configuration_pure_qwen2.py

from transformers import PretrainedConfig
from transformers.utils import logging
from transformers.modeling_rope_utils import rope_config_validation

logger = logging.get_logger(__name__)


class PureQwen2Config(PretrainedConfig):

    model_type = "pure_qwen2"
    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,
        svd_rank=5, # A slightly overestimated rank of token embedding matrix
        num_pc_to_remove=1, # Number of principal component to remove
        center=False, # If True, centre the input token embedding matrix
        num_iters=2, # Number of subspace iterations to conduct
        alpha=1, # Feature expression factor in parameter-free self-attention module
        disable_pcr=False, 
        disable_pfsa=False, 
        disable_covariance=True, 
        **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
        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
        # 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.svd_rank = svd_rank
        self.num_pc_to_remove = num_pc_to_remove
        self.center = center
        self.num_iters = num_iters
        self.alpha = alpha
        self.disable_pcr = disable_pcr
        self.disable_pfsa = disable_pfsa
        self.disable_covariance = disable_covariance

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