BioReason/bioreason/models/pl/configuration_pl.py

from transformers import PretrainedConfig

class ProteinLLMESMConfig(PretrainedConfig):
    model_type = "protein_llm"
    base_config_key = "esm_config"

    def __init__(
        self,
        # ESM2 related configurations
        esm_hidden_size=1280,
        esm_num_layers=33,
        esm_num_attention_heads=20,
        esm_vocab_size=33,
        esm_max_position_embeddings=1026,
        esm_layer_norm_eps=1e-5,
        esm_hidden_dropout_prob=0.1,
        esm_attention_probs_dropout_prob=0.1,
        esm_intermediate_size=5120,
        esm_hidden_act="gelu",
        esm_initializer_range=0.02,
        esm_layer_norm_eps=1e-5,
        **kwargs,
    ):
        super().__init__(**kwargs)

        # ESM2 configurations
        self.esm_hidden_size = esm_hidden_size
        self.esm_num_layers = esm_num_layers
        self.esm_num_attention_heads = esm_num_attention_heads
        self.esm_vocab_size = esm_vocab_size
        self.esm_max_position_embeddings = esm_max_position_embeddings
        self.esm_layer_norm_eps = esm_layer_norm_eps
        self.esm_hidden_dropout_prob = esm_hidden_dropout_prob
        self.esm_attention_probs_dropout_prob = esm_attention_probs_dropout_prob
        self.esm_intermediate_size = esm_intermediate_size
        self.esm_hidden_act = esm_hidden_act
        self.esm_initializer_range = esm_initializer_range

class ProteinLLMQFormerConfig(PretrainedConfig):
    model_type = "protein_llm"
    base_config_key = "qformer_config"

    def __init__(
        self,
        # Q-Former configurations
        qformer_hidden_size=768,
        qformer_num_hidden_layers=12,
        qformer_num_attention_heads=12,
        qformer_intermediate_size=3072,
        qformer_hidden_act="gelu",
        qformer_hidden_dropout_prob=0.1,
        qformer_attention_probs_dropout_prob=0.1,
        qformer_max_position_embeddings=512,
        qformer_layer_norm_eps=1e-12,
        qformer_initializer_range=0.02,
        qformer_vocab_size=30522,
        qformer_pad_token_id=0,
        qformer_position_embedding_type="absolute",
        qformer_use_cache=True,
        # Query tokens
        num_query_tokens=32,
        **kwargs,
    ):
        super().__init__(**kwargs)

        # Q-Former configurations
        self.qformer_hidden_size = qformer_hidden_size
        self.qformer_num_hidden_layers = qformer_num_hidden_layers
        self.qformer_num_attention_heads = qformer_num_attention_heads
        self.qformer_intermediate_size = qformer_intermediate_size
        self.qformer_hidden_act = qformer_hidden_act
        self.qformer_hidden_dropout_prob = qformer_hidden_dropout_prob
        self.qformer_attention_probs_dropout_prob = qformer_attention_probs_dropout_prob
        self.qformer_max_position_embeddings = qformer_max_position_embeddings
        self.qformer_layer_norm_eps = qformer_layer_norm_eps
        self.qformer_initializer_range = qformer_initializer_range
        self.qformer_vocab_size = qformer_vocab_size
        self.qformer_pad_token_id = qformer_pad_token_id
        self.qformer_position_embedding_type = qformer_position_embedding_type
        self.qformer_use_cache = qformer_use_cache
        self.num_query_tokens = num_query_tokens

class ProteinLLMConfig(PretrainedConfig):
    r"""
    This is the configuration class to store the configuration of a [`ProteinLLMModel`]. It is used to instantiate a
    Protein-LLM model according to the specified arguments, defining the model architecture. The model combines
    ESM2 protein encoder, Q-Former, and a language model for protein understanding and generation.

    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 language model. Defines the number of different tokens that can be represented by the
            `inputs_ids` passed when calling the model.
        hidden_size (`int`, *optional*, defaults to 8192):
            Dimension of the hidden representations in the language model.
        intermediate_size (`int`, *optional*, defaults to 29568):
            Dimension of the MLP representations in the language model.
        num_hidden_layers (`int`, *optional*, defaults to 80):
            Number of hidden layers in the Transformer encoder of the language model.
        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.
        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.
        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.
        max_window_layers (`int`, *optional*, defaults to 80):
            The number of layers that use SWA.
        attention_dropout (`float`, *optional*, defaults to 0.0):
            The dropout ratio for the attention probabilities.
        esm_config (`Dict`, *optional*):
            The config for the ESM2 protein encoder initialization.
        qformer_config (`Dict`, *optional*):
            The config for the Q-Former initialization.
        rope_scaling (`Dict`, *optional*):
            Dictionary containing the scaling configuration for the RoPE embeddings.
    """

    model_type = "protein_llm"
    sub_configs = {
        "esm_config": ProteinLLMESMConfig,
        "qformer_config": ProteinLLMQFormerConfig
    }
    keys_to_ignore_at_inference = ["past_key_values"]
    
    # Default tensor parallel plan for base model
    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=152064,
        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,
        esm_config=None,
        qformer_config=None,
        rope_scaling=None,
        protein_token_id=None,
        **kwargs,
    ):
        # Initialize ESM config
        if isinstance(esm_config, dict):
            self.esm_config = self.sub_configs["esm_config"](**esm_config)
        elif esm_config is None:
            self.esm_config = self.sub_configs["esm_config"]()
        else:
            self.esm_config = esm_config

        # Initialize Q-Former config
        if isinstance(qformer_config, dict):
            self.qformer_config = self.sub_configs["qformer_config"](**qformer_config)
        elif qformer_config is None:
            self.qformer_config = self.sub_configs["qformer_config"]()
        else:
            self.qformer_config = qformer_config

        # Language model configurations
        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
        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

        self.protein_token_id = protein_token_id

        # Validate the correctness of rotary position embeddings parameters
        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"]

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

__all__ = ["ProteinLLMConfig"]