{
    "architectures": [
        "Gemma4ForCausalLM"
    ],
    "attention_bias": false,
    "attention_dropout": 0.0,
    "attention_k_eq_v": false,
    "bos_token_id": 2,
    "dtype": "bfloat16",
    "enable_moe_block": false,
    "eos_token_id": [
        1,
        106
    ],
    "final_logit_softcapping": 30.0,
    "global_head_dim": 512,
    "head_dim": 256,
    "hidden_activation": "gelu_pytorch_tanh",
    "hidden_size": 1536,
    "hidden_size_per_layer_input": 256,
    "initializer_range": 0.02,
    "intermediate_size": 6144,
    "layer_types": [
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "full_attention",
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "full_attention",
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "full_attention",
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "full_attention",
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "full_attention",
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "full_attention",
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "sliding_attention",
        "full_attention"
    ],
    "max_position_embeddings": 131072,
    "model_type": "gemma4_text",
    "moe_intermediate_size": null,
    "num_attention_heads": 8,
    "num_experts": null,
    "num_global_key_value_heads": null,
    "num_hidden_layers": 35,
    "num_key_value_heads": 1,
    "num_kv_shared_layers": 20,
    "pad_token_id": 0,
    "quantization": {
        "group_size": 64,
        "bits": 4,
        "mode": "affine",
        "model.layers.0.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.0.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.1.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.1.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.2.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.2.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.3.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.3.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.4.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.4.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.5.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.5.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.6.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.6.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.7.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.7.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.8.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.8.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.9.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.9.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.10.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.10.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.11.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.11.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.12.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.12.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.13.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.13.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.14.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.14.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.15.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.15.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.16.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.16.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.17.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.17.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.18.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.18.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.19.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.19.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.20.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.20.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.21.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.21.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.22.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.22.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.23.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.23.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.24.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.24.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.25.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.25.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.26.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.26.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.27.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.27.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.28.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.28.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.29.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.29.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.30.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.30.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.31.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.31.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.32.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.32.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.33.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.33.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.34.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.34.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        }
    },
    "quantization_config": {
        "group_size": 64,
        "bits": 4,
        "mode": "affine",
        "model.layers.0.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.0.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.1.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.1.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.2.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.2.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.3.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.3.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.4.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.4.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.5.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.5.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.6.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.6.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.7.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.7.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.8.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.8.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.9.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.9.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.10.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.10.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.11.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.11.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.12.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.12.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.13.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.13.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.14.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.14.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.15.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.15.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.16.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.16.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.17.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.17.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.18.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.18.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.19.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.19.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.20.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.20.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.21.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.21.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.22.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.22.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.23.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.23.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.24.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.24.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.25.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.25.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.26.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.26.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.27.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.27.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.28.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.28.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.29.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.29.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.30.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.30.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.31.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.31.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.32.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.32.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.33.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.33.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.34.per_layer_input_gate": {
            "group_size": 64,
            "bits": 8
        },
        "model.layers.34.per_layer_projection": {
            "group_size": 64,
            "bits": 8
        }
    },
    "rms_norm_eps": 1e-06,
    "rope_parameters": {
        "full_attention": {
            "partial_rotary_factor": 0.25,
            "rope_theta": 1000000.0,
            "rope_type": "proportional"
        },
        "sliding_attention": {
            "rope_theta": 10000.0,
            "rope_type": "default"
        }
    },
    "sliding_window": 512,
    "tie_word_embeddings": true,
    "top_k_experts": null,
    "torch_dtype": "bfloat16",
    "transformers_version": "5.6.2",
    "use_bidirectional_attention": null,
    "use_cache": true,
    "use_double_wide_mlp": true,
    "vocab_size": 262144,
    "vocab_size_per_layer_input": 262144
}