| """DeepSeek-V4 model configuration. |
| |
| Adapted from the DeepSeek-V4 inference config (deepseek-ai/DeepSeek-V4-Pro) |
| and the HF Transformers DeepSeek-V3 config for HF compatibility. |
| |
| Key V4-specific features vs V3: |
| - Hyper-Connections (HC): multi-copy hidden states with Sinkhorn routing |
| - Compressed Sparse Attention (CSA): compression + sliding window + sparse indexing |
| - New MoE routing: sqrtsoftplus scoring, hash-based routing for first layers |
| - Large head_dim (512), o_groups/o_lora_rank for grouped output projection |
| - No kv_lora_rank (replaced by compress_ratios) |
| - No v_head_dim/qk_nope_head_dim (replaced by head_dim) |
| """ |
|
|
| from transformers.configuration_utils import PretrainedConfig |
|
|
|
|
| class DeepseekV4Config(PretrainedConfig): |
| model_type = "deepseek_v4" |
| keys_to_ignore_at_inference = ["past_key_values"] |
|
|
| def __init__( |
| self, |
| vocab_size=129280, |
| hidden_size=4096, |
| num_hidden_layers=43, |
| num_attention_heads=64, |
| num_key_value_heads=1, |
| |
| moe_intermediate_size=2048, |
| n_routed_experts=256, |
| n_shared_experts=1, |
| num_experts_per_tok=6, |
| norm_topk_prob=True, |
| scoring_func="sqrtsoftplus", |
| routed_scaling_factor=1.5, |
| topk_method="noaux_tc", |
| num_hash_layers=3, |
| swiglu_limit=10.0, |
| |
| q_lora_rank=1024, |
| head_dim=512, |
| qk_rope_head_dim=64, |
| o_groups=8, |
| o_lora_rank=1024, |
| sliding_window=128, |
| |
| compress_ratios=None, |
| compress_rope_theta=160000.0, |
| |
| index_n_heads=64, |
| index_head_dim=128, |
| index_topk=512, |
| |
| hc_mult=4, |
| hc_sinkhorn_iters=20, |
| hc_eps=1e-6, |
| |
| num_nextn_predict_layers=1, |
| |
| hidden_act="silu", |
| max_position_embeddings=4096, |
| initializer_range=0.02, |
| rms_norm_eps=1e-6, |
| use_cache=True, |
| pad_token_id=None, |
| bos_token_id=0, |
| eos_token_id=1, |
| tie_word_embeddings=False, |
| rope_theta=10000.0, |
| rope_scaling=None, |
| attention_bias=False, |
| attention_dropout=0.0, |
| **kwargs, |
| ): |
| self.vocab_size = vocab_size |
| self.hidden_size = hidden_size |
| self.num_hidden_layers = num_hidden_layers |
| self.num_attention_heads = num_attention_heads |
| self.num_key_value_heads = num_key_value_heads or num_attention_heads |
|
|
| |
| self.moe_intermediate_size = moe_intermediate_size |
| self.n_routed_experts = n_routed_experts |
| self.n_shared_experts = n_shared_experts |
| self.num_experts_per_tok = num_experts_per_tok |
| self.norm_topk_prob = norm_topk_prob |
| self.scoring_func = scoring_func |
| self.routed_scaling_factor = routed_scaling_factor |
| self.topk_method = topk_method |
| self.num_hash_layers = num_hash_layers |
| self.swiglu_limit = swiglu_limit |
|
|
| |
| self.q_lora_rank = q_lora_rank |
| self.head_dim = head_dim |
| self.qk_rope_head_dim = qk_rope_head_dim |
| self.nope_head_dim = head_dim - qk_rope_head_dim |
| self.o_groups = o_groups |
| self.o_lora_rank = o_lora_rank |
| self.sliding_window = sliding_window |
|
|
| |
| if compress_ratios is None: |
| |
| compress_ratios = [0] * (num_hidden_layers + 1) |
| self.compress_ratios = compress_ratios |
| self.compress_rope_theta = compress_rope_theta |
|
|
| |
| self.index_n_heads = index_n_heads |
| self.index_head_dim = index_head_dim |
| self.index_topk = index_topk |
|
|
| |
| self.hc_mult = hc_mult |
| self.hc_sinkhorn_iters = hc_sinkhorn_iters |
| self.hc_eps = hc_eps |
|
|
| |
| self.num_nextn_predict_layers = num_nextn_predict_layers |
|
|
| |
| self.hidden_act = hidden_act |
| self.max_position_embeddings = max_position_embeddings |
| 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_bias = attention_bias |
| self.attention_dropout = attention_dropout |
|
|
| 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, |
| ) |
|
|
