config.json

{
  "architectures": [
    "KimiLinearForCausalLM"
  ],
  "model_type": "kimi_linear",
  "auto_map": {
    "AutoConfig": "configuration_kimi_k3.KimiLinearConfig",
    "AutoModelForCausalLM": "modeling_kimi_linear.KimiLinearForCausalLM"
  },
  "vocab_size": 32768,
  "hidden_size": 2560,
  "num_hidden_layers": 32,
  "num_attention_heads": 20,
  "num_key_value_heads": 20,
  "max_position_embeddings": 1048576,
  "rms_norm_eps": 1e-06,
  "tie_word_embeddings": true,
  "linear_attn_config": {
    "full_attn_layers": [
      4,
      8,
      12,
      16,
      20,
      24,
      28,
      32
    ],
    "kda_layers": [
      1,
      2,
      3,
      5,
      6,
      7,
      9,
      10,
      11,
      13,
      14,
      15,
      17,
      18,
      19,
      21,
      22,
      23,
      25,
      26,
      27,
      29,
      30,
      31
    ],
    "num_heads": 20,
    "head_dim": 128,
    "short_conv_kernel_size": 4,
    "gate_lower_bound": -5.0,
    "use_full_rank_gate": true
  },
  "q_lora_rank": 1280,
  "kv_lora_rank": 640,
  "qk_nope_head_dim": 128,
  "qk_rope_head_dim": 0,
  "v_head_dim": 128,
  "mla_use_nope": true,
  "mla_use_output_gate": true,
  "num_experts": 0,
  "num_experts_per_tok": 0,
  "num_shared_experts": 0,
  "moe_intermediate_size": 0,
  "routed_expert_hidden_size": 0,
  "first_k_dense_replace": 32,
  "routed_scaling_factor": 1.0,
  "moe_renormalize": false,
  "moe_router_activation_func": "sigmoid",
  "dense_ffn_hidden": 6144,
  "intermediate_size": 6144,
  "hidden_act": "situ",
  "activation_situ_beta": 4.0,
  "activation_situ_linear_beta": 25.0,
  "attn_res_block_size": 4,
  "initializer_range": 0.02
}