tencent/Sequential-Hidden-Decoding-8B-n8-Instruct

1"""Qwen3ScaleSeq model configuration.2 3Extends Qwen3Config with scale_seq_times for embedding replication4to scale effective sequence length. See Scale_SeqLen_via_Embedding_Replication.md.5"""6 7from transformers import Qwen3Config8 9 10class Qwen3ScaleSeqConfig(Qwen3Config):11    """12    Configuration for Qwen3 with scaled sequence length via embedding replication.13 14    Adds one parameter on top of Qwen3Config:15        scale_seq_times (int): Number of additional embedding copies (n-1 in the doc).16            0 means no scaling (standard Qwen3 behavior).17            1 means 2x sequence length (original + 1 copy), etc.18    """19 20    model_type = "qwen3_scale_seq"21 22    def __init__(self, scale_seq_times=0, **kwargs):23        self.scale_seq_times = scale_seq_times24        super().__init__(**kwargs)25 26 27__all__ = ["Qwen3ScaleSeqConfig"]28