{
"architectures": [
"Gemma3ForConditionalGeneration"
],
"boi_token_index": 255999,
"eoi_token_index": 256000,
"eos_token_id": [
1,
106
],
"image_token_index": 262144,
"initializer_range": 0.02,
"mm_tokens_per_image": 256,
"model_type": "gemma3",
"quantization_config": {
"config_groups": {
"group_0": {
"input_activations": {
"actorder": null,
"block_structure": null,
"dynamic": true,
"group_size": null,
"num_bits": 8,
"observer": null,
"observer_kwargs": {},
"strategy": "token",
"symmetric": true,
"type": "float"
},
"output_activations": null,
"targets": [
"Linear"
],
"weights": {
"actorder": null,
"block_structure": null,
"dynamic": false,
"group_size": null,
"num_bits": 8,
"observer": "minmax",
"observer_kwargs": {},
"strategy": "channel",
"symmetric": true,
"type": "float"
}
}
},
"format": "float-quantized",
"global_compression_ratio": 1.2400234132867625,
"ignore": [
"vision_tower.vision_model.encoder.layers.0.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.0.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.0.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.0.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.0.mlp.fc1",
"vision_tower.vision_model.encoder.layers.0.mlp.fc2",
"vision_tower.vision_model.encoder.layers.1.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.1.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.1.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.1.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.1.mlp.fc1",
"vision_tower.vision_model.encoder.layers.1.mlp.fc2",
"vision_tower.vision_model.encoder.layers.2.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.2.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.2.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.2.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.2.mlp.fc1",
"vision_tower.vision_model.encoder.layers.2.mlp.fc2",
"vision_tower.vision_model.encoder.layers.3.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.3.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.3.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.3.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.3.mlp.fc1",
"vision_tower.vision_model.encoder.layers.3.mlp.fc2",
"vision_tower.vision_model.encoder.layers.4.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.4.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.4.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.4.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.4.mlp.fc1",
"vision_tower.vision_model.encoder.layers.4.mlp.fc2",
"vision_tower.vision_model.encoder.layers.5.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.5.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.5.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.5.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.5.mlp.fc1",
"vision_tower.vision_model.encoder.layers.5.mlp.fc2",
"vision_tower.vision_model.encoder.layers.6.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.6.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.6.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.6.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.6.mlp.fc1",
"vision_tower.vision_model.encoder.layers.6.mlp.fc2",
"vision_tower.vision_model.encoder.layers.7.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.7.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.7.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.7.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.7.mlp.fc1",
"vision_tower.vision_model.encoder.layers.7.mlp.fc2",
"vision_tower.vision_model.encoder.layers.8.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.8.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.8.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.8.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.8.mlp.fc1",
"vision_tower.vision_model.encoder.layers.8.mlp.fc2",
"vision_tower.vision_model.encoder.layers.9.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.9.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.9.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.9.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.9.mlp.fc1",
"vision_tower.vision_model.encoder.layers.9.mlp.fc2",
"vision_tower.vision_model.encoder.layers.10.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.10.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.10.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.10.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.10.mlp.fc1",
"vision_tower.vision_model.encoder.layers.10.mlp.fc2",
"vision_tower.vision_model.encoder.layers.11.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.11.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.11.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.11.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.11.mlp.fc1",
"vision_tower.vision_model.encoder.layers.11.mlp.fc2",
"vision_tower.vision_model.encoder.layers.12.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.12.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.12.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.12.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.12.mlp.fc1",
"vision_tower.vision_model.encoder.layers.12.mlp.fc2",
"vision_tower.vision_model.encoder.layers.13.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.13.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.13.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.13.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.13.mlp.fc1",
"vision_tower.vision_model.encoder.layers.13.mlp.fc2",
"vision_tower.vision_model.encoder.layers.14.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.14.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.14.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.14.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.14.mlp.fc1",
"vision_tower.vision_model.encoder.layers.14.mlp.fc2",
"vision_tower.vision_model.encoder.layers.15.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.15.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.15.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.15.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.15.mlp.fc1",
"vision_tower.vision_model.encoder.layers.15.mlp.fc2",
"vision_tower.vision_model.encoder.layers.16.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.16.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.16.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.16.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.16.mlp.fc1",
"vision_tower.vision_model.encoder.layers.16.mlp.fc2",
"vision_tower.vision_model.encoder.layers.17.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.17.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.17.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.17.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.17.mlp.fc1",
"vision_tower.vision_model.encoder.layers.17.mlp.fc2",
"vision_tower.vision_model.encoder.layers.18.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.18.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.18.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.18.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.18.mlp.fc1",
"vision_tower.vision_model.encoder.layers.18.mlp.fc2",
"vision_tower.vision_model.encoder.layers.19.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.19.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.19.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.19.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.19.mlp.fc1",
"vision_tower.vision_model.encoder.layers.19.mlp.fc2",
"vision_tower.vision_model.encoder.layers.20.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.20.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.20.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.20.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.20.mlp.fc1",
"vision_tower.vision_model.encoder.layers.20.mlp.fc2",
"vision_tower.vision_model.encoder.layers.21.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.21.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.21.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.21.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.21.mlp.fc1",
"vision_tower.vision_model.encoder.layers.21.mlp.fc2",
"vision_tower.vision_model.encoder.layers.22.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.22.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.22.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.22.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.22.mlp.fc1",
"vision_tower.vision_model.encoder.layers.22.mlp.fc2",
"vision_tower.vision_model.encoder.layers.23.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.23.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.23.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.23.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.23.mlp.fc1",
"vision_tower.vision_model.encoder.layers.23.mlp.fc2",
"vision_tower.vision_model.encoder.layers.24.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.24.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.24.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.24.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.24.mlp.fc1",
"vision_tower.vision_model.encoder.layers.24.mlp.fc2",
"vision_tower.vision_model.encoder.layers.25.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.25.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.25.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.25.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.25.mlp.fc1",
"vision_tower.vision_model.encoder.layers.25.mlp.fc2",
"vision_tower.vision_model.encoder.layers.26.self_attn.k_proj",
"vision_tower.vision_model.encoder.layers.26.self_attn.v_proj",
"vision_tower.vision_model.encoder.layers.26.self_attn.q_proj",
"vision_tower.vision_model.encoder.layers.26.self_attn.out_proj",
"vision_tower.vision_model.encoder.layers.26.mlp.fc1",
"vision_tower.vision_model.encoder.layers.26.mlp.fc2",
"language_model.lm_head"
],
"kv_cache_scheme": null,
"quant_method": "compressed-tensors",
"quantization_status": "compressed"
},
"text_config": {
"attention_bias": false,
"attention_dropout": 0.0,
"attn_logit_softcapping": null,
"cache_implementation": "hybrid",
"final_logit_softcapping": null,
"head_dim": 128,
"hidden_activation": "gelu_pytorch_tanh",
"hidden_size": 5376,
"initializer_range": 0.02,
"intermediate_size": 21504,
"max_position_embeddings": 131072,
"model_type": "gemma3_text",
"num_attention_heads": 32,
"num_hidden_layers": 62,
"num_key_value_heads": 16,
"query_pre_attn_scalar": 168,
"rms_norm_eps": 1e-06,
"rope_local_base_freq": 10000.0,
"rope_scaling": {
"factor": 8.0,
"rope_type": "linear"
},
"rope_theta": 1000000.0,
"sliding_window": 1024,
"sliding_window_pattern": 6,
"use_cache": true,
"vocab_size": 262208
},
"torch_dtype": "bfloat16",
"transformers_version": "4.50.0.dev0",
"vision_config": {
"attention_dropout": 0.0,
"hidden_act": "gelu_pytorch_tanh",
"hidden_size": 1152,
"image_size": 896,
"intermediate_size": 4304,
"layer_norm_eps": 1e-06,
"model_type": "siglip_vision_model",
"num_attention_heads": 16,
"num_channels": 3,
"num_hidden_layers": 27,
"patch_size": 14,
"vision_use_head": false
}
}