{
  "architectures": [
    "DoryForCausalLM"
  ],
  "model_type": "dory",
  "vocab_size": 131072,
  "hidden_size": 2560,
  "num_hidden_layers": 72,
  "intermediate_size": 7168,
  "num_attention_heads": 16,
  "num_key_value_heads": 4,
  "max_position_embeddings": 262144,
  "tie_word_embeddings": false,
  "ngpt_alpha_init": 0.05,
  "attention_output_gate_logit_scale_init": null,
  "mlp_bias": false,
  "attention_bias": false,
  "n_input_layers": 16,
  "n_recurrent_layers": 40,
  "n_output_layers": 16,
  "n_recurrent_loops": 2,
  "head_dim": 256,
  "hidden_act": "silu",
  "initializer_range": 0.0177,
  "dtype": "bfloat16",
  "sliding_window": 1024,
  "rope_parameters": {
    "sliding_attention": {
      "rope_type": "default",
      "rope_theta": 10000,
      "partial_rotary_factor": 1.0
    },
    "full_attention": {
      "rope_type": "default",
      "rope_theta": 10000000,
      "partial_rotary_factor": 1.0
    }
  },
  "layer_types": [
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "full_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "full_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "full_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "full_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "full_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "full_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "full_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "full_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "sliding_attention",
    "mlp",
    "full_attention",
    "mlp"
  ],
  "quantization_config": {
    "quant_method": "modelopt",
    "quant_algo": "W4A16_NVFP4",
    "producer": {
      "name": "modelopt",
      "version": "0.0.1.dev1+gfcc59e3a9"
    },
    "ignore": [
      "backbone.embeddings",
      "lm_head"
    ],
    "config_groups": {
      "group_0": {
        "weights": {
          "dynamic": false,
          "num_bits": 4,
          "type": "float",
          "group_size": 16
        },
        "targets": [
          "Linear"
        ]
      }
    }
  }
}