{
  "architectures": [
    "SparseMoEDecoderForCausalLM"
  ],
  "attention_backend": "flash_attention_2",
  "attention_bias": false,
  "attention_dropout_prob": 0.1,
  "bos_token_id": 50256,
  "capacity_factor": 1.25,
  "classifier_dropout": null,
  "dtype": "float32",
  "eos_token_id": 50256,
  "expert_dropout_prob": 0.1,
  "ffn_decompose": false,
  "ffn_rank": 192,
  "has_no_defaults_at_init": true,
  "hidden_dropout_prob": 0.1,
  "hidden_size": 768,
  "initializer_range": 0.02,
  "intermediate_size": 2048,
  "kv_shared_dim": null,
  "layer_norm_eps": 1e-12,
  "max_position_embeddings": 1024,
  "model_type": "sparse_subspace_decoder",
  "moe_intermediate_size": 512,
  "nope_dims": 32,
  "norm_type": "rmsnorm",
  "num_attention_heads": 12,
  "num_dense_layers": 0,
  "num_experts": 4,
  "num_hidden_layers": 8,
  "o_shared_dim": 192,
  "pad_token_id": 50256,
  "q_shared_dim": null,
  "qk_private_dim": 64,
  "rms_norm_eps": 1e-06,
  "rope_dims": 32,
  "rope_scaling": null,
  "rope_theta": 10000.0,
  "router_noise_std": 1.0,
  "router_top_k": 2,
  "transformers_version": "4.56.0",
  "use_sparse_moe": true,
  "vo_private_dim": 64,
  "vocab_rank": null,
  "vocab_size": 50257,
  "vocab_subspace": false
}