can be represented by the
            `inputs_ids` passed when calling [`Olmo3Model`]
        hidden_size (`int`, *optional*, defaults to 4096):
            Dimension of the hidden representations.
        intermediate_size (`int`, *optional*, defaults to 11008):
            Dimension of the MLP representations.
        num_hidden_layers (`int`, *optional*, defaults to 32):
            Number of hidden layers in the Transformer decoder.
        num_attention_heads (`int`, *optional*, defaults to 32):
            Number of attention heads for each attention layer in the Transformer decoder.
        num_key_value_heads (`int`, *optional*):
            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
            `num_attention_heads`.
        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 2048):
            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.
        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`.
        pad_token_id (`int`, *optional*, defaults to 1):
            Padding token id.
        bos_token_id (`int`, *optional*):
            Beginning of stream token id.
        eos_token_id (`int`, *optional*, defaults to 50279):
            End of stream token id.
        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
            Whether to tie weight embeddings
        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.
        attention_dropout (`float`, *optional*, defaults to 0.0):
            The dropout ratio for the attention probabilities.
        rms_norm_eps (`float`, *optional*, defaults to 1e-05):
            The epsilon used by the rms normalization layers.
        sliding_window (`int`, *optional*, defaults to 4096):
            Size of the sliding window for sliding window attention.
        layer_types (`list`, *optional*):
            Attention pattern for each layer. Defaults to sliding window attention
            for 3 out of 4 layers, and full attention for every 4th layer.

    ```python
    >>> from transformers import Olmo3Model, Olmo3Config

    >>> # Initializing a Olmo3 7B style configuration
    >>> configuration = Olmo3Config()

    >>> # Initializing a model from the Olmo3 7B style configuration
    >>> model = Olmo3Model(configuration)

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

    model_type = "bolmo"
    keys_to_ignore_at_inference = ["past_key_values"]
    base_model_tp_plan = {
        "layers.*.self_attn.q_proj": "colwise_rep",  # we need to replicate here due to the added norm on q and k
        "layers.*.self_attn.k_proj": "colwise_rep",  # we need to replicate here due to the added norm on q and k
        "layers.*.self_attn.v_proj": "colwise_rep",  # we need to replicate here due to the added norm on q and k
        "layers.*.self_attn.o_proj": "rowwise_rep",  # we need to replicate here due to the added norm on q and k
        "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=520,
        hidden_size=4096,
        intermediate_size=11008,
        num_hidden_layers=32,
        num_attention_heads=32,
        num_key_value_heads=None,
        hidden_act="silu",
        max_position_embeddings=2048,
        initializer_range=0.02,
        use_cache=True,
        pad_token_id=1,
        bos_token_id=None,
        eos_token_id=50279,
        tie_word_embeddings=False,
        rope_theta=10000.0,
        rope_scaling=None,
        attention_bias=False,
        attention_dropout=0.0,
        rms_norm_eps=1e-5,
        sliding_window=4096,
        layer_types=None,
        # bolmo config
        add_expanded_embeddings: bool = True,
        boundary_predictor_lookahead: int = 1,
        boundary_threshold: str = "sample:0",
        num_local_encoder_layers: int = 1,
        num_local_decoder_layers: int = 4,
        num_local_heads: int = 16,
        local_intermediate_size: int = 5504,
        local_rms_norm_eps=1e-5,
        subword_vocab_size: int = 100278, # dolma2_tokenizer subword vocab size
        tokenizer_config: BolmoTokenizerConfig | dict[str, Any] | None = None,
        **kwargs,
    ):
        super().__init__(
            pad_token_id=pad_token_id,
            bos_token_id=bos_token_id,
            eos_token_id=eos_token_id,
            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

        # 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.use_cache = use_cache
        self.rope_theta = rope_theta
        self.rope_scaling = rope_scaling
        self._rope_scaling_validation()
        self.attention_bias = attention_bias
        self.attention_dropout = attention_dropout

        self.rms_norm_eps = rms_norm_eps

        self.sliding_window = sliding_window
        self.layer_types = layer_types
        if self.layer_types is None:
            self.layer_types = [
                "sliding_attention" if (i + 1) % 4 != 0 else "full_attention" for i in range(self.num_hidden_layers)
            ]
        layer_type_validation(self.layer_types)

        # bolmo configuration
        self.add_expanded_embeddings = add_expanded_embeddings
        self.boundary_predictor_lookahead = boundary_predictor_lookahead
        self.boundary_threshold = boundary_threshold
        self.num_local_encoder_layers = num_local_encoder_layers
        self.num_local_decoder_layers = num_local_decoder_layers
        self.num_local_heads = num_local_heads
        self.local_intermediate_size = local_intermediate_size
        self.local_rms_norm_eps = local_rms_norm_eps
        self.subword_vocab_size = subword_vocab_size

        if tokenizer_config is None:
            self.tokenizer_config = asdict(BolmoTokenizerConfig.bolmo())
        elif isinstance(tokenizer_config, BolmoTokenizerConfig):
            self.tokenizer_config = asdict(tokenizer_config)
        else:
            self.tokenizer_config = tokenizer_config

    def _rope_scaling_validation(self):
        """
        Validate the `rope_scaling` configuration.
        """
        rope_config_validation(self)

__all__ = ["BolmoConfig"]