Spaces:
Running
Running
File size: 42,293 Bytes
eceb7f9 0657db8 f3995ad 0657db8 eceb7f9 104db8a eceb7f9 c30ebde 0657db8 c30ebde 0657db8 eceb7f9 d0bbc1a eceb7f9 d0bbc1a eceb7f9 c30ebde 0657db8 c30ebde 0657db8 104db8a d0bbc1a eceb7f9 c30ebde 0657db8 c30ebde 0657db8 eceb7f9 104db8a eceb7f9 c94f789 0657db8 eceb7f9 c94f789 eceb7f9 c94f789 eceb7f9 0657db8 eceb7f9 c30ebde | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 | {
"models": [
{
"id": "iq2_m",
"label": "IQ2_M",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "e8fe5e28e1b1c1f0cd0a39b85b16b631f17ca14e",
"platform": "gguf",
"description": "GGUF language weights for llama.cpp. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-IQ2_M.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-IQ2_M.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 CUDA sample validation; published copies use qwen2 metadata. Full quality measurements used the documented qwen35 protocol. CPU and Mac execution were not established by these tests.",
"file": "occamy-1.0-IQ2_M.gguf",
"bytes": 11659235328,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-IQ2_M.gguf?download=true"
},
{
"id": "q2_k",
"label": "Q2_K",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "e8fe5e28e1b1c1f0cd0a39b85b16b631f17ca14e",
"platform": "gguf",
"description": "GGUF language weights for llama.cpp. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-Q2_K.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-Q2_K.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 CUDA sample validation; published copies use qwen2 metadata. Full quality measurements used the documented qwen35 protocol. CPU and Mac execution were not established by these tests.",
"file": "occamy-1.0-Q2_K.gguf",
"bytes": 12939593728,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-Q2_K.gguf?download=true"
},
{
"id": "apex-i-mini",
"label": "APEX I-Mini",
"repo": "Accio-Lab/occamy-1.0-APEX-GGUF",
"revision": "7c731908b05f5cbd19002b23d421aa102c0dd0de",
"platform": "gguf",
"description": "Upstream APEX I-Mini mixed precision, directly from BF16 with an independently collected importance matrix; all 256 experts covered in every layer. Normalize prompt content to NFC.",
"status": "Experimental quality",
"command": "hf download Accio-Lab/occamy-1.0-APEX-GGUF occamy-1.0-APEX-I-Mini.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-APEX-I-Mini.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 native qwen2 + NFC: 20/20 authored cases; all authored cases passed; subset PPL 9.5643. All 733 tensors finite and 16/16 tokenizer probes. Full calibration coverage; calibrated does not mean better quality. Image, CPU and Mac were not retested.",
"file": "occamy-1.0-APEX-I-Mini.gguf",
"bytes": 13467210784,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF/resolve/main/occamy-1.0-APEX-I-Mini.gguf?download=true"
},
{
"id": "iq3_xs",
"label": "IQ3_XS",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "e8fe5e28e1b1c1f0cd0a39b85b16b631f17ca14e",
"platform": "gguf",
"description": "GGUF language weights for llama.cpp. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-IQ3_XS.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-IQ3_XS.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 CUDA sample validation; published copies use qwen2 metadata. Full quality measurements used the documented qwen35 protocol. CPU and Mac execution were not established by these tests.",
"file": "occamy-1.0-IQ3_XS.gguf",
"bytes": 14484144128,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-IQ3_XS.gguf?download=true"
},
{
"id": "q3_k_s",
"label": "Q3_K_S",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "6210e42ce97bc1b7c7ea58cab597e9b49ed523e5",
"platform": "gguf",
"description": "Standard mixed K-quant GGUF, generated directly from BF16 without calibration. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-Q3_K_S.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-Q3_K_S.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "Native qwen2 + NFC validation on B200: 19/20 authored cases, including executed Python and native tool round trips. Same Chinese JSON failure as BF16. 16/16 NFC tokenizer probes with special-token parsing. Vision, CPU and Mac execution were not tested for this addition.",
"file": "occamy-1.0-Q3_K_S.gguf",
"bytes": 15182184192,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-Q3_K_S.gguf?download=true"
},
{
"id": "iq3_m",
"label": "IQ3_M",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "e8fe5e28e1b1c1f0cd0a39b85b16b631f17ca14e",
"platform": "gguf",
"description": "GGUF language weights for llama.cpp. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-IQ3_M.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-IQ3_M.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 CUDA sample validation; published copies use qwen2 metadata. Full quality measurements used the documented qwen35 protocol. CPU and Mac execution were not established by these tests.",
"file": "occamy-1.0-IQ3_M.gguf",
"bytes": 15440519168,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-IQ3_M.gguf?download=true"
},
{
"id": "apex-compact",
"label": "APEX Compact",
"repo": "Accio-Lab/occamy-1.0-APEX-GGUF",
"revision": "7c731908b05f5cbd19002b23d421aa102c0dd0de",
"platform": "gguf",
"description": "Upstream APEX Compact MoE mixed-precision allocation, generated directly from BF16 without an importance matrix. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-APEX-GGUF occamy-1.0-APEX-Compact.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-APEX-Compact.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 native qwen2 + NFC: 20/20 authored functional cases, including executed Python and tool round trips; subset PPL 8.4567. All 733 tensors finite and 16/16 tokenizer probes. Bounded checks; image, CPU and Mac execution were not tested.",
"file": "occamy-1.0-APEX-Compact.gguf",
"bytes": 16538851072,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF/resolve/main/occamy-1.0-APEX-Compact.gguf?download=true"
},
{
"id": "apex-i-compact",
"label": "APEX I-Compact",
"repo": "Accio-Lab/occamy-1.0-APEX-GGUF",
"revision": "7c731908b05f5cbd19002b23d421aa102c0dd0de",
"platform": "gguf",
"description": "Upstream APEX I-Compact mixed precision, directly from BF16 with an independently collected importance matrix; all 256 experts covered in every layer. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-APEX-GGUF occamy-1.0-APEX-I-Compact.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-APEX-I-Compact.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 native qwen2 + NFC: 19/20 authored cases; failed: json_4; subset PPL 8.5074. All 733 tensors finite and 16/16 tokenizer probes. Full calibration coverage; calibrated does not mean better quality. Image, CPU and Mac were not retested.",
"file": "occamy-1.0-APEX-I-Compact.gguf",
"bytes": 16538851360,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF/resolve/main/occamy-1.0-APEX-I-Compact.gguf?download=true"
},
{
"id": "q3_k_m",
"label": "Q3_K_M",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "e8fe5e28e1b1c1f0cd0a39b85b16b631f17ca14e",
"platform": "gguf",
"description": "GGUF language weights for llama.cpp. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-Q3_K_M.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-Q3_K_M.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 CUDA sample validation; published copies use qwen2 metadata. Full quality measurements used the documented qwen35 protocol. CPU and Mac execution were not established by these tests.",
"file": "occamy-1.0-Q3_K_M.gguf",
"bytes": 16764764160,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-Q3_K_M.gguf?download=true"
},
{
"id": "q3_k_l",
"label": "Q3_K_L",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "ba279c80f36eadc81d9cc182d16237abbeafea1e",
"platform": "gguf",
"description": "Standard GGUF recipe, generated directly from BF16 without calibration. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-Q3_K_L.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-Q3_K_L.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "Native qwen2 + NFC checks on B200: 20/20 authored functional cases; subset PPL 9.1110. Full finite-tensor scan and 16/16 NFC tokenizer probes with special parsing. Bounded tests; image, CPU and Mac execution were not tested for this addition.",
"file": "occamy-1.0-Q3_K_L.gguf",
"bytes": 18115329792,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-Q3_K_L.gguf?download=true"
},
{
"id": "iq4_xs",
"label": "IQ4_XS",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "e8fe5e28e1b1c1f0cd0a39b85b16b631f17ca14e",
"platform": "gguf",
"description": "GGUF language weights for llama.cpp. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-IQ4_XS.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-IQ4_XS.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 CUDA sample validation; published copies use qwen2 metadata. Full quality measurements used the documented qwen35 protocol. CPU and Mac execution were not established by these tests.",
"file": "occamy-1.0-IQ4_XS.gguf",
"bytes": 18728777728,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-IQ4_XS.gguf?download=true"
},
{
"id": "q4_0",
"label": "Q4_0",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "ba279c80f36eadc81d9cc182d16237abbeafea1e",
"platform": "gguf",
"description": "Standard GGUF recipe, generated directly from BF16 without calibration. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-Q4_0.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-Q4_0.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "Native qwen2 + NFC checks on B200: 20/20 authored functional cases; subset PPL 8.4754. Full finite-tensor scan and 16/16 NFC tokenizer probes with special parsing. Bounded tests; image, CPU and Mac execution were not tested for this addition.",
"file": "occamy-1.0-Q4_0.gguf",
"bytes": 19715053312,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-Q4_0.gguf?download=true"
},
{
"id": "iq4_nl",
"label": "IQ4_NL",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "e8fe5e28e1b1c1f0cd0a39b85b16b631f17ca14e",
"platform": "gguf",
"description": "GGUF language weights for llama.cpp. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-IQ4_NL.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-IQ4_NL.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 CUDA sample validation; published copies use qwen2 metadata. Full quality measurements used the documented qwen35 protocol. CPU and Mac execution were not established by these tests.",
"file": "occamy-1.0-IQ4_NL.gguf",
"bytes": 19779278848,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-IQ4_NL.gguf?download=true"
},
{
"id": "q4_k_s",
"label": "Q4_K_S",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "e8fe5e28e1b1c1f0cd0a39b85b16b631f17ca14e",
"platform": "gguf",
"description": "GGUF language weights for llama.cpp. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-Q4_K_S.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-Q4_K_S.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 CUDA sample validation; published copies use qwen2 metadata. Full quality measurements used the documented qwen35 protocol. CPU and Mac execution were not established by these tests.",
"file": "occamy-1.0-Q4_K_S.gguf",
"bytes": 19889903616,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-Q4_K_S.gguf?download=true"
},
{
"id": "q4_k_m",
"label": "Q4_K_M",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "e8fe5e28e1b1c1f0cd0a39b85b16b631f17ca14e",
"platform": "gguf",
"description": "GGUF language weights for llama.cpp. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-Q4_K_M.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-Q4_K_M.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 CUDA sample validation; published copies use qwen2 metadata. Full quality measurements used the documented qwen35 protocol. CPU and Mac execution were not established by these tests.",
"file": "occamy-1.0-Q4_K_M.gguf",
"bytes": 21166757696,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-Q4_K_M.gguf?download=true"
},
{
"id": "apex-quality",
"label": "APEX Quality",
"repo": "Accio-Lab/occamy-1.0-APEX-GGUF",
"revision": "7c731908b05f5cbd19002b23d421aa102c0dd0de",
"platform": "gguf",
"description": "Upstream APEX Quality MoE mixed-precision allocation, generated directly from BF16 without an importance matrix. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-APEX-GGUF occamy-1.0-APEX-Quality.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-APEX-Quality.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 native qwen2 + NFC: 19/20 authored functional cases, including executed Python and tool round trips; subset PPL 8.2568. All 733 tensors finite and 16/16 tokenizer probes. Bounded checks; image, CPU and Mac execution were not tested.",
"file": "occamy-1.0-APEX-Quality.gguf",
"bytes": 22819401472,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF/resolve/main/occamy-1.0-APEX-Quality.gguf?download=true"
},
{
"id": "apex-i-quality",
"label": "APEX I-Quality",
"repo": "Accio-Lab/occamy-1.0-APEX-GGUF",
"revision": "7c731908b05f5cbd19002b23d421aa102c0dd0de",
"platform": "gguf",
"description": "Upstream APEX I-Quality mixed precision, directly from BF16 with an independently collected importance matrix; all 256 experts covered in every layer. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-APEX-GGUF occamy-1.0-APEX-I-Quality.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-APEX-I-Quality.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 native qwen2 + NFC: 19/20 authored cases; failed: json_4; subset PPL 8.3743. All 733 tensors finite and 16/16 tokenizer probes. Full calibration coverage; calibrated does not mean better quality. Image, CPU and Mac were not retested.",
"file": "occamy-1.0-APEX-I-Quality.gguf",
"bytes": 22819401760,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF/resolve/main/occamy-1.0-APEX-I-Quality.gguf?download=true"
},
{
"id": "q5_k_s",
"label": "Q5_K_S",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "6210e42ce97bc1b7c7ea58cab597e9b49ed523e5",
"platform": "gguf",
"description": "Standard mixed K-quant GGUF, generated directly from BF16 without calibration. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-Q5_K_S.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-Q5_K_S.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "Native qwen2 + NFC validation on B200: 19/20 authored cases, including executed Python and native tool round trips. Same Chinese JSON failure as BF16. 16/16 NFC tokenizer probes with special-token parsing. Vision, CPU and Mac execution were not tested for this addition.",
"file": "occamy-1.0-Q5_K_S.gguf",
"bytes": 23981283072,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-Q5_K_S.gguf?download=true"
},
{
"id": "q5_0",
"label": "Q5_0",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "ba279c80f36eadc81d9cc182d16237abbeafea1e",
"platform": "gguf",
"description": "Standard GGUF recipe, generated directly from BF16 without calibration. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-Q5_0.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-Q5_0.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "Native qwen2 + NFC checks on B200: 19/20 authored functional cases; subset PPL 8.2284. Full finite-tensor scan and 16/16 NFC tokenizer probes with special parsing. Bounded tests; image, CPU and Mac execution were not tested for this addition.",
"file": "occamy-1.0-Q5_0.gguf",
"bytes": 23981283072,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-Q5_0.gguf?download=true"
},
{
"id": "q5_k_m",
"label": "Q5_K_M",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "e8fe5e28e1b1c1f0cd0a39b85b16b631f17ca14e",
"platform": "gguf",
"description": "GGUF language weights for llama.cpp. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-Q5_K_M.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-Q5_K_M.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 CUDA sample validation; published copies use qwen2 metadata. Full quality measurements used the documented qwen35 protocol. CPU and Mac execution were not established by these tests.",
"file": "occamy-1.0-Q5_K_M.gguf",
"bytes": 24729131008,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-Q5_K_M.gguf?download=true"
},
{
"id": "apex-balanced",
"label": "APEX Balanced",
"repo": "Accio-Lab/occamy-1.0-APEX-GGUF",
"revision": "7c731908b05f5cbd19002b23d421aa102c0dd0de",
"platform": "gguf",
"description": "Upstream APEX Balanced MoE mixed-precision allocation, generated directly from BF16 without an importance matrix. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-APEX-GGUF occamy-1.0-APEX-Balanced.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-APEX-Balanced.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 native qwen2 + NFC: 19/20 authored functional cases, including executed Python and tool round trips; subset PPL 8.2994. All 733 tensors finite and 16/16 tokenizer probes. Bounded checks; image, CPU and Mac execution were not tested.",
"file": "occamy-1.0-APEX-Balanced.gguf",
"bytes": 25335983872,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF/resolve/main/occamy-1.0-APEX-Balanced.gguf?download=true"
},
{
"id": "apex-i-balanced",
"label": "APEX I-Balanced",
"repo": "Accio-Lab/occamy-1.0-APEX-GGUF",
"revision": "7c731908b05f5cbd19002b23d421aa102c0dd0de",
"platform": "gguf",
"description": "Upstream APEX I-Balanced mixed precision, directly from BF16 with an independently collected importance matrix; all 256 experts covered in every layer. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-APEX-GGUF occamy-1.0-APEX-I-Balanced.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-APEX-I-Balanced.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 native qwen2 + NFC: 19/20 authored cases; failed: json_4; subset PPL 8.3482. All 733 tensors finite and 16/16 tokenizer probes. Full calibration coverage; calibrated does not mean better quality. Image, CPU and Mac were not retested.",
"file": "occamy-1.0-APEX-I-Balanced.gguf",
"bytes": 25335984160,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-APEX-GGUF/resolve/main/occamy-1.0-APEX-I-Balanced.gguf?download=true"
},
{
"id": "q6_k",
"label": "Q6_K",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "e8fe5e28e1b1c1f0cd0a39b85b16b631f17ca14e",
"platform": "gguf",
"description": "GGUF language weights for llama.cpp. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-Q6_K.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-Q6_K.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 CUDA sample validation; published copies use qwen2 metadata. Full quality measurements used the documented qwen35 protocol. CPU and Mac execution were not established by these tests.",
"file": "occamy-1.0-Q6_K.gguf",
"bytes": 28514152448,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-Q6_K.gguf?download=true"
},
{
"id": "q8_0",
"label": "Q8_0",
"repo": "Accio-Lab/occamy-1.0-GGUF",
"revision": "e8fe5e28e1b1c1f0cd0a39b85b16b631f17ca14e",
"platform": "gguf",
"description": "GGUF language weights for llama.cpp. Normalize prompt content to NFC.",
"status": "Published",
"command": "hf download Accio-Lab/occamy-1.0-GGUF occamy-1.0-Q8_0.gguf --local-dir ./occamy-gguf\n\nllama-server -m ./occamy-gguf/occamy-1.0-Q8_0.gguf \\\n -ngl 999 -c 8192 -np 1 -fa on --jinja \\\n --host 127.0.0.1 --port 8000 --alias occamy",
"evidence": "B200 CUDA sample validation; published copies use qwen2 metadata. Full quality measurements used the documented qwen35 protocol. CPU and Mac execution were not established by these tests.",
"file": "occamy-1.0-Q8_0.gguf",
"bytes": 36903139456,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF",
"download": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/occamy-1.0-Q8_0.gguf?download=true"
},
{
"id": "bf16",
"label": "BF16",
"repo": "Accio-Lab/occamy-1.0",
"revision": "f9e2771699f14d9c4bbbc0c7b913e58ebe1442d0",
"platform": "nvidia",
"description": "Original full-precision checkpoint. Text and vision.",
"status": "Published",
"command": "vllm serve Accio-Lab/occamy-1.0 \\\n --host 127.0.0.1 --port 8000 \\\n --tensor-parallel-size 1 --max-model-len 8192 \\\n --reasoning-parser qwen3 --enable-auto-tool-choice \\\n --tool-call-parser qwen3_coder",
"evidence": "BF16, FP8 and NVFP4 passed nine-case H200 vLLM compatibility checks. NVFP4 used Marlin W4A16 on H200. SGLang commands and sample evaluations are documented in each model card. Set memory/context limits for your hardware.",
"file": null,
"bytes": 70218761392,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0",
"download": null
},
{
"id": "fp8",
"label": "FP8",
"repo": "Accio-Lab/occamy-1.0-FP8",
"revision": "ac48ca63988ac0f9717bc6c65b11dd77e7cf6ba8",
"platform": "nvidia",
"description": "128 × 128 blockwise FP8 with preserved vision weights.",
"status": "Published",
"command": "SGLANG_JIT_DEEPGEMM_PRECOMPILE=0 python -m sglang.launch_server \\\n --model-path Accio-Lab/occamy-1.0-FP8 --served-model-name occamy \\\n --host 127.0.0.1 --port 18400 --dtype bfloat16 \\\n --context-length 16384 --max-running-requests 4 \\\n --max-total-tokens 65536 --max-mamba-cache-size 16 \\\n --mem-fraction-static 0.60 --attention-backend fa3 \\\n --mm-attention-backend fa3 --mamba-backend triton \\\n --disable-cuda-graph --random-seed 42 \\\n --tool-call-parser qwen3_coder --enable-metrics",
"evidence": "BF16, FP8 and NVFP4 passed nine-case H200 vLLM compatibility checks. NVFP4 used Marlin W4A16 on H200. SGLang commands and sample evaluations are documented in each model card. Set memory/context limits for your hardware.",
"file": null,
"bytes": 36613923840,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-FP8",
"download": null
},
{
"id": "nvfp4",
"label": "NVFP4",
"repo": "Accio-Lab/occamy-1.0-NVFP4",
"revision": "2e53d70641794e441cb038d2499a9e4b8cad306a",
"platform": "nvidia",
"description": "Routed experts in NVFP4; other weights retain BF16. Native FP4 needs compatible hardware and kernels.",
"status": "Published",
"command": "python -m sglang.launch_server \\\n --model-path Accio-Lab/occamy-1.0-NVFP4 \\\n --quantization modelopt_fp4 --dtype bfloat16 \\\n --moe-runner-backend flashinfer_cutlass \\\n --attention-backend triton --mamba-backend triton \\\n --mm-attention-backend sdpa \\\n --tool-call-parser qwen3_coder --reasoning-parser qwen3 \\\n --context-length 2048 --max-total-tokens 4096 \\\n --max-running-requests 1 --max-mamba-cache-size 4 \\\n --disable-cuda-graph --disable-radix-cache --disable-overlap-schedule",
"evidence": "BF16, FP8 and NVFP4 passed nine-case H200 vLLM compatibility checks. NVFP4 used Marlin W4A16 on H200. SGLang commands and sample evaluations are documented in each model card. Set memory/context limits for your hardware.",
"file": null,
"bytes": 23926133080,
"vision": true,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-NVFP4",
"download": null
},
{
"id": "mlx-8-bit",
"label": "MLX 8-bit",
"repo": "Accio-Lab/occamy-1.0-MLX-8bit",
"revision": "7ab5091817d5e62f8790cc8e3f84bb8463d0bfb7",
"platform": "mlx",
"description": "Native affine 8-bit from BF16, group size 64; routers and shared-expert gates use 8-bit. Text only.",
"status": "Mac validation pending",
"command": "python -m pip install \"mlx==0.32.2\" \"mlx-lm==0.31.3\" \"transformers==5.8.1\"\n\npython - <<'PY'\nfrom mlx_lm import load, generate\nmodel, tokenizer = load(\"Accio-Lab/occamy-1.0-MLX-8bit\")\nprompt = tokenizer.apply_chat_template(\n [{\"role\": \"user\", \"content\": \"Compute 2+2. Answer briefly.\"}],\n tokenize=False, add_generation_prompt=True, enable_thinking=False,\n)\nprint(generate(model, tokenizer, prompt=prompt, max_tokens=128))\nPY",
"evidence": "Strict stock reload, all floating values and all dequantized rows checked. Native Linux CUDA: 8/8 authored cached greedy fixtures, finite logits and 2/2 HTTP API checks. Mac Metal and broad quality benchmarks remain pending.",
"file": null,
"bytes": 36828222561,
"vision": false,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-MLX-8bit",
"download": null
},
{
"id": "mlx-6-bit",
"label": "MLX 6-bit",
"repo": "Accio-Lab/occamy-1.0-MLX-6bit",
"revision": "9459767612766f7095a15ab72b2181816caab336",
"platform": "mlx",
"description": "Native affine 6-bit from BF16, group size 64; routers and shared-expert gates use 8-bit. Text only.",
"status": "Mac validation pending",
"command": "python -m pip install \"mlx==0.32.2\" \"mlx-lm==0.31.3\" \"transformers==5.8.1\"\n\npython - <<'PY'\nfrom mlx_lm import load, generate\nmodel, tokenizer = load(\"Accio-Lab/occamy-1.0-MLX-6bit\")\nprompt = tokenizer.apply_chat_template(\n [{\"role\": \"user\", \"content\": \"Compute 2+2. Answer briefly.\"}],\n tokenize=False, add_generation_prompt=True, enable_thinking=False,\n)\nprint(generate(model, tokenizer, prompt=prompt, max_tokens=128))\nPY",
"evidence": "Strict stock reload, all floating values and all dequantized rows checked. Native Linux CUDA: 8/8 authored cached greedy fixtures, finite logits and 2/2 HTTP API checks. Mac Metal and broad quality benchmarks remain pending.",
"file": null,
"bytes": 28168623119,
"vision": false,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-MLX-6bit",
"download": null
},
{
"id": "mlx-5-bit",
"label": "MLX 5-bit",
"repo": "Accio-Lab/occamy-1.0-MLX-5bit",
"revision": "b3a65ecc9be13f0c73b3ea5ae9bc8ae91b207bc3",
"platform": "mlx",
"description": "Native MLX affine weight format, group 64, directly from BF16. Router/shared gates use affine 8-bit. Text only; Apple Metal acceptance is pending.",
"status": "Mac validation pending",
"command": "python -m pip install \"mlx==0.32.2\" \"mlx-lm==0.31.3\" \"transformers==5.8.1\"\n\npython - <<'PY'\nfrom mlx_lm import load, generate\nmodel, tokenizer = load(\"Accio-Lab/occamy-1.0-MLX-5bit\")\nprompt = tokenizer.apply_chat_template(\n [{\"role\": \"user\", \"content\": \"Compute 2+2. Answer briefly.\"}],\n tokenize=False, add_generation_prompt=True, enable_thinking=False,\n)\nprint(generate(model, tokenizer, prompt=prompt, max_tokens=128))\nPY",
"evidence": "Native Linux MLX on B200: 8/8 authored inference fixtures, 2/2 HTTP checks, every stored float and dequantized row checked. Paired subset PPL 8.3075 versus BF16 8.3074; bounded check, no speed ranking. Mac Metal acceptance remains pending.",
"bytes": 23838823629,
"vision": false,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-MLX-5bit",
"download": null
},
{
"id": "mlx-4-bit",
"label": "MLX 4-bit",
"repo": "Accio-Lab/occamy-1.0-MLX-4bit",
"revision": "7a442e16d5700f775f02a844d6e93a68cf95ad58",
"platform": "mlx",
"description": "Native affine 4-bit, group size 64; routers and shared-expert gates use 8-bit. Text only.",
"status": "Mac validation pending",
"command": "python -m pip install \"mlx==0.32.2\" \"mlx-lm==0.31.3\" \"transformers==5.8.1\"\n\npython - <<'PY'\nfrom mlx_lm import load, generate\nmodel, tokenizer = load(\"Accio-Lab/occamy-1.0-MLX-4bit\")\nprompt = tokenizer.apply_chat_template(\n [{\"role\": \"user\", \"content\": \"Compute 2+2. Answer briefly.\"}],\n tokenize=False, add_generation_prompt=True, enable_thinking=False,\n)\nprint(generate(model, tokenizer, prompt=prompt, max_tokens=128))\nPY",
"evidence": "Strict reload, all quantized-row finite checks and one cached greedy generation passed on Linux native MLX CPU. Mac Metal acceptance and broad quality evaluation remain pending.",
"file": null,
"bytes": 19509024201,
"vision": false,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-MLX-4bit",
"download": null
},
{
"id": "mlx-3-bit",
"label": "MLX 3-bit",
"repo": "Accio-Lab/occamy-1.0-MLX-3bit",
"revision": "0349815457382560473322f1322de36f668aefea",
"platform": "mlx",
"description": "Native affine 3-bit, group size 64; routers and shared-expert gates use 8-bit. Text only.",
"status": "Mac validation pending",
"command": "python -m pip install \"mlx==0.32.2\" \"mlx-lm==0.31.3\" \"transformers==5.8.1\"\n\npython - <<'PY'\nfrom mlx_lm import load, generate\nmodel, tokenizer = load(\"Accio-Lab/occamy-1.0-MLX-3bit\")\nprompt = tokenizer.apply_chat_template(\n [{\"role\": \"user\", \"content\": \"Compute 2+2. Answer briefly.\"}],\n tokenize=False, add_generation_prompt=True, enable_thinking=False,\n)\nprint(generate(model, tokenizer, prompt=prompt, max_tokens=128))\nPY",
"evidence": "Strict reload, all quantized-row finite checks and one cached greedy generation passed on Linux native MLX CPU. Mac Metal acceptance and broad quality evaluation remain pending.",
"file": null,
"bytes": 15179224395,
"vision": false,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-MLX-3bit",
"download": null
},
{
"id": "mlx-mxfp8",
"label": "MLX MXFP8",
"repo": "Accio-Lab/occamy-1.0-MLX-mxfp8",
"revision": "7d8347131140ee8d345c8a561f7af657e2569a72",
"platform": "mlx",
"description": "Native MLX mxfp8 weight format, group 32, directly from BF16. Router/shared gates use affine 8-bit. Text only; Apple Metal acceptance is pending.",
"status": "Mac validation pending",
"command": "python -m pip install \"mlx==0.32.2\" \"mlx-lm==0.31.3\" \"transformers==5.8.1\"\n\npython - <<'PY'\nfrom mlx_lm import load, generate\nmodel, tokenizer = load(\"Accio-Lab/occamy-1.0-MLX-mxfp8\")\nprompt = tokenizer.apply_chat_template(\n [{\"role\": \"user\", \"content\": \"Compute 2+2. Answer briefly.\"}],\n tokenize=False, add_generation_prompt=True, enable_thinking=False,\n)\nprint(generate(model, tokenizer, prompt=prompt, max_tokens=128))\nPY",
"evidence": "Native Linux MLX on B200: 8/8 authored inference fixtures, 2/2 HTTP checks, every stored float and dequantized row checked. Paired subset PPL 8.3866 versus BF16 8.3074; bounded check, no speed ranking. Mac Metal acceptance remains pending.",
"bytes": 35745712868,
"vision": false,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-MLX-mxfp8",
"download": null
},
{
"id": "mlx-mxfp4",
"label": "MLX MXFP4",
"repo": "Accio-Lab/occamy-1.0-MLX-mxfp4",
"revision": "4366fb2d80574254e77b677a80fa92215fb63a85",
"platform": "mlx",
"description": "Native MLX mxfp4 weight format, group 32, directly from BF16. Router/shared gates use affine 8-bit. Text only; Apple Metal acceptance is pending.",
"status": "Mac validation pending",
"command": "python -m pip install \"mlx==0.32.2\" \"mlx-lm==0.31.3\" \"transformers==5.8.1\"\n\npython - <<'PY'\nfrom mlx_lm import load, generate\nmodel, tokenizer = load(\"Accio-Lab/occamy-1.0-MLX-mxfp4\")\nprompt = tokenizer.apply_chat_template(\n [{\"role\": \"user\", \"content\": \"Compute 2+2. Answer briefly.\"}],\n tokenize=False, add_generation_prompt=True, enable_thinking=False,\n)\nprint(generate(model, tokenizer, prompt=prompt, max_tokens=128))\nPY",
"evidence": "Native Linux MLX on B200: 8/8 authored inference fixtures, 2/2 HTTP checks, every stored float and dequantized row checked. Paired subset PPL 8.8886 versus BF16 8.3074; bounded check, no speed ranking. Mac Metal acceptance remains pending.",
"bytes": 18426513876,
"vision": false,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-MLX-mxfp4",
"download": null
},
{
"id": "mlx-nvfp4",
"label": "MLX NVFP4",
"repo": "Accio-Lab/occamy-1.0-MLX-nvfp4",
"revision": "475bf75de5b3f1b5f7cce36cecd55fe5d2855dca",
"platform": "mlx",
"description": "Native MLX nvfp4 weight format, group 16, directly from BF16. Router/shared gates use affine 8-bit. Text only; Apple Metal acceptance is pending.",
"status": "Mac validation pending",
"command": "python -m pip install \"mlx==0.32.2\" \"mlx-lm==0.31.3\" \"transformers==5.8.1\"\n\npython - <<'PY'\nfrom mlx_lm import load, generate\nmodel, tokenizer = load(\"Accio-Lab/occamy-1.0-MLX-nvfp4\")\nprompt = tokenizer.apply_chat_template(\n [{\"role\": \"user\", \"content\": \"Compute 2+2. Answer briefly.\"}],\n tokenize=False, add_generation_prompt=True, enable_thinking=False,\n)\nprint(generate(model, tokenizer, prompt=prompt, max_tokens=128))\nPY",
"evidence": "Native Linux MLX on B200: 8/8 authored inference fixtures, 2/2 HTTP checks, every stored float and dequantized row checked. Paired subset PPL 8.9242 versus BF16 8.3074; bounded check, no speed ranking. Mac Metal acceptance remains pending.",
"bytes": 19508964296,
"vision": false,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-MLX-nvfp4",
"download": null
},
{
"id": "mtp-head",
"label": "MTP head",
"repo": "Accio-Lab/occamy-1.0-MTP",
"revision": "2fd68d371e0b9ae5ac1c7d945ae1d40a9c616ec6",
"platform": "research",
"description": "Experimental BF16 draft head. Download a base separately; use the supplied patched SGLang runtime.",
"status": "Experimental",
"command": "hf download Accio-Lab/occamy-1.0-MTP --local-dir ./occamy-mtp\n\n# Follow NATIVE-MTP3.md for the exact source revision,\n# runtime patch, base assembly and native MTP3 launch command.",
"evidence": "Recorded native MTP3 H200 BF16 TP1 greedy checks passed 4/4 cold/warm and 20/20 fixed-prompt token/logprob comparisons. Performance after the final fix is unmeasured. Other precision/backend combinations require separate validation.",
"file": null,
"bytes": 1689283656,
"vision": false,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-MTP",
"download": null
}
],
"collection": "https://huggingface.co/collections/Accio-Lab/occamy-10-6ac00729659e8a86ec11e216",
"mlx_collection": "https://huggingface.co/collections/Accio-Lab/occamy-10-mlx-6ac0072a7c1cdbc5e418e90b",
"space": "https://huggingface.co/spaces/Accio-Lab/Occamy-Explorer",
"base": "Accio-Lab/occamy-1.0",
"project": "https://accio-lab.github.io/occamy/",
"paper": "https://arxiv.org/abs/2609.11977",
"projector": {
"bytes": 899282944,
"url": "https://huggingface.co/Accio-Lab/occamy-1.0-GGUF/resolve/main/mmproj-occamy-1.0-F16.gguf?download=true"
}
}
|