config.json
| 1 | { |
| 2 | "architectures": [ |
| 3 | "Gemma4ForConditionalGeneration" |
| 4 | ], |
| 5 | "audio_config": null, |
| 6 | "audio_token_id": 258881, |
| 7 | "boa_token_id": 256000, |
| 8 | "boi_token_id": 255999, |
| 9 | "dtype": "bfloat16", |
| 10 | "eoa_token_id": 258883, |
| 11 | "eoa_token_index": 258883, |
| 12 | "eoi_token_id": 258882, |
| 13 | "eos_token_id": [ |
| 14 | 1, |
| 15 | 106 |
| 16 | ], |
| 17 | "image_token_id": 258880, |
| 18 | "initializer_range": 0.02, |
| 19 | "model_type": "gemma4", |
| 20 | "quantization_config": { |
| 21 | "config_groups": { |
| 22 | "FP8_BLOCK": { |
| 23 | "format": "float-quantized", |
| 24 | "input_activations": { |
| 25 | "actorder": null, |
| 26 | "block_structure": null, |
| 27 | "dynamic": true, |
| 28 | "group_size": 128, |
| 29 | "num_bits": 8, |
| 30 | "observer": null, |
| 31 | "observer_kwargs": {}, |
| 32 | "scale_dtype": null, |
| 33 | "strategy": "group", |
| 34 | "symmetric": true, |
| 35 | "type": "float", |
| 36 | "zp_dtype": null |
| 37 | }, |
| 38 | "output_activations": null, |
| 39 | "targets": [ |
| 40 | "Linear" |
| 41 | ], |
| 42 | "weights": { |
| 43 | "actorder": null, |
| 44 | "block_structure": [ |
| 45 | 128, |
| 46 | 128 |
| 47 | ], |
| 48 | "dynamic": false, |
| 49 | "group_size": null, |
| 50 | "num_bits": 8, |
| 51 | "observer": "memoryless_minmax", |
| 52 | "observer_kwargs": {}, |
| 53 | "scale_dtype": null, |
| 54 | "strategy": "block", |
| 55 | "symmetric": true, |
| 56 | "type": "float", |
| 57 | "zp_dtype": null |
| 58 | } |
| 59 | } |
| 60 | }, |
| 61 | "format": "float-quantized", |
| 62 | "global_compression_ratio": null, |
| 63 | "ignore": [ |
| 64 | "re:.*vision.*", |
| 65 | "lm_head", |
| 66 | "re:.*embed_tokens.*" |
| 67 | ], |
| 68 | "kv_cache_scheme": null, |
| 69 | "quant_method": "compressed-tensors", |
| 70 | "quantization_status": "compressed", |
| 71 | "sparsity_config": {}, |
| 72 | "transform_config": {}, |
| 73 | "version": "0.14.0.1" |
| 74 | }, |
| 75 | "text_config": { |
| 76 | "attention_bias": false, |
| 77 | "attention_dropout": 0.0, |
| 78 | "attention_k_eq_v": true, |
| 79 | "bos_token_id": 2, |
| 80 | "dtype": "bfloat16", |
| 81 | "enable_moe_block": false, |
| 82 | "eos_token_id": 1, |
| 83 | "expert_intermediate_size": null, |
| 84 | "final_logit_softcapping": 30.0, |
| 85 | "global_head_dim": 512, |
| 86 | "head_dim": 256, |
| 87 | "hidden_activation": "gelu_pytorch_tanh", |
| 88 | "hidden_size": 5376, |
| 89 | "hidden_size_per_layer_input": 0, |
| 90 | "initializer_range": 0.02, |
| 91 | "intermediate_size": 21504, |
| 92 | "layer_types": [ |
| 93 | "sliding_attention", |
| 94 | "sliding_attention", |
| 95 | "sliding_attention", |
| 96 | "sliding_attention", |
| 97 | "sliding_attention", |
| 98 | "full_attention", |
| 99 | "sliding_attention", |
| 100 | "sliding_attention", |
| 101 | "sliding_attention", |
| 102 | "sliding_attention", |
| 103 | "sliding_attention", |
| 104 | "full_attention", |
| 105 | "sliding_attention", |
| 106 | "sliding_attention", |
| 107 | "sliding_attention", |
| 108 | "sliding_attention", |
| 109 | "sliding_attention", |
| 110 | "full_attention", |
| 111 | "sliding_attention", |
| 112 | "sliding_attention", |
| 113 | "sliding_attention", |
| 114 | "sliding_attention", |
| 115 | "sliding_attention", |
| 116 | "full_attention", |
| 117 | "sliding_attention", |
| 118 | "sliding_attention", |
| 119 | "sliding_attention", |
| 120 | "sliding_attention", |
| 121 | "sliding_attention", |
| 122 | "full_attention", |
| 123 | "sliding_attention", |
| 124 | "sliding_attention", |
| 125 | "sliding_attention", |
| 126 | "sliding_attention", |
| 127 | "sliding_attention", |
| 128 | "full_attention", |
| 129 | "sliding_attention", |
| 130 | "sliding_attention", |
| 131 | "sliding_attention", |
| 132 | "sliding_attention", |
| 133 | "sliding_attention", |
| 134 | "full_attention", |
| 135 | "sliding_attention", |
| 136 | "sliding_attention", |
| 137 | "sliding_attention", |
| 138 | "sliding_attention", |
| 139 | "sliding_attention", |
| 140 | "full_attention", |
| 141 | "sliding_attention", |
| 142 | "sliding_attention", |
| 143 | "sliding_attention", |
| 144 | "sliding_attention", |
| 145 | "sliding_attention", |
| 146 | "full_attention", |
| 147 | "sliding_attention", |
| 148 | "sliding_attention", |
| 149 | "sliding_attention", |
| 150 | "sliding_attention", |
| 151 | "sliding_attention", |
| 152 | "full_attention" |
| 153 | ], |
| 154 | "max_position_embeddings": 262144, |
| 155 | "model_type": "gemma4_text", |
| 156 | "num_attention_heads": 32, |
| 157 | "num_experts": null, |
| 158 | "num_global_key_value_heads": 4, |
| 159 | "num_hidden_layers": 60, |
| 160 | "num_key_value_heads": 16, |
| 161 | "num_kv_shared_layers": 0, |
| 162 | "pad_token_id": 0, |
| 163 | "rms_norm_eps": 1e-06, |
| 164 | "rope_parameters": { |
| 165 | "full_attention": { |
| 166 | "partial_rotary_factor": 0.25, |
| 167 | "rope_theta": 1000000.0, |
| 168 | "rope_type": "proportional" |
| 169 | }, |
| 170 | "sliding_attention": { |
| 171 | "rope_theta": 10000.0, |
| 172 | "rope_type": "default" |
| 173 | } |
| 174 | }, |
| 175 | "sliding_window": 1024, |
| 176 | "tie_word_embeddings": true, |
| 177 | "top_k_experts": null, |
| 178 | "use_bidirectional_attention": "vision", |
| 179 | "use_cache": true, |
| 180 | "use_double_wide_mlp": false, |
| 181 | "vocab_size": 262144, |
| 182 | "vocab_size_per_layer_input": 262144 |
| 183 | }, |
| 184 | "tie_word_embeddings": true, |
| 185 | "transformers_version": "5.5.0.dev0", |
| 186 | "video_token_id": 258884, |
| 187 | "vision_config": { |
| 188 | "_name_or_path": "", |
| 189 | "architectures": null, |
| 190 | "attention_bias": false, |
| 191 | "attention_dropout": 0.0, |
| 192 | "chunk_size_feed_forward": 0, |
| 193 | "default_output_length": 280, |
| 194 | "dtype": "bfloat16", |
| 195 | "global_head_dim": 72, |
| 196 | "head_dim": 72, |
| 197 | "hidden_activation": "gelu_pytorch_tanh", |
| 198 | "hidden_size": 1152, |
| 199 | "id2label": { |
| 200 | "0": "LABEL_0", |
| 201 | "1": "LABEL_1" |
| 202 | }, |
| 203 | "initializer_range": 0.02, |
| 204 | "intermediate_size": 4304, |
| 205 | "is_encoder_decoder": false, |
| 206 | "label2id": { |
| 207 | "LABEL_0": 0, |
| 208 | "LABEL_1": 1 |
| 209 | }, |
| 210 | "max_position_embeddings": 131072, |
| 211 | "model_type": "gemma4_vision", |
| 212 | "num_attention_heads": 16, |
| 213 | "num_hidden_layers": 27, |
| 214 | "num_key_value_heads": 16, |
| 215 | "output_attentions": false, |
| 216 | "output_hidden_states": false, |
| 217 | "patch_size": 16, |
| 218 | "pooling_kernel_size": 3, |
| 219 | "position_embedding_size": 10240, |
| 220 | "problem_type": null, |
| 221 | "return_dict": true, |
| 222 | "rms_norm_eps": 1e-06, |
| 223 | "rope_parameters": { |
| 224 | "rope_theta": 100.0, |
| 225 | "rope_type": "default" |
| 226 | }, |
| 227 | "standardize": true, |
| 228 | "use_clipped_linears": false |
| 229 | }, |
| 230 | "vision_soft_tokens_per_image": 280 |
| 231 | } |