config.json

{
  "_attn_implementation_autoset": false,
  "add_cross_attention": false,
  "architectures": [
    "KimiLinearForCausalLM"
  ],
  "auto_map": {
    "AutoConfig": "configuration_kimi_linear.KimiLinearConfig",
    "AutoModelForCausalLM": "modeling_kimi_linear.KimiLinearForCausalLM"
  },
  "bos_token_id": 163584,
  "cross_attention_hidden_size": null,
  "decoder_start_token_id": null,
  "dtype": "float32",
  "eos_token_id": 163586,
  "finetuning_task": null,
  "first_k_dense_replace": 1,
  "head_dim": 32,
  "hidden_act": "silu",
  "hidden_size": 512,
  "initializer_range": 0.02,
  "intermediate_size": 256,
  "is_decoder": false,
  "kv_lora_rank": 32,
  "linear_attn_config": {
    "full_attn_layers": [
      4
    ],
    "head_dim": 64,
    "kda_layers": [
      1,
      2,
      3
    ],
    "num_heads": 8,
    "short_conv_kernel_size": 4
  },
  "mla_use_nope": true,
  "model_type": "kimi_linear",
  "moe_intermediate_size": 256,
  "moe_layer_freq": 1,
  "moe_renormalize": true,
  "moe_router_activation_func": "sigmoid",
  "num_attention_heads": 8,
  "num_expert_group": 1,
  "num_experts": 4,
  "num_experts_per_token": 2,
  "num_hidden_layers": 5,
  "num_key_value_heads": 8,
  "num_nextn_predict_layers": 0,
  "num_shared_experts": 1,
  "pad_token_id": 163839,
  "prefix": null,
  "pruned_heads": {},
  "q_lora_rank": null,
  "qk_nope_head_dim": 32,
  "qk_rope_head_dim": 16,
  "rms_norm_eps": 1e-05,
  "rope_parameters": {
    "rope_theta": 10000.0,
    "rope_type": "default"
  },
  "rope_theta": 10000.0,
  "routed_scaling_factor": 2.446,
  "sep_token_id": null,
  "task_specific_params": null,
  "tf_legacy_loss": false,
  "tie_encoder_decoder": false,
  "tie_word_embeddings": false,
  "tokenizer_class": null,
  "topk_group": 1,
  "torchscript": false,
  "transformers_version": "5.3.0",
  "use_bfloat16": false,
  "use_cache": true,
  "use_grouped_topk": true,
  "v_head_dim": 32,
  "vocab_size": 163840
}