{
    "architectures": [
        "NumbatForCausalLM"
    ],
    "attention_bias": false,
    "auto_map": {
        "AutoConfig": "configuration_numbat.NumbatConfig",
        "AutoModel": "modeling_numbat.NumbatModel",
        "AutoModelForCausalLM": "modeling_numbat.NumbatForCausalLM"
    },
    "bos_token_id": 0,
    "dense_layer_indices": [
        0,
        1,
        109,
        110,
        111,
        112,
        113
    ],
    "dtype": "bfloat16",
    "eos_token_id": 1,
    "fp32_buffer_suffixes": [
        "mlp.gate.e_score_correction_bias"
    ],
    "fp32_parameter_suffixes": [
        "self_attn.A_log",
        "self_attn.dt_bias"
    ],
    "gdn_config": {
        "chunk_size": 64,
        "conv_bias": false,
        "head_dim_k": 128,
        "head_dim_v": 256,
        "num_heads": 48,
        "qk_l2norm": true,
        "short_conv_kernel": 4,
        "state_dtype": "float32"
    },
    "hidden_act": "silu",
    "hidden_size": 12288,
    "intermediate_size": 24576,
    "is_encoder_decoder": false,
    "max_position_embeddings": 131072,
    "mla_config": {
        "kv_lora_rank": 512,
        "num_heads": 96,
        "q_lora_rank": 1536,
        "qk_nope_head_dim": 128,
        "qk_rope_head_dim": 64,
        "rope_theta": 1000000.0,
        "v_head_dim": 128
    },
    "mla_layer_indices": [
        3,
        7,
        11,
        15,
        19,
        23,
        27,
        31,
        35,
        39,
        43,
        47,
        51,
        55,
        59,
        63,
        67,
        71,
        75,
        79,
        83,
        87,
        91,
        95,
        99,
        103,
        107,
        111
    ],
    "mlp_bias": false,
    "model_type": "numbat",
    "moe_config": {
        "group_limited_routing": {
            "n_group": 72,
            "topk_group": 8
        },
        "load_balancing": {
            "dropless": true,
            "freeze_correction_at_token_fraction": 0.98
        },
        "moe_intermediate_size": 2816,
        "n_routed_experts": 1224,
        "n_shared_experts": 1,
        "norm_topk_prob": true,
        "num_experts_per_tok": 48,
        "num_moe_layers": 107,
        "routed_scaling_factor": 1.0,
        "router_dtype": "bfloat16",
        "router_score_dtype": "float32",
        "scoring_func": "sigmoid"
    },
    "mtp_config": {
        "depth": 1,
        "inference_use": "self-speculative drafter (1 draft token)",
        "layer_index_in_checkpoint": 114,
        "loss_weight": 0.1
    },
    "num_hidden_layers": 114,
    "pad_token_id": 2,
    "rms_norm_eps": 1e-06,
    "tie_word_embeddings": false,
    "transformers_version": "5.19.0",
    "use_cache": false,
    "vocab_size": 131072,
    "quantization_config": {
        "config_groups": {
            "group_0": {
                "input_activations": {
                    "dynamic": false,
                    "group_size": 16,
                    "num_bits": 4,
                    "type": "float"
                },
                "targets": [
                    "Linear"
                ],
                "weights": {
                    "dynamic": false,
                    "group_size": 16,
                    "num_bits": 4,
                    "type": "float"
                }
            }
        },
        "group_size": 16,
        "ignore": [
            "lm_head",
            "model.embed_tokens",
            "model.norm",
            "model.layers.0.*",
            "model.layers.1.*",
            "model.layers.109.*",
            "model.layers.110.*",
            "model.layers.111.*",
            "model.layers.112.*",
            "model.layers.113.*",
            "model.layers.114.*",
            "model.layers.*.mlp.gate",
            "model.layers.*.self_attn.a_proj",
            "model.layers.*.self_attn.b_proj",
            "model.layers.*.self_attn.*conv1d",
            "model.layers.*.*layernorm",
            "model.layers.*.self_attn.*layernorm",
            "model.layers.*.self_attn.o_norm"
        ],
        "producer": {
            "name": "modelopt",
            "version": "0.47.0"
        },
        "quant_algo": "NVFP4",
        "quant_method": "modelopt"
    }
}