| 1 | { |
| 2 | "state-spaces/mamba-130m-hf": { |
| 3 | "architectures": [ |
| 4 | "MambaForCausalLM" |
| 5 | ], |
| 6 | "model_type": "mamba", |
| 7 | "text_model_type": "mamba", |
| 8 | "hidden_size": 768, |
| 9 | "total_num_hidden_layers": 24, |
| 10 | "total_num_attention_heads": 0, |
| 11 | "head_size": 0, |
| 12 | "vocab_size": 50280, |
| 13 | "total_num_kv_heads": 0, |
| 14 | "num_experts": 0, |
| 15 | "is_deepseek_mla": false, |
| 16 | "is_multimodal_model": false, |
| 17 | "dtype": "torch.float32" |
| 18 | }, |
| 19 | "mistralai/Mamba-Codestral-7B-v0.1": { |
| 20 | "architectures": [ |
| 21 | "Mamba2ForCausalLM" |
| 22 | ], |
| 23 | "model_type": "mamba", |
| 24 | "text_model_type": "mamba", |
| 25 | "hidden_size": 4096, |
| 26 | "total_num_hidden_layers": 64, |
| 27 | "total_num_attention_heads": 0, |
| 28 | "head_size": 0, |
| 29 | "vocab_size": 32768, |
| 30 | "total_num_kv_heads": 0, |
| 31 | "num_experts": 0, |
| 32 | "is_deepseek_mla": false, |
| 33 | "is_multimodal_model": false, |
| 34 | "dtype": "torch.bfloat16" |
| 35 | }, |
| 36 | "ibm-nasa-geospatial/Prithvi-EO-2.0-300M-TL-Sen1Floods11": { |
| 37 | "architectures": [ |
| 38 | "Terratorch" |
| 39 | ], |
| 40 | "model_type": "timm_wrapper", |
| 41 | "text_model_type": "timm_wrapper", |
| 42 | "hidden_size": 0, |
| 43 | "total_num_hidden_layers": 0, |
| 44 | "total_num_attention_heads": 0, |
| 45 | "head_size": 0, |
| 46 | "vocab_size": 0, |
| 47 | "total_num_kv_heads": 0, |
| 48 | "num_experts": 0, |
| 49 | "is_deepseek_mla": false, |
| 50 | "is_multimodal_model": true, |
| 51 | "dtype": "torch.float32" |
| 52 | }, |
| 53 | "tiiuae/falcon-mamba-7b-instruct": { |
| 54 | "architectures": [ |
| 55 | "FalconMambaForCausalLM" |
| 56 | ], |
| 57 | "model_type": "falcon_mamba", |
| 58 | "text_model_type": "falcon_mamba", |
| 59 | "hidden_size": 4096, |
| 60 | "total_num_hidden_layers": 64, |
| 61 | "total_num_attention_heads": 0, |
| 62 | "head_size": 0, |
| 63 | "vocab_size": 65024, |
| 64 | "total_num_kv_heads": 0, |
| 65 | "num_experts": 0, |
| 66 | "is_deepseek_mla": false, |
| 67 | "is_multimodal_model": false, |
| 68 | "dtype": "torch.bfloat16" |
| 69 | }, |
| 70 | "Zyphra/Zamba2-7B-instruct": { |
| 71 | "architectures": [ |
| 72 | "Zamba2ForCausalLM" |
| 73 | ], |
| 74 | "model_type": "zamba2", |
| 75 | "text_model_type": "zamba2", |
| 76 | "hidden_size": 3584, |
| 77 | "total_num_hidden_layers": 81, |
| 78 | "total_num_attention_heads": 32, |
| 79 | "head_size": 224, |
| 80 | "vocab_size": 32000, |
| 81 | "total_num_kv_heads": 32, |
| 82 | "num_experts": 0, |
| 83 | "is_deepseek_mla": false, |
| 84 | "is_multimodal_model": false, |
| 85 | "dtype": "torch.bfloat16" |
| 86 | }, |
| 87 | "databricks/dbrx-instruct": { |
| 88 | "architectures": [ |
| 89 | "DbrxForCausalLM" |
| 90 | ], |
| 91 | "model_type": "dbrx", |
| 92 | "text_model_type": "dbrx", |
| 93 | "hidden_size": 6144, |
| 94 | "total_num_hidden_layers": 40, |
| 95 | "total_num_attention_heads": 48, |
| 96 | "head_size": 128, |
| 97 | "vocab_size": 100352, |
| 98 | "total_num_kv_heads": 8, |
| 99 | "num_experts": 0, |
| 100 | "is_deepseek_mla": false, |
| 101 | "is_multimodal_model": false, |
| 102 | "dtype": "torch.bfloat16" |
| 103 | }, |
| 104 | "tiiuae/falcon-7b": { |
| 105 | "architectures": [ |
| 106 | "FalconForCausalLM" |
| 107 | ], |
| 108 | "model_type": "falcon", |
| 109 | "text_model_type": "falcon", |
| 110 | "hidden_size": 4544, |
| 111 | "total_num_hidden_layers": 32, |
| 112 | "total_num_attention_heads": 71, |
| 113 | "head_size": 64, |
| 114 | "vocab_size": 65024, |
| 115 | "total_num_kv_heads": 1, |
| 116 | "num_experts": 0, |
| 117 | "is_deepseek_mla": false, |
| 118 | "is_multimodal_model": false, |
| 119 | "dtype": "torch.bfloat16" |
| 120 | }, |
| 121 | "tiiuae/falcon-40b": { |
| 122 | "architectures": [ |
| 123 | "FalconForCausalLM" |
| 124 | ], |
| 125 | "model_type": "falcon", |
| 126 | "text_model_type": "falcon", |
| 127 | "hidden_size": 8192, |
| 128 | "total_num_hidden_layers": 60, |
| 129 | "total_num_attention_heads": 128, |
| 130 | "head_size": 64, |
| 131 | "vocab_size": 65024, |
| 132 | "total_num_kv_heads": 8, |
| 133 | "num_experts": 0, |
| 134 | "is_deepseek_mla": false, |
| 135 | "is_multimodal_model": false, |
| 136 | "dtype": "torch.bfloat16" |
| 137 | }, |
| 138 | "luccafong/deepseek_mtp_main_random": { |
| 139 | "architectures": [ |
| 140 | "DeepseekV3ForCausalLM" |
| 141 | ], |
| 142 | "model_type": "deepseek_v3", |
| 143 | "text_model_type": "deepseek_v3", |
| 144 | "hidden_size": 2560, |
| 145 | "total_num_hidden_layers": 5, |
| 146 | "total_num_attention_heads": 32, |
| 147 | "head_size": 576, |
| 148 | "vocab_size": 129280, |
| 149 | "total_num_kv_heads": 32, |
| 150 | "num_experts": 72, |
| 151 | "is_deepseek_mla": true, |
| 152 | "is_multimodal_model": false, |
| 153 | "dtype": "torch.bfloat16" |
| 154 | }, |
| 155 | "luccafong/deepseek_mtp_draft_random": { |
| 156 | "architectures": [ |
| 157 | "DeepseekV3ForCausalLM" |
| 158 | ], |
| 159 | "model_type": "deepseek_v3", |
| 160 | "text_model_type": "deepseek_v3", |
| 161 | "hidden_size": 2560, |
| 162 | "total_num_hidden_layers": 10, |
| 163 | "total_num_attention_heads": 32, |
| 164 | "head_size": 576, |
| 165 | "vocab_size": 129280, |
| 166 | "total_num_kv_heads": 32, |
| 167 | "num_experts": 72, |
| 168 | "is_deepseek_mla": true, |
| 169 | "is_multimodal_model": false, |
| 170 | "dtype": "torch.bfloat16" |
| 171 | }, |
| 172 | "Qwen/Qwen3-Next-80B-A3B-Instruct": { |
| 173 | "architectures": [ |
| 174 | "Qwen3NextForCausalLM" |
| 175 | ], |
| 176 | "model_type": "qwen3_next", |
| 177 | "text_model_type": "qwen3_next", |
| 178 | "hidden_size": 2048, |
| 179 | "total_num_hidden_layers": 48, |
| 180 | "total_num_attention_heads": 16, |
| 181 | "head_size": 256, |
| 182 | "vocab_size": 151936, |
| 183 | "total_num_kv_heads": 2, |
| 184 | "num_experts": 512, |
| 185 | "is_deepseek_mla": false, |
| 186 | "is_multimodal_model": false, |
| 187 | "dtype": "torch.bfloat16" |
| 188 | }, |
| 189 | "tiny-random/qwen3-next-moe": { |
| 190 | "architectures": [ |
| 191 | "Qwen3NextForCausalLM" |
| 192 | ], |
| 193 | "model_type": "qwen3_next", |
| 194 | "text_model_type": "qwen3_next", |
| 195 | "hidden_size": 8, |
| 196 | "total_num_hidden_layers": 4, |
| 197 | "total_num_attention_heads": 16, |
| 198 | "head_size": 32, |
| 199 | "vocab_size": 151936, |
| 200 | "total_num_kv_heads": 8, |
| 201 | "num_experts": 32, |
| 202 | "is_deepseek_mla": false, |
| 203 | "is_multimodal_model": false, |
| 204 | "dtype": "torch.bfloat16" |
| 205 | }, |
| 206 | "zai-org/GLM-4.5": { |
| 207 | "architectures": [ |
| 208 | "Glm4MoeForCausalLM" |
| 209 | ], |
| 210 | "model_type": "glm4_moe", |
| 211 | "text_model_type": "glm4_moe", |
| 212 | "hidden_size": 5120, |
| 213 | "total_num_hidden_layers": 92, |
| 214 | "total_num_attention_heads": 96, |
| 215 | "head_size": 128, |
| 216 | "vocab_size": 151552, |
| 217 | "total_num_kv_heads": 8, |
| 218 | "num_experts": 160, |
| 219 | "is_deepseek_mla": false, |
| 220 | "is_multimodal_model": false, |
| 221 | "dtype": "torch.bfloat16" |
| 222 | }, |
| 223 | "baidu/ERNIE-4.5-21B-A3B-PT": { |
| 224 | "architectures": [ |
| 225 | "Ernie4_5_MoeForCausalLM" |
| 226 | ], |
| 227 | "model_type": "ernie4_5_moe", |
| 228 | "text_model_type": "ernie4_5_moe", |
| 229 | "hidden_size": 2560, |
| 230 | "total_num_hidden_layers": 28, |
| 231 | "total_num_attention_heads": 20, |
| 232 | "head_size": 128, |
| 233 | "vocab_size": 103424, |
| 234 | "total_num_kv_heads": 4, |
| 235 | "num_experts": 64, |
| 236 | "is_deepseek_mla": false, |
| 237 | "is_multimodal_model": false, |
| 238 | "dtype": "torch.bfloat16" |
| 239 | }, |
| 240 | "lmsys/gpt-oss-20b-bf16": { |
| 241 | "architectures": [ |
| 242 | "GptOssForCausalLM" |
| 243 | ], |
| 244 | "model_type": "gpt_oss", |
| 245 | "text_model_type": "gpt_oss", |
| 246 | "hidden_size": 2880, |
| 247 | "total_num_hidden_layers": 24, |
| 248 | "total_num_attention_heads": 64, |
| 249 | "head_size": 64, |
| 250 | "vocab_size": 201088, |
| 251 | "total_num_kv_heads": 8, |
| 252 | "num_experts": 32, |
| 253 | "is_deepseek_mla": false, |
| 254 | "is_multimodal_model": false, |
| 255 | "dtype": "torch.bfloat16" |
| 256 | }, |
| 257 | "deepseek-ai/DeepSeek-V3.2-Exp": { |
| 258 | "architectures": [ |
| 259 | "DeepseekV32ForCausalLM" |
| 260 | ], |
| 261 | "model_type": "deepseek_v32", |
| 262 | "text_model_type": "deepseek_v32", |
| 263 | "hidden_size": 7168, |
| 264 | "total_num_hidden_layers": 61, |
| 265 | "total_num_attention_heads": 128, |
| 266 | "head_size": 576, |
| 267 | "vocab_size": 129280, |
| 268 | "total_num_kv_heads": 128, |
| 269 | "num_experts": 256, |
| 270 | "is_deepseek_mla": true, |
| 271 | "is_multimodal_model": false, |
| 272 | "dtype": "torch.bfloat16" |
| 273 | }, |
| 274 | "meta-llama/Llama-4-Scout-17B-16E-Instruct": { |
| 275 | "architectures": [ |
| 276 | "Llama4ForConditionalGeneration" |
| 277 | ], |
| 278 | "model_type": "llama4", |
| 279 | "text_model_type": "llama4_text", |
| 280 | "hidden_size": 5120, |
| 281 | "total_num_hidden_layers": 48, |
| 282 | "total_num_attention_heads": 40, |
| 283 | "head_size": 128, |
| 284 | "vocab_size": 202048, |
| 285 | "total_num_kv_heads": 8, |
| 286 | "num_experts": 16, |
| 287 | "is_deepseek_mla": false, |
| 288 | "is_multimodal_model": true, |
| 289 | "dtype": "torch.bfloat16" |
| 290 | }, |
| 291 | "nvidia/Llama-3_3-Nemotron-Super-49B-v1": { |
| 292 | "architectures": [ |
| 293 | "DeciLMForCausalLM" |
| 294 | ], |
| 295 | "model_type": "nemotron-nas", |
| 296 | "text_model_type": "nemotron-nas", |
| 297 | "hidden_size": 8192, |
| 298 | "total_num_hidden_layers": 80, |
| 299 | "total_num_attention_heads": 64, |
| 300 | "head_size": 128, |
| 301 | "vocab_size": 128256, |
| 302 | "total_num_kv_heads": 8, |
| 303 | "num_experts": 0, |
| 304 | "is_deepseek_mla": false, |
| 305 | "is_multimodal_model": false, |
| 306 | "dtype": "torch.bfloat16" |
| 307 | }, |
| 308 | "XiaomiMiMo/MiMo-7B-RL": { |
| 309 | "architectures": [ |
| 310 | "MiMoForCausalLM" |
| 311 | ], |
| 312 | "model_type": "mimo", |
| 313 | "text_model_type": "mimo", |
| 314 | "hidden_size": 4096, |
| 315 | "total_num_hidden_layers": 36, |
| 316 | "total_num_attention_heads": 32, |
| 317 | "head_size": 128, |
| 318 | "vocab_size": 151680, |
| 319 | "total_num_kv_heads": 8, |
| 320 | "num_experts": 0, |
| 321 | "is_deepseek_mla": false, |
| 322 | "is_multimodal_model": false, |
| 323 | "dtype": "torch.bfloat16" |
| 324 | }, |
| 325 | "meituan-longcat/LongCat-Flash-Chat": { |
| 326 | "architectures": [ |
| 327 | "LongcatFlashForCausalLM" |
| 328 | ], |
| 329 | "model_type": "longcat_flash", |
| 330 | "text_model_type": "longcat_flash", |
| 331 | "hidden_size": 6144, |
| 332 | "total_num_hidden_layers": 28, |
| 333 | "total_num_attention_heads": 64, |
| 334 | "head_size": 576, |
| 335 | "vocab_size": 131072, |
| 336 | "total_num_kv_heads": 64, |
| 337 | "num_experts": 512, |
| 338 | "is_deepseek_mla": true, |
| 339 | "is_multimodal_model": false, |
| 340 | "dtype": "torch.float32" |
| 341 | }, |
| 342 | "stepfun-ai/Step-3.5-Flash": { |
| 343 | "architectures": [ |
| 344 | "Step3p5ForCausalLM" |
| 345 | ], |
| 346 | "model_type": "step3p5", |
| 347 | "text_model_type": "step3p5", |
| 348 | "hidden_size": 4096, |
| 349 | "total_num_hidden_layers": 45, |
| 350 | "total_num_attention_heads": 64, |
| 351 | "head_size": 128, |
| 352 | "vocab_size": 128896, |
| 353 | "total_num_kv_heads": 8, |
| 354 | "num_experts": 288, |
| 355 | "is_deepseek_mla": false, |
| 356 | "is_multimodal_model": false, |
| 357 | "dtype": "torch.bfloat16" |
| 358 | }, |
| 359 | "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16": { |
| 360 | "architectures": [ |
| 361 | "NemotronHForCausalLM" |
| 362 | ], |
| 363 | "model_type": "nemotron_h", |
| 364 | "text_model_type": "nemotron_h", |
| 365 | "hidden_size": 2688, |
| 366 | "total_num_hidden_layers": 52, |
| 367 | "total_num_attention_heads": 32, |
| 368 | "head_size": 128, |
| 369 | "vocab_size": 131072, |
| 370 | "total_num_kv_heads": 2, |
| 371 | "num_experts": 128, |
| 372 | "is_deepseek_mla": false, |
| 373 | "is_multimodal_model": false, |
| 374 | "dtype": "torch.bfloat16" |
| 375 | } |
| 376 | } |