config.json

{
  "architectures": [
    "Gemma3nForConditionalGeneration"
  ],
  "audio_config": {
    "conf_attention_chunk_size": 4,
    "conf_attention_context_left": 5,
    "conf_attention_context_right": 0,
    "conf_attention_logit_cap": 50.0,
    "conf_conv_kernel_size": 3,
    "conf_num_attention_heads": 4,
    "conf_num_hidden_layers": 2,
    "conf_reduction_factor": 4,
    "conf_residual_weight": 0.5,
    "gradient_clipping": 10000000000.0,
    "hidden_size": 32,
    "input_feat_size": 32,
    "model_type": "gemma3n_audio",
    "rms_norm_eps": 1e-06,
    "sscp_conv_channel_size": [
      16,
      8
    ],
    "sscp_conv_group_norm_eps": 0.001,
    "sscp_conv_kernel_size": [
      [
        3,
        3
      ],
      [
        3,
        3
      ]
    ],
    "sscp_conv_stride_size": [
      [
        2,
        2
      ],
      [
        2,
        2
      ]
    ],
    "vocab_offset": 262272,
    "vocab_size": 128
  },
  "audio_soft_tokens_per_image": 4,
  "audio_token_id": 8,
  "boa_token_id": 6,
  "boi_token_id": 4,
  "dtype": "float32",
  "eoa_token_id": 7,
  "eoi_token_id": 5,
  "image_token_id": 3,
  "initializer_range": 0.02,
  "mm_tokens_per_image": 2,
  "model_type": "gemma3n",
  "text_config": {
    "activation_sparsity_pattern": [
      0.0,
      0.0,
      0.0,
      0.0
    ],
    "altup_active_idx": 0,
    "altup_coef_clip": 120.0,
    "altup_correct_scale": true,
    "altup_num_inputs": 2,
    "attention_bias": false,
    "attention_dropout": 0.0,
    "bos_token_id": 2,
    "eos_token_id": 1,
    "final_logit_softcapping": 30.0,
    "head_dim": 8,
    "hidden_activation": "gelu_pytorch_tanh",
    "hidden_size": 16,
    "hidden_size_per_layer_input": 16,
    "initializer_range": 0.02,
    "intermediate_size": [
      22,
      22,
      22,
      22
    ],
    "laurel_rank": 64,
    "layer_types": [
      "full_attention",
      "sliding_attention",
      "full_attention",
      "sliding_attention"
    ],
    "max_position_embeddings": 512,
    "model_type": "gemma3n_text",
    "num_attention_heads": 2,
    "num_hidden_layers": 4,
    "num_key_value_heads": 2,
    "num_kv_shared_layers": 2,
    "pad_token_id": 0,
    "rms_norm_eps": 1e-06,
    "rope_parameters": {
      "full_attention": {
        "rope_theta": 1000000.0,
        "rope_type": "default"
      },
      "sliding_attention": {
        "rope_theta": 10000.0,
        "rope_type": "default"
      }
    },
    "sliding_window": 512,
    "tie_word_embeddings": true,
    "use_cache": true,
    "vocab_size": 262400,
    "vocab_size_per_layer_input": 99
  },
  "tie_word_embeddings": true,
  "transformers_version": "5.16.0.dev0",
  "use_cache": true,
  "vision_config": {
    "architecture": "vit_pe_core_large_patch14_336",
    "do_pooling": false,
    "hidden_size": 32,
    "image_size": 20,
    "initializer_range": 0.02,
    "intermediate_size": 37,
    "is_training": true,
    "label_names": [
      "LABEL_0",
      "LABEL_1"
    ],
    "model_args": {
      "depth": 2,
      "embed_dim": 64,
      "global_pool": "",
      "img_size": [
        20,
        20
      ],
      "init_values": 0.1,
      "ref_feat_shape": [
        1,
        1
      ],
      "use_post_transformer_norm": false
    },
    "model_type": "gemma3n_vision",
    "num_attention_heads": 4,
    "num_channels": 3,
    "num_classes": 2,
    "num_hidden_layers": 2,
    "num_key_value_heads": 1,
    "patch_size": 5,
    "rms_norm_eps": 1e-06,
    "use_labels": true,
    "vocab_offset": 262144,
    "vocab_size": 128
  },
  "vision_soft_tokens_per_image": 4
}