appvoid commited on
Commit
1b191e4
·
verified ·
1 Parent(s): 1b948d5

Upload data.json

Browse files
Files changed (1) hide show
  1. data.json +187 -75
data.json CHANGED
@@ -88,6 +88,46 @@
88
  }
89
  ]
90
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
91
  {
92
  "id": "liquidai",
93
  "name": "LiquidAI",
@@ -162,6 +202,109 @@
162
  }
163
  ]
164
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
165
  {
166
  "id": "tiiuae",
167
  "name": "TII",
@@ -407,7 +550,7 @@
407
  "evaluation_status": "evaluated",
408
  "date_submitted": "2026-11-06",
409
  "date_added": "2026-11-06",
410
- "author_note": "A 144M-parameter dense Boris 2 language-model checkpoint from a 100B-token pretraining run, evaluated with a 2K-token context.",
411
  "comment_source": "dotlabs",
412
  "checkpoint_revision": "f2615d19c87b92b5d64a611030d7bbfb7c1da5b1",
413
  "scores": {
@@ -430,77 +573,6 @@
430
  }
431
  ]
432
  },
433
- {
434
- "id": "huggingface-tb",
435
- "name": "HuggingFaceTB",
436
- "link": "https://huggingface.co/HuggingFaceTB",
437
- "color": "#F5D18C",
438
- "models": [
439
- {
440
- "id": "HuggingFaceTB/SmolLM2-135M",
441
- "name": "SmolLM2-135M",
442
- "parameters": "135M",
443
- "model_card_url": "https://huggingface.co/HuggingFaceTB/SmolLM2-135M",
444
- "model_type": "pretrained",
445
- "evaluation_status": "evaluated",
446
- "date_submitted": "2026-11-04",
447
- "date_added": "2026-11-04",
448
- "author_note": "A small open foundation model designed to make language modeling accessible within a tight compute budget.",
449
- "comment_source": "dotlabs",
450
- "checkpoint_revision": "93efa2f097d58c2a74874c7e644dbc9b0cee75a2",
451
- "scores": {
452
- "ttm": 46.7,
453
- "marc": 5.3125,
454
- "procedural_pile": 0.004785877737822812,
455
- "smolcua": 0.0,
456
- "smolscript": 0.5374255374255374
457
- },
458
- "cognirank_index": 7.125861081745702,
459
- "evaluation": {
460
- "protocol": "cognirank-2026-11-update",
461
- "requested_context": 4096,
462
- "component_protocols": {
463
- "reasoning": "cognirank-v9",
464
- "smolscript": "smolscript-0.5.0"
465
- },
466
- "source_result_sha256": {
467
- "reasoning": "b863c457cb2b5d6f32650f8a55b0cf90df150391aca002c923eef1473ceb054d",
468
- "smolscript": "b4e8f5fa96434823307edee7e14b66fe8b9fc9be048ebcf7c00800ae4474270a"
469
- }
470
- }
471
- },
472
- {
473
- "id": "HuggingFaceTB/SmolLM-135M",
474
- "name": "SmolLM-135M",
475
- "parameters": "135M",
476
- "model_card_url": "https://huggingface.co/HuggingFaceTB/SmolLM-135M",
477
- "model_type": "pretrained",
478
- "evaluation_status": "evaluated",
479
- "date_submitted": "2026-11-06",
480
- "date_added": "2026-11-06",
481
- "author_note": "A compact SmolLM foundation model pretrained on educational and web text, designed for accessible small-model experimentation.",
482
- "comment_source": "dotlabs",
483
- "checkpoint_revision": "1d461723eec654e65efdc40cf49301c89c0c92f4",
484
- "scores": {
485
- "smolcua": 0.0,
486
- "marc": 1.5625,
487
- "procedural_pile": 0.007275037924876215,
488
- "ttm": 47.3,
489
- "smolscript": 0.5374255374255374
490
- },
491
- "cognirank_index": 6.412700904942976,
492
- "evaluation": {
493
- "protocol": "cognirank2-v5",
494
- "requested_context": 2048,
495
- "component_protocols": {
496
- "smolcua": "smolcua-v1.4-three-passes",
497
- "smolcua_scoring": "v2-outer-whitespace-insensitive"
498
- },
499
- "smolcua_dataset_sha256": "74fb0a12a8dd75cbd8675d76e8bcf4a16e1b424668a195f4ff9a31da2becf065"
500
- }
501
- }
502
- ]
503
- },
504
  {
505
  "id": "axiomic-labs",
506
  "name": "AxiomicLabs",
@@ -745,7 +817,7 @@
745
  "evaluation_status": "evaluated",
746
  "date_submitted": "2026-11-07",
747
  "date_added": "2026-11-07",
748
- "author_note": "A 97.7M-parameter hybrid recurrent model combining Gated DeltaNet-2 and full attention, evaluated with a 4,096-token context extension beyond its declared 2,048-token window.",
749
  "comment_source": "dotlabs",
750
  "checkpoint_revision": "bd2dd851bcb14d0bff5b3a78ca58246b99c441f4",
751
  "scores": {
@@ -791,7 +863,7 @@
791
  "evaluation_status": "evaluated",
792
  "date_submitted": "2026-11-07",
793
  "date_added": "2026-11-07",
794
- "author_note": "A 49.3M-parameter language model pretrained from scratch, evaluated with a 4,096-token context extension beyond its declared 1,024-token window.",
795
  "comment_source": "dotlabs",
796
  "checkpoint_revision": "3f6e4f20bbf6d69dc43ab95ef9376925de8a3160",
797
  "scores": {
@@ -822,6 +894,46 @@
822
  }
823
  ]
824
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
825
  {
826
  "id": "algodriveai",
827
  "name": "AlgoDriveAI",
@@ -837,7 +949,7 @@
837
  "evaluation_status": "evaluated",
838
  "date_submitted": "2026-11-07",
839
  "date_added": "2026-11-07",
840
- "author_note": "A 76.2M-parameter math-focused base language model trained by AlgoDriveAI, evaluated with a 4,096-token context extension beyond its declared 2,048-token window.",
841
  "comment_source": "dotlabs",
842
  "checkpoint_revision": "4b3de728668c6159529b63b2dc63a35d97095046",
843
  "scores": {
 
88
  }
89
  ]
90
  },
91
+ {
92
+ "id": "qwen",
93
+ "name": "Qwen",
94
+ "link": "https://huggingface.co/Qwen",
95
+ "color": "#C3ADE8",
96
+ "models": [
97
+ {
98
+ "id": "Qwen/Qwen2.5-Coder-0.5B-Instruct",
99
+ "name": "Qwen2.5-Coder-0.5B-Instruct",
100
+ "parameters": "0.5B",
101
+ "model_card_url": "https://huggingface.co/Qwen/Qwen2.5-Coder-0.5B-Instruct",
102
+ "model_type": "creator_posttrained",
103
+ "evaluation_status": "evaluated",
104
+ "date_submitted": "2026-11-07",
105
+ "date_added": "2026-11-07",
106
+ "author_note": "An instruction-tuned Qwen2.5 Coder model evaluated on CogniRank reasoning, agentic, and programming benchmarks.",
107
+ "comment_source": "dotlabs",
108
+ "checkpoint_revision": "ea3f2471cf1b1f0db85067f1ef93848e38e88c25",
109
+ "scores": {
110
+ "ttm": 47.35,
111
+ "marc": 7.1875,
112
+ "procedural_pile": 13.450502730198362,
113
+ "smolcua": 6.7121,
114
+ "smolscript": 17.558830058830058
115
+ },
116
+ "cognirank_index": 15.53334974443935,
117
+ "evaluation": {
118
+ "protocol": "cognirank2-v5-context-4096",
119
+ "requested_context": 4096,
120
+ "component_protocols": {
121
+ "reasoning": "cognirank2-v5-context-4096",
122
+ "smolcua": "SmolCUA v1.4",
123
+ "smolcua_subset": "smolcua-v1.4-three-passes",
124
+ "smolscript": "smolscript-full-350"
125
+ },
126
+ "source_file_sha256": "b5eed5fc69303033e8c364aa0de66f5f2169cc5e482ab65327bfecdd32b3dfba"
127
+ }
128
+ }
129
+ ]
130
+ },
131
  {
132
  "id": "liquidai",
133
  "name": "LiquidAI",
 
202
  }
203
  ]
204
  },
205
+ {
206
+ "id": "huggingface-tb",
207
+ "name": "HuggingFaceTB",
208
+ "link": "https://huggingface.co/HuggingFaceTB",
209
+ "color": "#F5D18C",
210
+ "models": [
211
+ {
212
+ "id": "HuggingFaceTB/SmolLM2-360M-Instruct",
213
+ "name": "SmolLM2-360M-Instruct",
214
+ "parameters": "360M",
215
+ "model_card_url": "https://huggingface.co/HuggingFaceTB/SmolLM2-360M-Instruct",
216
+ "model_type": "creator_posttrained",
217
+ "evaluation_status": "evaluated",
218
+ "date_submitted": "2026-11-07",
219
+ "date_added": "2026-11-07",
220
+ "author_note": "An instruction-tuned SmolLM2 checkpoint from its original creators, evaluated across all five CogniRank benchmarks.",
221
+ "comment_source": "dotlabs",
222
+ "checkpoint_revision": "a10cc1512eabd3dde888204e902eca88bddb4951",
223
+ "scores": {
224
+ "ttm": 53.1,
225
+ "marc": 11.5625,
226
+ "procedural_pile": 8.35849782325648,
227
+ "smolcua": 0.2298,
228
+ "smolscript": 4.436859436859438
229
+ },
230
+ "cognirank_index": 11.846689323846936,
231
+ "evaluation": {
232
+ "protocol": "cognirank2-v5-context-4096",
233
+ "requested_context": 4096,
234
+ "component_protocols": {
235
+ "reasoning": "cognirank2-v5-context-4096",
236
+ "smolcua": "SmolCUA v1.4",
237
+ "smolcua_subset": "smolcua-v1.4-three-passes",
238
+ "smolscript": "smolscript-full-350"
239
+ },
240
+ "source_file_sha256": "3d5b2d29bacd10635c2e47bd97f7520503c865296141ada7162f5d7b851c76de"
241
+ }
242
+ },
243
+ {
244
+ "id": "HuggingFaceTB/SmolLM2-135M",
245
+ "name": "SmolLM2-135M",
246
+ "parameters": "135M",
247
+ "model_card_url": "https://huggingface.co/HuggingFaceTB/SmolLM2-135M",
248
+ "model_type": "pretrained",
249
+ "evaluation_status": "evaluated",
250
+ "date_submitted": "2026-11-04",
251
+ "date_added": "2026-11-04",
252
+ "author_note": "A small open foundation model designed to make language modeling accessible within a tight compute budget.",
253
+ "comment_source": "dotlabs",
254
+ "checkpoint_revision": "93efa2f097d58c2a74874c7e644dbc9b0cee75a2",
255
+ "scores": {
256
+ "ttm": 46.7,
257
+ "marc": 5.3125,
258
+ "procedural_pile": 0.004785877737822812,
259
+ "smolcua": 0.0,
260
+ "smolscript": 0.5374255374255374
261
+ },
262
+ "cognirank_index": 7.125861081745702,
263
+ "evaluation": {
264
+ "protocol": "cognirank-2026-11-update",
265
+ "requested_context": 4096,
266
+ "component_protocols": {
267
+ "reasoning": "cognirank-v9",
268
+ "smolscript": "smolscript-0.5.0"
269
+ },
270
+ "source_result_sha256": {
271
+ "reasoning": "b863c457cb2b5d6f32650f8a55b0cf90df150391aca002c923eef1473ceb054d",
272
+ "smolscript": "b4e8f5fa96434823307edee7e14b66fe8b9fc9be048ebcf7c00800ae4474270a"
273
+ }
274
+ }
275
+ },
276
+ {
277
+ "id": "HuggingFaceTB/SmolLM-135M",
278
+ "name": "SmolLM-135M",
279
+ "parameters": "135M",
280
+ "model_card_url": "https://huggingface.co/HuggingFaceTB/SmolLM-135M",
281
+ "model_type": "pretrained",
282
+ "evaluation_status": "evaluated",
283
+ "date_submitted": "2026-11-06",
284
+ "date_added": "2026-11-06",
285
+ "author_note": "A compact SmolLM foundation model pretrained on educational and web text, designed for accessible small-model experimentation.",
286
+ "comment_source": "dotlabs",
287
+ "checkpoint_revision": "1d461723eec654e65efdc40cf49301c89c0c92f4",
288
+ "scores": {
289
+ "smolcua": 0.0,
290
+ "marc": 1.5625,
291
+ "procedural_pile": 0.007275037924876215,
292
+ "ttm": 47.3,
293
+ "smolscript": 0.5374255374255374
294
+ },
295
+ "cognirank_index": 6.412700904942976,
296
+ "evaluation": {
297
+ "protocol": "cognirank2-v5",
298
+ "requested_context": 2048,
299
+ "component_protocols": {
300
+ "smolcua": "smolcua-v1.4-three-passes",
301
+ "smolcua_scoring": "v2-outer-whitespace-insensitive"
302
+ },
303
+ "smolcua_dataset_sha256": "74fb0a12a8dd75cbd8675d76e8bcf4a16e1b424668a195f4ff9a31da2becf065"
304
+ }
305
+ }
306
+ ]
307
+ },
308
  {
309
  "id": "tiiuae",
310
  "name": "TII",
 
550
  "evaluation_status": "evaluated",
551
  "date_submitted": "2026-11-06",
552
  "date_added": "2026-11-06",
553
+ "author_note": "A 144M-parameter dense Boris 2 language-model checkpoint from a 100B-token pretraining run.",
554
  "comment_source": "dotlabs",
555
  "checkpoint_revision": "f2615d19c87b92b5d64a611030d7bbfb7c1da5b1",
556
  "scores": {
 
573
  }
574
  ]
575
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
576
  {
577
  "id": "axiomic-labs",
578
  "name": "AxiomicLabs",
 
817
  "evaluation_status": "evaluated",
818
  "date_submitted": "2026-11-07",
819
  "date_added": "2026-11-07",
820
+ "author_note": "A 97.7M-parameter hybrid recurrent model combining Gated DeltaNet-2 and full attention.",
821
  "comment_source": "dotlabs",
822
  "checkpoint_revision": "bd2dd851bcb14d0bff5b3a78ca58246b99c441f4",
823
  "scores": {
 
863
  "evaluation_status": "evaluated",
864
  "date_submitted": "2026-11-07",
865
  "date_added": "2026-11-07",
866
+ "author_note": "A 49.3M-parameter language model pretrained from scratch.",
867
  "comment_source": "dotlabs",
868
  "checkpoint_revision": "3f6e4f20bbf6d69dc43ab95ef9376925de8a3160",
869
  "scores": {
 
894
  }
895
  ]
896
  },
897
+ {
898
+ "id": "meta",
899
+ "name": "Meta",
900
+ "link": "https://huggingface.co/facebook",
901
+ "color": "#B5CCDF",
902
+ "models": [
903
+ {
904
+ "id": "facebook/MobileLLM-R1-140M",
905
+ "name": "MobileLLM-R1-140M",
906
+ "parameters": "140M",
907
+ "model_card_url": "https://huggingface.co/facebook/MobileLLM-R1-140M",
908
+ "model_type": "pretrained",
909
+ "evaluation_status": "evaluated",
910
+ "date_submitted": "2026-11-07",
911
+ "date_added": "2026-11-07",
912
+ "author_note": "A 140M-parameter MobileLLM-R1 checkpoint evaluated across the five CogniRank benchmarks.",
913
+ "comment_source": "dotlabs",
914
+ "checkpoint_revision": "df21db6bd739b9b82c2caf2647b724ed63b8521e",
915
+ "scores": {
916
+ "ttm": 38.7,
917
+ "marc": 3.125,
918
+ "procedural_pile": 0.017009287573051233,
919
+ "smolcua": 0.0,
920
+ "smolscript": 0.0
921
+ },
922
+ "cognirank_index": 5.549896692120643,
923
+ "evaluation": {
924
+ "protocol": "cognirank2-v5-context-4096",
925
+ "requested_context": 4096,
926
+ "component_protocols": {
927
+ "reasoning": "cognirank2-v5-context-4096",
928
+ "smolcua": "SmolCUA v1.4",
929
+ "smolcua_subset": "smolcua-v1.4-three-passes",
930
+ "smolscript": "smolscript-full-350"
931
+ },
932
+ "source_file_sha256": "34d4f01fded5e7bd8b49e93c16a59d123bf0144b6c208a9d02e67737ff46d758"
933
+ }
934
+ }
935
+ ]
936
+ },
937
  {
938
  "id": "algodriveai",
939
  "name": "AlgoDriveAI",
 
949
  "evaluation_status": "evaluated",
950
  "date_submitted": "2026-11-07",
951
  "date_added": "2026-11-07",
952
+ "author_note": "A 76.2M-parameter math-focused base language model trained by AlgoDriveAI.",
953
  "comment_source": "dotlabs",
954
  "checkpoint_revision": "4b3de728668c6159529b63b2dc63a35d97095046",
955
  "scores": {