{
  "architectures": [
    "QRRTDeltaLatentMoEForCausalLM"
  ],
  "attention_dropout": 0.0,
  "attention_scalar_gate": true,
  "bos_token_id": 1,
  "delta_backend": "auto",
  "delta_conv_kernel": 4,
  "dtype": "float32",
  "eos_token_id": 2,
  "expert_backend": "auto",
  "expert_intermediate_size": 176,
  "experts_per_token": 4,
  "gate_multiplier": 2.0,
  "head_dim": 64,
  "hidden_size": 512,
  "intermediate_size": 1792,
  "latent_size": 256,
  "layer_types": [
    "full_attention",
    "linear_attention",
    "linear_attention",
    "linear_attention",
    "full_attention",
    "linear_attention",
    "linear_attention",
    "linear_attention",
    "full_attention",
    "linear_attention",
    "linear_attention",
    "linear_attention",
    "full_attention",
    "linear_attention",
    "linear_attention",
    "linear_attention",
    "full_attention",
    "full_attention"
  ],
  "lora_alpha": 16,
  "lora_rank": 16,
  "max_position_embeddings": 2048,
  "mlp_scalar_gate": true,
  "model_type": "q_rrt_delta_latent_moe",
  "moe_budget": "total",
  "nope_every_n": null,
  "nope_layers": [],
  "num_attention_heads": 8,
  "num_core_layers": 8,
  "num_experts": 16,
  "num_hidden_layers": 18,
  "num_key_value_heads": 2,
  "num_passes": 2,
  "pad_token_id": 0,
  "qk_norm": true,
  "relaxed": true,
  "rms_norm_eps": 1e-05,
  "rope_theta": 10000.0,
  "routed_scaling_factor": 1.0,
  "router_bias_speed": 0.001,
  "router_seq_aux_coef": 0.0001,
  "shared_intermediate_size": 256,
  "swiglu_limit": 10.0,
  "tie_word_embeddings": true,
  "transformers_version": "5.17.0",
  "vocab_size": 32768
}