# 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨 # This file was automatically generated from src/transformers/models/glm_moe_dsa/modular_glm_moe_dsa.py. # Do NOT edit this file manually as any edits will be overwritten by the generation of # the file from the modular. If any change should be done, please apply the change to the # modular_glm_moe_dsa.py file directly. One of our CI enforces this. # 🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨 # Copyright 2026 the HuggingFace Team. All rights reserved. # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. from ...configuration_utils import PreTrainedConfig, layer_type_validation from ...modeling_rope_utils import RopeParameters class GlmMoeDsaConfig(PreTrainedConfig): r""" This is the configuration class to store the configuration of a [`GlmMoeDsaModel`]. It is used to instantiate a GLM-5 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 GLM-5. e.g. [zai-org/GLM-5](https://huggingface.co/zai-org/GLM-5) 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 154880): Vocabulary size of the model. Defines the number of different tokens that can be represented by the `inputs_ids` passed when calling [`GlmMoeDsaModel`]. hidden_size (`int`, *optional*, defaults to 6144): Dimension of the hidden representations. intermediate_size (`int`, *optional*, defaults to 12288): Dimension of the dense MLP representations. moe_intermediate_size (`int`, *optional*, defaults to 2048): Dimension of the MoE expert representations. num_hidden_layers (`int`, *optional*, defaults to 78): Number of hidden layers in the Transformer decoder. num_attention_heads (`int`, *optional*, defaults to 64): Number of attention heads for each attention layer in the Transformer decoder. num_key_value_heads (`int`, *optional*, defaults to 64): Number of key-value heads for Grouped Query Attention. If equal to `num_attention_heads`, uses MHA. n_shared_experts (`int`, *optional*, defaults to 1): Number of shared experts in MoE layers. n_routed_experts (`int`, *optional*, defaults to 256): Number of routed experts in MoE layers. routed_scaling_factor (`float`, *optional*, defaults to 2.5): Scaling factor for routed experts. kv_lora_rank (`int`, *optional*, defaults to 512): Rank of the LoRA matrices for key and value projections (MLA). q_lora_rank (`int`, *optional*, defaults to 2048): Rank of the LoRA matrices for query projections (MLA). qk_rope_head_dim (`int`, *optional*, defaults to 64): Dimension of the query/key heads that use rotary position embeddings. qk_nope_head_dim (`int`, *optional*, defaults to 192): Dimension of the query/key heads that don't use rotary position embeddings. v_head_dim (`int`, *optional*, defaults to 256): Dimension of the value heads. 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. num_experts_per_tok (`int`, *optional*, defaults to 8): Number of experts selected per token. norm_topk_prob (`bool`, *optional*, defaults to `True`): Whether to normalize the weights of the routed experts. hidden_act (`str` or `function`, *optional*, defaults to `"silu"`): The non-linear activation function in the decoder. max_position_embeddings (`int`, *optional*, defaults to 202752): 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. pad_token_id (`int`, *optional*): Padding token id. bos_token_id (`int`, *optional*, defaults to 0): Beginning of stream token id. eos_token_id (`int`, *optional*, defaults to 1): End of stream token id. tie_word_embeddings (`bool`, *optional*, defaults to `False`): Whether to tie weight embeddings. rope_parameters (`RopeParameters`, *optional*): Configuration parameters for the RoPE embeddings, including `rope_theta` and optional scaling parameters. rope_interleave (`bool`, *optional*, defaults to `True`): Whether to interleave the rotary position embeddings. mlp_layer_types (`list`, *optional*): MLP type pattern for each layer (`"dense"` or `"sparse"`). Defaults to 3 dense + rest sparse. attention_bias (`bool`, *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. index_topk (`int`, *optional*, defaults to 2048): Number of top tokens selected by the indexer for sparse attention. index_head_dim (`int`, *optional*, defaults to 128): Head dimension for the indexer projections (DSA). index_n_heads (`int | None`, *optional*, defaults to 32): Number of heads for the indexer projections (DSA). indexer_rope_interleave (`bool`, *optional*, defaults to `True`): Whether the indexer uses interleaved rotary position embeddings. ```python >>> from transformers import GlmMoeDsaConfig, GlmMoeDsaModel >>> # Initializing a GLM-MoE-DSA configuration >>> configuration = GlmMoeDsaConfig() >>> # Initializing a model from the configuration >>> model = GlmMoeDsaModel(configuration) >>> # Accessing the model configuration >>> configuration = model.config ```""" model_type = "glm_moe_dsa" keys_to_ignore_at_inference = ["past_key_values"] base_model_tp_plan = { "layers.*.self_attn.o_proj": "rowwise", "layers.*.mlp.experts.gate_up_proj": "packed_colwise", "layers.*.mlp.experts.down_proj": "rowwise", "layers.*.mlp.experts": "moe_tp_experts", "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"]), } attribute_map = { "num_local_experts": "n_routed_experts", } def __init__( self, vocab_size: int | None = 154880, hidden_size: int | None = 6144, intermediate_size: int | None = 12288, moe_intermediate_size: int | None = 2048, num_hidden_layers: int | None = 78, num_attention_heads: int | None = 64, num_key_value_heads: int | None = 64, n_shared_experts: int | None = 1, n_routed_experts: int | None = 256, routed_scaling_factor: float | None = 2.5, kv_lora_rank: int | None = 512, q_lora_rank: int | None = 2048, qk_rope_head_dim: int | None = 64, qk_nope_head_dim: int | None = 192, v_head_dim: int | None = 256, n_group: int | None = 1, topk_group: int | None = 1, num_experts_per_tok: int | None = 8, norm_topk_prob: bool | None = True, hidden_act: str | None = "silu", max_position_embeddings: int | None = 202752, initializer_range: float | None = 0.02, rms_norm_eps: float | None = 1e-5, use_cache: bool | None = True, pad_token_id: int | None = None, bos_token_id: int | None = 0, eos_token_id: int | None = 1, tie_word_embeddings: bool | None = False, rope_parameters: RopeParameters | dict[str, RopeParameters] | None = None, mlp_layer_types=None, attention_bias: bool | None = False, attention_dropout: float | None = 0.0, index_topk: int | None = 2048, index_head_dim: int | None = 128, index_n_heads: int | None = 32, **kwargs, ): # Model dimensions self.vocab_size = vocab_size self.hidden_size = hidden_size self.intermediate_size = intermediate_size self.moe_intermediate_size = moe_intermediate_size self.num_hidden_layers = num_hidden_layers self.max_position_embeddings = max_position_embeddings # Attention dimensions (MLA) self.num_attention_heads = num_attention_heads self.num_key_value_heads = num_key_value_heads self.kv_lora_rank = kv_lora_rank self.q_lora_rank = q_lora_rank self.qk_rope_head_dim = qk_rope_head_dim self.qk_nope_head_dim = qk_nope_head_dim self.qk_head_dim = qk_nope_head_dim + qk_rope_head_dim self.v_head_dim = v_head_dim self.head_dim = qk_rope_head_dim # MoE parameters self.n_shared_experts = n_shared_experts self.n_routed_experts = n_routed_experts self.routed_scaling_factor = routed_scaling_factor self.n_group = n_group self.topk_group = topk_group self.num_experts_per_tok = num_experts_per_tok self.norm_topk_prob = norm_topk_prob # MLP layer types: first 3 dense, rest sparse self.mlp_layer_types = mlp_layer_types if self.mlp_layer_types is None: self.mlp_layer_types = ["dense"] * min(3, num_hidden_layers) + ["sparse"] * (num_hidden_layers - 3) layer_type_validation(self.mlp_layer_types, self.num_hidden_layers, attention=False) # Indexer (DSA) parameters self.index_topk = index_topk self.index_head_dim = index_head_dim self.index_n_heads = index_n_heads # General config self.hidden_act = hidden_act self.initializer_range = initializer_range self.rms_norm_eps = rms_norm_eps self.use_cache = use_cache self.attention_bias = attention_bias self.attention_dropout = attention_dropout self.rope_parameters = rope_parameters 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, ) __all__ = ["GlmMoeDsaConfig"]