rh`'`ddlmZmZddlmZej eZGddeZdgZ y))PretrainedConfiglayer_type_validation)loggingceZdZdZdZdgZidddddddd d d d d d ddddd dd dddddd dd ddddZdgdgfddgdgfdgdgfdZ d fd ZxZ S)! Dots1Configa This is the configuration class to store the configuration of a [`Dots1Model`]. It is used to instantiate a `dots.llm1` model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of [rednote-hilab/dots.llm1.base](https://huggingface.co/rednote-hilab/dots.llm1.base). 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 model. Defines the number of different tokens that can be represented by the `input_ids` passed when calling [`Dots1Model`]. hidden_size (`int`, *optional*, defaults to 4608): Dimension of the hidden representations. intermediate_size (`int`, *optional*, defaults to 10944): Dimension of the MLP representations. moe_intermediate_size (`int`, *optional*, defaults to 1408): Dimension of the MoE representations. num_hidden_layers (`int`, *optional*, defaults to 62): 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*, defaults to 32): Number of key/value heads for Grouped Query Attention. If `num_key_value_heads=num_attention_heads`, Multi Head Attention (MHA) is used. If `num_key_value_heads=1`, Multi Query Attention (MQA) is used. Otherwise, Grouped Query Attention (GQA) is used. If not specified, defaults to `num_attention_heads`. n_shared_experts (`int`, *optional*, default=None): Number of shared experts. None means dense model. n_routed_experts (`int`, *optional*, default=None): Number of routed experts. None means dense model. n_group (`int`, *optional*, defaults to 1): Number of groups for routed experts. topk_group (`int`, *optional*, defaults to 1): Number of selected groups for each token (selected experts only within `topk_group` groups). num_experts_per_tok (`int`, *optional*, default=None): Number of selected experts. None means dense model. first_k_dense_replace (`int`, *optional*, defaults to 0): Number of dense layers at the beginning of the model before the first MoE layer. norm_topk_prob (`bool`, *optional*, defaults to `False`): Whether to normalize the weights of the routed experts. hidden_act (`str` or `function`, *optional*, defaults to `"silu"`): The non-linear activation function (function or string). max_position_embeddings (`int`, *optional*, defaults to 2048): Maximum sequence length the model might ever be used with. initializer_range (`float`, *optional*, defaults to 0.02): Standard deviation of the truncated_normal_initializer for initializing all weight matrices. rms_norm_eps (`float`, *optional*, defaults to 1e-06): 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. Only relevant if `config.is_decoder=True`. tie_word_embeddings (`bool`, *optional*, defaults to `False`): Whether to tie the input and output word embeddings. rope_theta (`float`, *optional*, defaults to 10000.0): The base period of the RoPE embeddings. rope_scaling (`dict`, *optional*): Dictionary for scaling RoPE embeddings. Supports `{"type": strategy name, "factor": scaling factor}`. attention_bias (`bool`, *optional*, defaults to `False`): Whether to use a bias in the self-attention projections. attention_dropout (`float`, *optional*, defaults to 0.0): Dropout ratio for the attention probabilities. routed_scaling_factor (`float`, *optional*, defaults to 1.0): Scaling factor for routed experts. sliding_window (`int`, *optional*, defaults to 4096): Size of the sliding window for attention. If not specified, defaults to `4096`. max_window_layers (`int`, *optional*, defaults to 62): The number of layers using full attention. The first `max_window_layers` layers will use full attention, while any additional layer afterwards will use SWA (Sliding Window Attention). layer_types (`list`, *optional*): Attention pattern for each layer. Examples: ```python >>> from transformers import Dots1Model, Dots1Config >>> # Initializing a Dots1 style configuration >>> configuration = Dots1Config() >>> # Accessing the model configuration >>> configuration = model.config ``` dots1past_key_valueszlayers.*.self_attn.q_projcolwisezlayers.*.self_attn.k_projzlayers.*.self_attn.v_projzlayers.*.self_attn.o_projrowwisez layers.*.mlp.experts.*.gate_proj local_colwisezlayers.*.mlp.experts.*.up_projz layers.*.mlp.experts.*.down_proj local_rowwisezlayers.*.mlp.experts.*localz%layers.*.mlp.shared_experts.gate_projz#layers.*.mlp.shared_experts.up_projz%layers.*.mlp.shared_experts.down_projzlayers.*.mlp.shared_expertszlayers.*.mlp.gate_projzlayers.*.mlp.up_projzlayers.*.mlp.down_projz layers.*.mlpgather input_ids inputs_embeds hidden_statesattention_mask) embed_tokenslayersnormc ||_||_||_||_||_||_||_||_| |_| |_ | |_ ||_ ||}| |_ | |_ ||_||_||_||_||_||_||_||_||_||_||_||_||_|j4Et7|j Dcgc]!}|j0||j2k\rdnd#c}|_t9|j4t;|xdd|i|ycc}w)Nsliding_attentionfull_attentiontie_word_embeddings) vocab_sizemax_position_embeddings hidden_sizeintermediate_sizemoe_intermediate_sizenum_hidden_layersnum_attention_headsn_shared_expertsn_routed_expertsnum_experts_per_tokfirst_k_dense_replacenorm_topk_probn_group topk_groupnum_key_value_heads hidden_actinitializer_range rms_norm_eps use_cache rope_theta rope_scalingattention_biasattention_dropoutrouted_scaling_factorsliding_windowmax_window_layers layer_typesrangersuper__init__) selfrrrr r!r"r*r#r$r(r)r%r&r'r+rr,r-r.rr/r0r1r2r3r4r5r6kwargsi __class__s /var/www/html/ai-insurance-compliance-backend/venv/lib/python3.12/site-packages/transformers/models/dots1/configuration_dots1.pyr9zDots1Config.__init__so@%'>$&!2%:"!2#6 0 0#6 %:",  &"5  $#6 $!2("$(,!2%:",!2&    # t556  &&2qD r@NNrANFsiluig{Gz?gư>TFg@NFgg?ir?N) __name__ __module__ __qualname____doc__ model_typekeys_to_ignore_at_inferencebase_model_tp_planbase_model_pp_planr9 __classcell__)r=s@r>rrsvQfJ#4"5#Y#Y $Y $Y  +O  )/  +O !' 0 . 0 &w !/  !/ !(&(9:#%568IJ!"_$56"  $!!;J J rN) configuration_utilsrrutilsr get_loggerrDloggerr__all__rrMr>rSs;K   H %z "z z /rM