jev-compatible-server 0.1.0__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- jev_compatible_server/__init__.py +5 -0
- jev_compatible_server/app.py +144 -0
- jev_compatible_server/backends.py +320 -0
- jev_compatible_server/batching.py +68 -0
- jev_compatible_server/bosun.py +229 -0
- jev_compatible_server/causal_options.py +228 -0
- jev_compatible_server/classifier_adapters.py +472 -0
- jev_compatible_server/cross_encoder.py +71 -0
- jev_compatible_server/custom_heads.py +704 -0
- jev_compatible_server/encoder_decoder.py +630 -0
- jev_compatible_server/gliner2.py +40 -0
- jev_compatible_server/hidden_state_probe.py +384 -0
- jev_compatible_server/laya.py +135 -0
- jev_compatible_server/native_systemone.py +248 -0
- jev_compatible_server/protocol.py +104 -0
- jev_compatible_server/public-models.json +981 -0
- jev_compatible_server/registry.py +283 -0
- jev_compatible_server/runtime.py +237 -0
- jev_compatible_server/sequence_classifier.py +219 -0
- jev_compatible_server-0.1.0.dist-info/METADATA +157 -0
- jev_compatible_server-0.1.0.dist-info/RECORD +23 -0
- jev_compatible_server-0.1.0.dist-info/WHEEL +4 -0
- jev_compatible_server-0.1.0.dist-info/entry_points.txt +2 -0
|
@@ -0,0 +1,981 @@
|
|
|
1
|
+
{
|
|
2
|
+
"default": "kev-4b",
|
|
3
|
+
"recipes": {
|
|
4
|
+
"bosun-decision-tokens-v1": {
|
|
5
|
+
"decision": {
|
|
6
|
+
"readout": "bosun_decision_tokens",
|
|
7
|
+
"question_types": [
|
|
8
|
+
"choice",
|
|
9
|
+
"score",
|
|
10
|
+
"noul"
|
|
11
|
+
],
|
|
12
|
+
"dtype": "bfloat16",
|
|
13
|
+
"seed": 0,
|
|
14
|
+
"loader": {
|
|
15
|
+
"dtype": "bfloat16",
|
|
16
|
+
"device_map": "auto",
|
|
17
|
+
"attn_implementation": "sdpa"
|
|
18
|
+
}
|
|
19
|
+
}
|
|
20
|
+
},
|
|
21
|
+
"hidden-state-probe-v1": {
|
|
22
|
+
"decision": {
|
|
23
|
+
"readout": "hidden_state_probe",
|
|
24
|
+
"hidden": {
|
|
25
|
+
"token": "last_non_pad"
|
|
26
|
+
},
|
|
27
|
+
"dtype": "bfloat16",
|
|
28
|
+
"batch_size": 8
|
|
29
|
+
}
|
|
30
|
+
},
|
|
31
|
+
"hidden-state-candidate-verifier-v1": {
|
|
32
|
+
"decision": {
|
|
33
|
+
"readout": "hidden_state_probe",
|
|
34
|
+
"question_types": [
|
|
35
|
+
"choice",
|
|
36
|
+
"score",
|
|
37
|
+
"noul"
|
|
38
|
+
],
|
|
39
|
+
"hidden": {
|
|
40
|
+
"token": "last_non_pad"
|
|
41
|
+
},
|
|
42
|
+
"dtype": "bfloat16",
|
|
43
|
+
"batch_size": 8,
|
|
44
|
+
"instruction_adapters": {
|
|
45
|
+
"choice": "",
|
|
46
|
+
"score": "",
|
|
47
|
+
"noul": ""
|
|
48
|
+
},
|
|
49
|
+
"candidates": {
|
|
50
|
+
"choice": {
|
|
51
|
+
"criterion_template": "{key}: {criterion}",
|
|
52
|
+
"empty_criterion_template": "{key}"
|
|
53
|
+
},
|
|
54
|
+
"score": {
|
|
55
|
+
"criterion_template": "{criterion}"
|
|
56
|
+
},
|
|
57
|
+
"noul": {
|
|
58
|
+
"no_criteria_options": {
|
|
59
|
+
"true": "yes",
|
|
60
|
+
"false": "no"
|
|
61
|
+
},
|
|
62
|
+
"criterion_template": "{criterion}"
|
|
63
|
+
}
|
|
64
|
+
},
|
|
65
|
+
"aggregation": {
|
|
66
|
+
"choice_temperature": 1.0,
|
|
67
|
+
"score_temperature": 1.0,
|
|
68
|
+
"noul_a": 1.0,
|
|
69
|
+
"noul_b": 0.0,
|
|
70
|
+
"noul_mode": "true_false_softmax",
|
|
71
|
+
"confidence": "normalized_entropy"
|
|
72
|
+
}
|
|
73
|
+
}
|
|
74
|
+
},
|
|
75
|
+
"encoder-decoder-yes-no-margin-v1": {
|
|
76
|
+
"decision": {
|
|
77
|
+
"readout": "encoder_decoder_margin",
|
|
78
|
+
"labels": {
|
|
79
|
+
"positive": "yes",
|
|
80
|
+
"negative": "no"
|
|
81
|
+
},
|
|
82
|
+
"encoder": {
|
|
83
|
+
"document_template": "<Document>: {document}",
|
|
84
|
+
"pooling": "mean_chunks",
|
|
85
|
+
"chunk_size": 4
|
|
86
|
+
},
|
|
87
|
+
"decoder": {
|
|
88
|
+
"template": "<bos><start_of_turn>user\nJudge whether the Document meets the requirements based on the Query and the Instruct provided. Note that the answer can only be \"yes\" or \"no\".\n\n<Instruct>: {instruction}\n<Query>: {query}<end_of_turn>\n<start_of_turn>model\n\n\n\n",
|
|
89
|
+
"pad_to_multiple_of": 8
|
|
90
|
+
},
|
|
91
|
+
"instruction_adapters": {
|
|
92
|
+
"choice": "Evaluate whether the candidate option described in the Document is an appropriate answer to the question above, given the information in the Query. Judge the candidate by its stated description, not merely by topical relevance.",
|
|
93
|
+
"score": "Evaluate whether the level described in the Document accurately characterizes the Query with respect to the assessment above. Judge this level by its stated criteria, not merely by topical relevance.",
|
|
94
|
+
"noul": "Given the Query, evaluate whether the candidate criterion in the Document correctly describes the answer to the question above. Answer yes if this candidate criterion is satisfied, otherwise no."
|
|
95
|
+
},
|
|
96
|
+
"candidates": {
|
|
97
|
+
"choice": {
|
|
98
|
+
"criterion_template": "{key}: {criterion}",
|
|
99
|
+
"empty_criterion_template": "{key}"
|
|
100
|
+
},
|
|
101
|
+
"score": {
|
|
102
|
+
"criterion_template": "{criterion}"
|
|
103
|
+
},
|
|
104
|
+
"noul": {
|
|
105
|
+
"no_criteria_template": "{state}",
|
|
106
|
+
"criterion_template": "{criterion}"
|
|
107
|
+
}
|
|
108
|
+
},
|
|
109
|
+
"aggregation": {
|
|
110
|
+
"choice_temperature": 1.0,
|
|
111
|
+
"score_temperature": 1.0,
|
|
112
|
+
"noul_a": 1.0,
|
|
113
|
+
"noul_b": 0.0,
|
|
114
|
+
"confidence": "normalized_entropy"
|
|
115
|
+
},
|
|
116
|
+
"limits": {
|
|
117
|
+
"query": 512,
|
|
118
|
+
"document": 1024,
|
|
119
|
+
"decoder": 2048
|
|
120
|
+
},
|
|
121
|
+
"batch_size": 4,
|
|
122
|
+
"cache": {
|
|
123
|
+
"max_mib": 256
|
|
124
|
+
}
|
|
125
|
+
}
|
|
126
|
+
},
|
|
127
|
+
"sequence-classifier-native-scalar-v1": {
|
|
128
|
+
"decision": {
|
|
129
|
+
"readout": "sequence_classifier_margin",
|
|
130
|
+
"loader": {
|
|
131
|
+
"num_labels": 1
|
|
132
|
+
},
|
|
133
|
+
"input": {
|
|
134
|
+
"state_template": "State:\n{state}",
|
|
135
|
+
"tail_template": "\n\nQuestion:\n{instructions}\n\nOption:\n{option}",
|
|
136
|
+
"truncation": "preserve_tail",
|
|
137
|
+
"max_length": 512
|
|
138
|
+
},
|
|
139
|
+
"instruction_adapters": {
|
|
140
|
+
"choice": "",
|
|
141
|
+
"score": "",
|
|
142
|
+
"noul": ""
|
|
143
|
+
},
|
|
144
|
+
"candidates": {
|
|
145
|
+
"choice": {
|
|
146
|
+
"criterion_template": "{key}: {criterion}",
|
|
147
|
+
"empty_criterion_template": "{key}"
|
|
148
|
+
},
|
|
149
|
+
"score": {
|
|
150
|
+
"criterion_template": "{criterion}"
|
|
151
|
+
},
|
|
152
|
+
"noul": {
|
|
153
|
+
"no_criteria_options": {
|
|
154
|
+
"true": "yes",
|
|
155
|
+
"false": "no"
|
|
156
|
+
},
|
|
157
|
+
"criterion_template": "{criterion}"
|
|
158
|
+
}
|
|
159
|
+
},
|
|
160
|
+
"aggregation": {
|
|
161
|
+
"choice_temperature": 1.0,
|
|
162
|
+
"score_temperature": 1.0,
|
|
163
|
+
"noul_a": 1.0,
|
|
164
|
+
"noul_b": 0.0,
|
|
165
|
+
"confidence": "normalized_entropy"
|
|
166
|
+
},
|
|
167
|
+
"batch_size": 32
|
|
168
|
+
}
|
|
169
|
+
},
|
|
170
|
+
"cross-encoder-reranker-v1": {
|
|
171
|
+
"decision": {
|
|
172
|
+
"readout": "cross_encoder_margin",
|
|
173
|
+
"max_length": 512,
|
|
174
|
+
"batch_size": 16,
|
|
175
|
+
"instruction_adapters": {
|
|
176
|
+
"choice": "",
|
|
177
|
+
"score": "",
|
|
178
|
+
"noul": ""
|
|
179
|
+
},
|
|
180
|
+
"candidates": {
|
|
181
|
+
"choice": {
|
|
182
|
+
"criterion_template": "{key}: {criterion}",
|
|
183
|
+
"empty_criterion_template": "{key}"
|
|
184
|
+
},
|
|
185
|
+
"score": {
|
|
186
|
+
"criterion_template": "{criterion}"
|
|
187
|
+
},
|
|
188
|
+
"noul": {
|
|
189
|
+
"no_criteria_options": {
|
|
190
|
+
"true": "yes",
|
|
191
|
+
"false": "no"
|
|
192
|
+
},
|
|
193
|
+
"criterion_template": "{criterion}"
|
|
194
|
+
}
|
|
195
|
+
},
|
|
196
|
+
"aggregation": {
|
|
197
|
+
"choice_temperature": 1.0,
|
|
198
|
+
"score_temperature": 1.0,
|
|
199
|
+
"noul_a": 1.0,
|
|
200
|
+
"noul_b": 0.0,
|
|
201
|
+
"confidence": "normalized_entropy"
|
|
202
|
+
}
|
|
203
|
+
}
|
|
204
|
+
},
|
|
205
|
+
"sequence-classifier-scalar-options-v1": {
|
|
206
|
+
"decision": {
|
|
207
|
+
"readout": "sequence_classifier_margin",
|
|
208
|
+
"loader": {
|
|
209
|
+
"base_model": "Qwen/Qwen3.5-4B-Base",
|
|
210
|
+
"adapter": "auto",
|
|
211
|
+
"num_labels": 1
|
|
212
|
+
},
|
|
213
|
+
"input": {
|
|
214
|
+
"state_template": "State:\n{state}",
|
|
215
|
+
"tail_template": "\n\nQuestion:\n{instructions}\n\nOption:\n{option}",
|
|
216
|
+
"truncation": "preserve_tail",
|
|
217
|
+
"max_length": 384
|
|
218
|
+
},
|
|
219
|
+
"instruction_adapters": {
|
|
220
|
+
"choice": "",
|
|
221
|
+
"score": "",
|
|
222
|
+
"noul": ""
|
|
223
|
+
},
|
|
224
|
+
"candidates": {
|
|
225
|
+
"choice": {
|
|
226
|
+
"criterion_template": "{key}: {criterion}",
|
|
227
|
+
"empty_criterion_template": "{key}"
|
|
228
|
+
},
|
|
229
|
+
"score": {
|
|
230
|
+
"criterion_template": "{criterion}"
|
|
231
|
+
},
|
|
232
|
+
"noul": {
|
|
233
|
+
"no_criteria_options": {
|
|
234
|
+
"true": "yes",
|
|
235
|
+
"false": "no"
|
|
236
|
+
},
|
|
237
|
+
"criterion_template": "{criterion}"
|
|
238
|
+
}
|
|
239
|
+
},
|
|
240
|
+
"aggregation": {
|
|
241
|
+
"choice_temperature": 1.75,
|
|
242
|
+
"score_temperature": 1.75,
|
|
243
|
+
"noul_a": 0.5714285714285714,
|
|
244
|
+
"noul_b": 0.0,
|
|
245
|
+
"noul_mode": "true_false_softmax",
|
|
246
|
+
"confidence": "max_probability"
|
|
247
|
+
},
|
|
248
|
+
"batch_size": 48
|
|
249
|
+
}
|
|
250
|
+
}
|
|
251
|
+
},
|
|
252
|
+
"models": {
|
|
253
|
+
"bosun-v3.1-0.6b": {
|
|
254
|
+
"backend": "transformers",
|
|
255
|
+
"model": "Hanno-Labs/bosun-v3.1-0.6b",
|
|
256
|
+
"recipe": "bosun-decision-tokens-v1",
|
|
257
|
+
"description": "Bosun 3.1 0.6B native typed-decision tokens",
|
|
258
|
+
"support_status": "supported",
|
|
259
|
+
"config": {
|
|
260
|
+
"decision": {
|
|
261
|
+
"loader": {
|
|
262
|
+
"revision": "87cea31da06559b0f6509f7b45e2724c38e45c03"
|
|
263
|
+
}
|
|
264
|
+
}
|
|
265
|
+
}
|
|
266
|
+
},
|
|
267
|
+
"bosun-v3.1-1.7b": {
|
|
268
|
+
"backend": "transformers",
|
|
269
|
+
"model": "Hanno-Labs/bosun-v3.1-1.7b",
|
|
270
|
+
"recipe": "bosun-decision-tokens-v1",
|
|
271
|
+
"description": "Bosun 3.1 1.7B native typed-decision tokens",
|
|
272
|
+
"support_status": "supported",
|
|
273
|
+
"config": {
|
|
274
|
+
"decision": {
|
|
275
|
+
"loader": {
|
|
276
|
+
"revision": "8c715a360af4f2d93ce4b7aafd85529f10e760f7"
|
|
277
|
+
}
|
|
278
|
+
}
|
|
279
|
+
}
|
|
280
|
+
},
|
|
281
|
+
"kev-0.5b": {
|
|
282
|
+
"backend": "transformers",
|
|
283
|
+
"model": "jaredpalmer/kev-0.5b",
|
|
284
|
+
"description": "Kev prototype, Qwen2.5-0.5B plus pointer head",
|
|
285
|
+
"support_status": "supported",
|
|
286
|
+
"config": {
|
|
287
|
+
"decision.readout": "pointer_head",
|
|
288
|
+
"decision.backbone": "Qwen/Qwen2.5-0.5B",
|
|
289
|
+
"decision.adapter": "auto",
|
|
290
|
+
"decision.head_path": "head.pt",
|
|
291
|
+
"decision.head_repo": "jaredpalmer/kev-0.5b",
|
|
292
|
+
"decision.packing": {
|
|
293
|
+
"state_token": "<|fim_prefix|>",
|
|
294
|
+
"question_token": "<|fim_middle|>",
|
|
295
|
+
"option_start": "<|box_start|>",
|
|
296
|
+
"option_end": "<|box_end|>",
|
|
297
|
+
"decision_token": "<|fim_suffix|>"
|
|
298
|
+
}
|
|
299
|
+
}
|
|
300
|
+
},
|
|
301
|
+
"kev-0.6b": {
|
|
302
|
+
"backend": "transformers",
|
|
303
|
+
"model": "jaredpalmer/kev-0.6b",
|
|
304
|
+
"description": "Kev Qwen3-0.6B pointer-head model",
|
|
305
|
+
"support_status": "supported",
|
|
306
|
+
"config": {
|
|
307
|
+
"decision.readout": "pointer_head",
|
|
308
|
+
"decision.backbone": "Qwen/Qwen3-0.6B-Base",
|
|
309
|
+
"decision.adapter": "auto",
|
|
310
|
+
"decision.head_path": "head.pt",
|
|
311
|
+
"decision.head_repo": "jaredpalmer/kev-0.6b",
|
|
312
|
+
"decision.packing": {
|
|
313
|
+
"state_token": "<|fim_prefix|>",
|
|
314
|
+
"question_token": "<|fim_middle|>",
|
|
315
|
+
"option_start": "<|box_start|>",
|
|
316
|
+
"option_end": "<|box_end|>",
|
|
317
|
+
"decision_token": "<|fim_suffix|>"
|
|
318
|
+
}
|
|
319
|
+
}
|
|
320
|
+
},
|
|
321
|
+
"kev-4b": {
|
|
322
|
+
"backend": "transformers",
|
|
323
|
+
"model": "jaredpalmer/kev-4b",
|
|
324
|
+
"description": "Kev Qwen3.5-4B pointer-head model",
|
|
325
|
+
"support_status": "supported",
|
|
326
|
+
"config": {
|
|
327
|
+
"decision.readout": "pointer_head",
|
|
328
|
+
"decision.backbone": "Qwen/Qwen3.5-4B-Base",
|
|
329
|
+
"decision.adapter": "auto",
|
|
330
|
+
"decision.head_path": "head.pt",
|
|
331
|
+
"decision.head_repo": "jaredpalmer/kev-4b",
|
|
332
|
+
"decision.packing": {
|
|
333
|
+
"state_token": "<|fim_prefix|>",
|
|
334
|
+
"question_token": "<|fim_middle|>",
|
|
335
|
+
"option_start": "<|box_start|>",
|
|
336
|
+
"option_end": "<|box_end|>",
|
|
337
|
+
"decision_token": "<|fim_suffix|>"
|
|
338
|
+
}
|
|
339
|
+
}
|
|
340
|
+
},
|
|
341
|
+
"kev-8b": {
|
|
342
|
+
"backend": "transformers",
|
|
343
|
+
"model": "jaredpalmer/kev-8b",
|
|
344
|
+
"description": "Kev Qwen3-8B pointer-head model",
|
|
345
|
+
"support_status": "supported",
|
|
346
|
+
"config": {
|
|
347
|
+
"decision.readout": "pointer_head",
|
|
348
|
+
"decision.backbone": "Qwen/Qwen3-8B-Base",
|
|
349
|
+
"decision.adapter": "auto",
|
|
350
|
+
"decision.head_path": "head.pt",
|
|
351
|
+
"decision.head_repo": "jaredpalmer/kev-8b",
|
|
352
|
+
"decision.packing": {
|
|
353
|
+
"state_token": "<|fim_prefix|>",
|
|
354
|
+
"question_token": "<|fim_middle|>",
|
|
355
|
+
"option_start": "<|box_start|>",
|
|
356
|
+
"option_end": "<|box_end|>",
|
|
357
|
+
"decision_token": "<|fim_suffix|>"
|
|
358
|
+
}
|
|
359
|
+
}
|
|
360
|
+
},
|
|
361
|
+
"kalm-jev-nano": {
|
|
362
|
+
"backend": "transformers",
|
|
363
|
+
"model": "KaLM-Embedding/KaLM-Reranker-V1-Nano-R2",
|
|
364
|
+
"recipe": "encoder-decoder-yes-no-margin-v1",
|
|
365
|
+
"description": "Nano encoder-decoder yes/no-margin decision model",
|
|
366
|
+
"support_status": "supported"
|
|
367
|
+
},
|
|
368
|
+
"kalm-jev-small": {
|
|
369
|
+
"backend": "transformers",
|
|
370
|
+
"model": "KaLM-Embedding/KaLM-Reranker-V1-Small-R2",
|
|
371
|
+
"recipe": "encoder-decoder-yes-no-margin-v1",
|
|
372
|
+
"description": "Small encoder-decoder yes/no-margin decision model",
|
|
373
|
+
"support_status": "supported"
|
|
374
|
+
},
|
|
375
|
+
"kalm-jev-large": {
|
|
376
|
+
"backend": "transformers",
|
|
377
|
+
"model": "KaLM-Embedding/KaLM-Reranker-V1-Large-R2",
|
|
378
|
+
"recipe": "encoder-decoder-yes-no-margin-v1",
|
|
379
|
+
"description": "Large encoder-decoder yes/no-margin decision model",
|
|
380
|
+
"support_status": "supported"
|
|
381
|
+
},
|
|
382
|
+
"nanojev": {
|
|
383
|
+
"backend": "transformers",
|
|
384
|
+
"model": "C-Tianyu/NanoJev",
|
|
385
|
+
"description": "Qwen3-0.6B attention-based candidate-set head",
|
|
386
|
+
"support_status": "pending",
|
|
387
|
+
"enabled": false
|
|
388
|
+
},
|
|
389
|
+
"open-jev-deberta-v3-large": {
|
|
390
|
+
"backend": "transformers",
|
|
391
|
+
"model": "com-kotobalabs/open-jev-deberta-v3-large",
|
|
392
|
+
"description": "DeBERTa encoder with option scoring head",
|
|
393
|
+
"support_status": "pending",
|
|
394
|
+
"enabled": false
|
|
395
|
+
},
|
|
396
|
+
"system-one-qwen3.5-4b-scorer": {
|
|
397
|
+
"backend": "transformers",
|
|
398
|
+
"model": "pngwn/system-one-qwen3.5-4b-scorer",
|
|
399
|
+
"recipe": "sequence-classifier-scalar-options-v1",
|
|
400
|
+
"description": "Qwen3.5 plus LoRA scalar candidate-scoring head",
|
|
401
|
+
"support_status": "supported"
|
|
402
|
+
},
|
|
403
|
+
"ztc-judge-4b": {
|
|
404
|
+
"backend": "transformers",
|
|
405
|
+
"model": "FINAL-Bench/ZTC-Judge-4B",
|
|
406
|
+
"recipe": "hidden-state-candidate-verifier-v1",
|
|
407
|
+
"description": "Question-answer verifier using a final-hidden-state RBF probe",
|
|
408
|
+
"support_status": "supported",
|
|
409
|
+
"config": {
|
|
410
|
+
"decision": {
|
|
411
|
+
"loader": {
|
|
412
|
+
"revision": "b0bb16038d7adbca6f4edda90ac72fb960d8319f",
|
|
413
|
+
"attn_implementation": "sdpa"
|
|
414
|
+
},
|
|
415
|
+
"input": {
|
|
416
|
+
"mode": "candidate_tasks",
|
|
417
|
+
"template": "다음은 어떤 문제와 그에 대한 답변이다. 이 답변이 옳은지 판단하라.\n\n[문제]\n{query}\n\n{instructions}\n\n[답변]\n{candidate}\n\n이 답변은 옳은가?",
|
|
418
|
+
"max_length": 2048
|
|
419
|
+
},
|
|
420
|
+
"probe": {
|
|
421
|
+
"kind": "rbf",
|
|
422
|
+
"repo": "auto",
|
|
423
|
+
"file": "ztc_curve_probe.npz"
|
|
424
|
+
}
|
|
425
|
+
}
|
|
426
|
+
}
|
|
427
|
+
},
|
|
428
|
+
"ztc-judge-9b": {
|
|
429
|
+
"backend": "transformers",
|
|
430
|
+
"model": "FINAL-Bench/ZTC-Judge-9B",
|
|
431
|
+
"recipe": "hidden-state-candidate-verifier-v1",
|
|
432
|
+
"description": "Question-answer verifier using a final-hidden-state RBF probe",
|
|
433
|
+
"support_status": "supported",
|
|
434
|
+
"config": {
|
|
435
|
+
"decision": {
|
|
436
|
+
"loader": {
|
|
437
|
+
"revision": "05d154a56b6d48ea7f2d999cabbafae3f26734b9",
|
|
438
|
+
"attn_implementation": "sdpa"
|
|
439
|
+
},
|
|
440
|
+
"input": {
|
|
441
|
+
"mode": "candidate_tasks",
|
|
442
|
+
"template": "다음은 어떤 문제와 그에 대한 답변이다. 이 답변이 옳은지 판단하라.\n\n[문제]\n{query}\n\n{instructions}\n\n[답변]\n{candidate}\n\n이 답변은 옳은가?",
|
|
443
|
+
"max_length": 2048
|
|
444
|
+
},
|
|
445
|
+
"probe": {
|
|
446
|
+
"kind": "rbf",
|
|
447
|
+
"repo": "auto",
|
|
448
|
+
"file": "ztc_curve_probe.npz"
|
|
449
|
+
}
|
|
450
|
+
}
|
|
451
|
+
}
|
|
452
|
+
},
|
|
453
|
+
"ztc-judge-27b": {
|
|
454
|
+
"backend": "transformers",
|
|
455
|
+
"model": "FINAL-Bench/ZTC-Judge-27B",
|
|
456
|
+
"recipe": "hidden-state-candidate-verifier-v1",
|
|
457
|
+
"description": "Question-answer verifier using a final-hidden-state RBF probe",
|
|
458
|
+
"support_status": "supported",
|
|
459
|
+
"config": {
|
|
460
|
+
"decision": {
|
|
461
|
+
"loader": {
|
|
462
|
+
"revision": "6cbd8acd12d185941271adfd0b302371b789b85a",
|
|
463
|
+
"attn_implementation": "sdpa"
|
|
464
|
+
},
|
|
465
|
+
"input": {
|
|
466
|
+
"mode": "candidate_tasks",
|
|
467
|
+
"template": "다음은 어떤 문제와 그에 대한 답변이다. 이 답변이 옳은지 판단하라.\n\n[문제]\n{query}\n\n{instructions}\n\n[답변]\n{candidate}\n\n이 답변은 옳은가?",
|
|
468
|
+
"max_length": 2048
|
|
469
|
+
},
|
|
470
|
+
"probe": {
|
|
471
|
+
"kind": "rbf",
|
|
472
|
+
"repo": "auto",
|
|
473
|
+
"file": "ztc_curve_probe.npz"
|
|
474
|
+
},
|
|
475
|
+
"batch_size": 1
|
|
476
|
+
}
|
|
477
|
+
}
|
|
478
|
+
},
|
|
479
|
+
"darwin-397b-ztc": {
|
|
480
|
+
"backend": "transformers",
|
|
481
|
+
"model": "FINAL-Bench/Darwin-397B-ZTC",
|
|
482
|
+
"recipe": "hidden-state-probe-v1",
|
|
483
|
+
"description": "Calibrated self-confidence probe over the final prompt hidden state",
|
|
484
|
+
"support_status": "supported",
|
|
485
|
+
"config": {
|
|
486
|
+
"decision": {
|
|
487
|
+
"question_types": [
|
|
488
|
+
"noul"
|
|
489
|
+
],
|
|
490
|
+
"loader": {
|
|
491
|
+
"revision": "7ff4f83a3d9873844cd6e066b691150f4c9cf501",
|
|
492
|
+
"attn_implementation": "sdpa",
|
|
493
|
+
"device_map": "auto"
|
|
494
|
+
},
|
|
495
|
+
"input": {
|
|
496
|
+
"mode": "shared_state",
|
|
497
|
+
"template": "{state}",
|
|
498
|
+
"fields": {},
|
|
499
|
+
"max_length": 2048
|
|
500
|
+
},
|
|
501
|
+
"probe": {
|
|
502
|
+
"kind": "linear",
|
|
503
|
+
"repo": "auto",
|
|
504
|
+
"file": "ztc/ztc_probe_darwin397b.npz"
|
|
505
|
+
},
|
|
506
|
+
"output": {
|
|
507
|
+
"transform": "probe_calibrated_sigmoid"
|
|
508
|
+
},
|
|
509
|
+
"batch_size": 1
|
|
510
|
+
}
|
|
511
|
+
}
|
|
512
|
+
},
|
|
513
|
+
"patronus-lynx-8b-instruct": {
|
|
514
|
+
"backend": "transformers",
|
|
515
|
+
"model": "PatronusAI/Llama-3-Patronus-Lynx-8B-Instruct",
|
|
516
|
+
"description": "Generative RAG faithfulness judge with PASS/FAIL output",
|
|
517
|
+
"support_status": "pending",
|
|
518
|
+
"enabled": false,
|
|
519
|
+
"config": {
|
|
520
|
+
"decision.readout": "generative_structured_judge"
|
|
521
|
+
}
|
|
522
|
+
},
|
|
523
|
+
"laya-multilingual": {
|
|
524
|
+
"backend": "transformers",
|
|
525
|
+
"model": "convaiinnovations/laya-multilingual",
|
|
526
|
+
"description": "mmBERT option-marker set head with act/escalate output",
|
|
527
|
+
"support_status": "pending",
|
|
528
|
+
"enabled": false,
|
|
529
|
+
"config": {
|
|
530
|
+
"decision.readout": "option_marker_set_head"
|
|
531
|
+
}
|
|
532
|
+
},
|
|
533
|
+
"laya": {
|
|
534
|
+
"backend": "transformers",
|
|
535
|
+
"model": "convaiinnovations/laya",
|
|
536
|
+
"description": "Published Laya non-autoregressive typed-decision model",
|
|
537
|
+
"support_status": "supported",
|
|
538
|
+
"enabled": true,
|
|
539
|
+
"config": {
|
|
540
|
+
"decision.readout": "laya_native",
|
|
541
|
+
"decision.question_types": ["choice", "score", "noul"]
|
|
542
|
+
}
|
|
543
|
+
},
|
|
544
|
+
"vectara-hhem-2.1": {
|
|
545
|
+
"backend": "transformers",
|
|
546
|
+
"model": "vectara/hallucination_evaluation_model",
|
|
547
|
+
"description": "Pairwise premise/hypothesis consistency classifier",
|
|
548
|
+
"support_status": "pending",
|
|
549
|
+
"enabled": false,
|
|
550
|
+
"config": {
|
|
551
|
+
"decision.readout": "pair_sequence_classifier"
|
|
552
|
+
}
|
|
553
|
+
},
|
|
554
|
+
"qwen3-next-80b-a3b-instruct": {
|
|
555
|
+
"backend": "transformers",
|
|
556
|
+
"model": "Qwen/Qwen3-Next-80B-A3B-Instruct",
|
|
557
|
+
"description": "Generic autoregressive instruction model without a decision readout recipe",
|
|
558
|
+
"support_status": "pending",
|
|
559
|
+
"enabled": false,
|
|
560
|
+
"config": {
|
|
561
|
+
"decision.readout": "generative_structured_judge"
|
|
562
|
+
}
|
|
563
|
+
},
|
|
564
|
+
"laya-mlx": {
|
|
565
|
+
"backend": "mlx",
|
|
566
|
+
"model": "aac6fef/laya-mlx",
|
|
567
|
+
"description": "MLX bidirectional decision encoder with marker scalar outputs",
|
|
568
|
+
"support_status": "pending",
|
|
569
|
+
"enabled": false,
|
|
570
|
+
"config": {
|
|
571
|
+
"decision.readout": "marker_scalar_head",
|
|
572
|
+
"decision.encoder": "bidirectional",
|
|
573
|
+
"decision.question_types": ["choice", "score", "noul"],
|
|
574
|
+
"decision.output": "option_probabilities"
|
|
575
|
+
}
|
|
576
|
+
},
|
|
577
|
+
"bge-reranker-v2-m3": {
|
|
578
|
+
"backend": "transformers",
|
|
579
|
+
"model": "BAAI/bge-reranker-v2-m3",
|
|
580
|
+
"recipe": "sequence-classifier-native-scalar-v1",
|
|
581
|
+
"description": "BGE multilingual reranker exposed as candidate scoring",
|
|
582
|
+
"support_status": "supported"
|
|
583
|
+
},
|
|
584
|
+
"gte-reranker-modernbert-base": {
|
|
585
|
+
"backend": "transformers",
|
|
586
|
+
"model": "Alibaba-NLP/gte-reranker-modernbert-base",
|
|
587
|
+
"recipe": "sequence-classifier-native-scalar-v1",
|
|
588
|
+
"description": "GTE ModernBERT reranker exposed as candidate scoring",
|
|
589
|
+
"support_status": "supported"
|
|
590
|
+
},
|
|
591
|
+
"certo": {
|
|
592
|
+
"backend": "transformers",
|
|
593
|
+
"model": "altslate/certo-decision-model",
|
|
594
|
+
"recipe": "sequence-classifier-native-scalar-v1",
|
|
595
|
+
"description": "Certo decision classifier exposed as candidate scoring",
|
|
596
|
+
"support_status": "supported"
|
|
597
|
+
},
|
|
598
|
+
"mxbai-rerank-base-v2": {
|
|
599
|
+
"backend": "transformers",
|
|
600
|
+
"model": "mixedbread-ai/mxbai-rerank-base-v2",
|
|
601
|
+
"recipe": "cross-encoder-reranker-v1",
|
|
602
|
+
"description": "Mixedbread reranker exposed as candidate scoring",
|
|
603
|
+
"support_status": "supported"
|
|
604
|
+
},
|
|
605
|
+
"qwen3-reranker-4b": {
|
|
606
|
+
"backend": "transformers",
|
|
607
|
+
"model": "Qwen/Qwen3-Reranker-4B",
|
|
608
|
+
"recipe": "cross-encoder-reranker-v1",
|
|
609
|
+
"description": "Qwen3 reranker exposed as candidate scoring",
|
|
610
|
+
"support_status": "supported"
|
|
611
|
+
},
|
|
612
|
+
"zerank-2": {
|
|
613
|
+
"backend": "transformers",
|
|
614
|
+
"model": "zeroentropy/zerank-2-reranker",
|
|
615
|
+
"recipe": "cross-encoder-reranker-v1",
|
|
616
|
+
"description": "ZeroEntropy reranker exposed as candidate scoring",
|
|
617
|
+
"support_status": "supported"
|
|
618
|
+
},
|
|
619
|
+
"jqv": {
|
|
620
|
+
"backend": "transformers",
|
|
621
|
+
"model": "Qwen/Qwen3-32B",
|
|
622
|
+
"description": "JQV option-letter causal logit judge",
|
|
623
|
+
"support_status": "supported",
|
|
624
|
+
"enabled": true,
|
|
625
|
+
"config": {
|
|
626
|
+
"decision": {
|
|
627
|
+
"readout": "causal_options",
|
|
628
|
+
"profile": "jqv",
|
|
629
|
+
"question_types": ["choice", "score", "noul"],
|
|
630
|
+
"temperature": 3.0225814579771493,
|
|
631
|
+
"loader": {
|
|
632
|
+
"model": "Qwen/Qwen3-32B",
|
|
633
|
+
"revision": "9216db5781bf21249d130ec9da846c4624c16137",
|
|
634
|
+
"dtype": "bf16",
|
|
635
|
+
"device_map": "auto",
|
|
636
|
+
"attn_implementation": "sdpa"
|
|
637
|
+
}
|
|
638
|
+
}
|
|
639
|
+
}
|
|
640
|
+
},
|
|
641
|
+
"litjev": {
|
|
642
|
+
"backend": "transformers",
|
|
643
|
+
"model": "Qwen/Qwen3.8-27B",
|
|
644
|
+
"description": "LitJev native no-thinking causal option-logit judge",
|
|
645
|
+
"support_status": "supported",
|
|
646
|
+
"enabled": true,
|
|
647
|
+
"config": {
|
|
648
|
+
"decision": {
|
|
649
|
+
"readout": "causal_options",
|
|
650
|
+
"profile": "litjev",
|
|
651
|
+
"question_types": ["choice", "score", "noul"],
|
|
652
|
+
"temperature": 1.0,
|
|
653
|
+
"loader": {
|
|
654
|
+
"model": "Qwen/Qwen3.8-27B",
|
|
655
|
+
"revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0",
|
|
656
|
+
"dtype": "bf16",
|
|
657
|
+
"device_map": "auto",
|
|
658
|
+
"attn_implementation": "sdpa"
|
|
659
|
+
}
|
|
660
|
+
}
|
|
661
|
+
}
|
|
662
|
+
},
|
|
663
|
+
"djev": {
|
|
664
|
+
"backend": "transformers",
|
|
665
|
+
"model": "google/diffusiongemma-26B-A4B-it",
|
|
666
|
+
"description": "DJeV structured DiffusionGemma decision distribution",
|
|
667
|
+
"support_status": "pending",
|
|
668
|
+
"enabled": false,
|
|
669
|
+
"config": {"decision.readout": "diffusion_structured_read"}
|
|
670
|
+
},
|
|
671
|
+
"djev-thinking": {
|
|
672
|
+
"backend": "transformers",
|
|
673
|
+
"model": "google/diffusiongemma-26B-A4B-it",
|
|
674
|
+
"description": "DJeV thinking variant with staged thought and distribution readout",
|
|
675
|
+
"support_status": "pending",
|
|
676
|
+
"enabled": false,
|
|
677
|
+
"config": {"decision.readout": "diffusion_thinking_read"}
|
|
678
|
+
},
|
|
679
|
+
"jeff": {
|
|
680
|
+
"backend": "transformers",
|
|
681
|
+
"model": "knowledgator/gliformer-large-v1",
|
|
682
|
+
"description": "Jeff GLiFormer native system-one decision service",
|
|
683
|
+
"support_status": "pending",
|
|
684
|
+
"enabled": false,
|
|
685
|
+
"config": {"decision.readout": "gliformer_native"}
|
|
686
|
+
},
|
|
687
|
+
"open-jev-zefan-2b": {
|
|
688
|
+
"backend": "transformers",
|
|
689
|
+
"model": "ZefanCai/Open-Jev-2B",
|
|
690
|
+
"description": "Open-Jev 2B packaged LoRA plus calibrated scalar head",
|
|
691
|
+
"support_status": "supported",
|
|
692
|
+
"enabled": true,
|
|
693
|
+
"config": {
|
|
694
|
+
"decision": {
|
|
695
|
+
"readout": "openjev_scalar_head",
|
|
696
|
+
"question_types": ["choice", "score", "noul"],
|
|
697
|
+
"dtype": "bfloat16",
|
|
698
|
+
"loader": {
|
|
699
|
+
"base_model": "Qwen/Qwen3.5-2B",
|
|
700
|
+
"revision": "15852e8c16360a2fea060d615a32b45270f8a8fc",
|
|
701
|
+
"tokenizer": "Qwen/Qwen3.5-2B",
|
|
702
|
+
"model_class": "image_text_to_text",
|
|
703
|
+
"backbone_path": "model.language_model",
|
|
704
|
+
"attn_implementation": "sdpa",
|
|
705
|
+
"adapter": {
|
|
706
|
+
"repo": "ZefanCai/Open-Jev-2B",
|
|
707
|
+
"revision": "0c7aa498b1627be8da4acf34c863ff0ee0a92785",
|
|
708
|
+
"subfolder": "package/checkpoint/adapter"
|
|
709
|
+
}
|
|
710
|
+
},
|
|
711
|
+
"input": {
|
|
712
|
+
"template": "{query}\n{instructions}\n{candidate}",
|
|
713
|
+
"max_length": 4096,
|
|
714
|
+
"chat_template": {
|
|
715
|
+
"add_generation_prompt": true,
|
|
716
|
+
"enable_thinking": false
|
|
717
|
+
}
|
|
718
|
+
},
|
|
719
|
+
"artifacts": {
|
|
720
|
+
"head": {
|
|
721
|
+
"repo": "ZefanCai/Open-Jev-2B",
|
|
722
|
+
"revision": "0c7aa498b1627be8da4acf34c863ff0ee0a92785",
|
|
723
|
+
"file": "package/checkpoint/head.pt",
|
|
724
|
+
"kind": "linear"
|
|
725
|
+
},
|
|
726
|
+
"calibration": {
|
|
727
|
+
"repo": "ZefanCai/Open-Jev-2B",
|
|
728
|
+
"revision": "0c7aa498b1627be8da4acf34c863ff0ee0a92785",
|
|
729
|
+
"file": "package/checkpoint/temperature.json",
|
|
730
|
+
"field": "temperature"
|
|
731
|
+
}
|
|
732
|
+
},
|
|
733
|
+
"batch_size": 32
|
|
734
|
+
}
|
|
735
|
+
}
|
|
736
|
+
},
|
|
737
|
+
"open-jev-zefan-9b": {
|
|
738
|
+
"backend": "transformers",
|
|
739
|
+
"model": "ZefanCai/Open-Jev-9B",
|
|
740
|
+
"description": "Open-Jev 9B packaged LoRA plus calibrated scalar head",
|
|
741
|
+
"support_status": "supported",
|
|
742
|
+
"enabled": true,
|
|
743
|
+
"config": {
|
|
744
|
+
"decision": {
|
|
745
|
+
"readout": "openjev_scalar_head",
|
|
746
|
+
"question_types": ["choice", "score", "noul"],
|
|
747
|
+
"dtype": "bfloat16",
|
|
748
|
+
"loader": {
|
|
749
|
+
"base_model": "Qwen/Qwen3.5-9B",
|
|
750
|
+
"revision": "c202236235762e1c871ad0ccb60c8ee5ba337b9a",
|
|
751
|
+
"tokenizer": "Qwen/Qwen3.5-9B",
|
|
752
|
+
"model_class": "image_text_to_text",
|
|
753
|
+
"backbone_path": "model.language_model",
|
|
754
|
+
"attn_implementation": "sdpa",
|
|
755
|
+
"adapter": {
|
|
756
|
+
"repo": "ZefanCai/Open-Jev-9B",
|
|
757
|
+
"revision": "47e966881e489511c0c7f5633a9e1960a676a551",
|
|
758
|
+
"subfolder": "package/checkpoint/adapter"
|
|
759
|
+
}
|
|
760
|
+
},
|
|
761
|
+
"input": {
|
|
762
|
+
"template": "{query}\n{instructions}\n{candidate}",
|
|
763
|
+
"max_length": 4096,
|
|
764
|
+
"chat_template": {
|
|
765
|
+
"add_generation_prompt": true,
|
|
766
|
+
"enable_thinking": false
|
|
767
|
+
}
|
|
768
|
+
},
|
|
769
|
+
"artifacts": {
|
|
770
|
+
"head": {
|
|
771
|
+
"repo": "ZefanCai/Open-Jev-9B",
|
|
772
|
+
"revision": "47e966881e489511c0c7f5633a9e1960a676a551",
|
|
773
|
+
"file": "package/checkpoint/head.pt",
|
|
774
|
+
"kind": "linear"
|
|
775
|
+
},
|
|
776
|
+
"calibration": {
|
|
777
|
+
"repo": "ZefanCai/Open-Jev-9B",
|
|
778
|
+
"revision": "47e966881e489511c0c7f5633a9e1960a676a551",
|
|
779
|
+
"file": "package/checkpoint/temperature.json",
|
|
780
|
+
"field": "temperature"
|
|
781
|
+
}
|
|
782
|
+
},
|
|
783
|
+
"batch_size": 32
|
|
784
|
+
}
|
|
785
|
+
}
|
|
786
|
+
},
|
|
787
|
+
"opendecision": {
|
|
788
|
+
"backend": "transformers",
|
|
789
|
+
"model": "MoritzLaurer/ModernBERT-large-zeroshot-v2.0",
|
|
790
|
+
"description": "OpenDecision NLI entailment candidate scorer",
|
|
791
|
+
"recipe": "hidden-state-candidate-verifier-v1",
|
|
792
|
+
"support_status": "supported",
|
|
793
|
+
"enabled": true,
|
|
794
|
+
"config": {
|
|
795
|
+
"decision": {
|
|
796
|
+
"readout": "nli_entailment",
|
|
797
|
+
"question_types": ["choice", "score", "noul"],
|
|
798
|
+
"loader": {
|
|
799
|
+
"base_model": "MoritzLaurer/ModernBERT-large-zeroshot-v2.0",
|
|
800
|
+
"tokenizer": "MoritzLaurer/ModernBERT-large-zeroshot-v2.0",
|
|
801
|
+
"revision": "a51e07b524299e309dd2b88d48b0cfa2bd9ec598"
|
|
802
|
+
},
|
|
803
|
+
"entailment_label": "entailment",
|
|
804
|
+
"input": {
|
|
805
|
+
"premise_template": "{state}",
|
|
806
|
+
"hypothesis_template": "{instructions}\n\nCandidate: {candidate}",
|
|
807
|
+
"max_length": 8192
|
|
808
|
+
},
|
|
809
|
+
"batch_size": 32
|
|
810
|
+
}
|
|
811
|
+
}
|
|
812
|
+
},
|
|
813
|
+
"openjev-thinking": {
|
|
814
|
+
"backend": "transformers",
|
|
815
|
+
"model": "google/diffusiongemma-26b-a4b-it",
|
|
816
|
+
"description": "OpenJev thinking DiffusionGemma typed server",
|
|
817
|
+
"support_status": "pending",
|
|
818
|
+
"enabled": false,
|
|
819
|
+
"config": {"decision.readout": "diffusion_thinking_read"}
|
|
820
|
+
},
|
|
821
|
+
"openjev-verdict": {
|
|
822
|
+
"backend": "transformers",
|
|
823
|
+
"model": "heman10x/rlcd-modernbert-151m",
|
|
824
|
+
"description": "OpenJev Verdict GLiClass distribution and calibrator",
|
|
825
|
+
"support_status": "supported",
|
|
826
|
+
"enabled": true,
|
|
827
|
+
"config": {
|
|
828
|
+
"decision": {
|
|
829
|
+
"readout": "gliclass_calibrated",
|
|
830
|
+
"question_types": ["choice", "score", "noul"],
|
|
831
|
+
"loader": {
|
|
832
|
+
"base_model": "heman10x/rlcd-modernbert-151m",
|
|
833
|
+
"tokenizer": "heman10x/rlcd-modernbert-151m",
|
|
834
|
+
"revision": "8af2496eb63c7fa66d7d234e1f62629380030eb4"
|
|
835
|
+
},
|
|
836
|
+
"input": {
|
|
837
|
+
"max_length": 512,
|
|
838
|
+
"max_candidates": 25
|
|
839
|
+
},
|
|
840
|
+
"batch_size": 32,
|
|
841
|
+
"calibrator": {
|
|
842
|
+
"repo": "heman10x/rlcd-modernbert-151m",
|
|
843
|
+
"file": "calibrator.json",
|
|
844
|
+
"revision": "8af2496eb63c7fa66d7d234e1f62629380030eb4",
|
|
845
|
+
"format": "rlcd-calibrator-v1"
|
|
846
|
+
}
|
|
847
|
+
}
|
|
848
|
+
}
|
|
849
|
+
},
|
|
850
|
+
"reflex-4b": {
|
|
851
|
+
"backend": "transformers",
|
|
852
|
+
"model": "Qwen/Qwen3.5-4B",
|
|
853
|
+
"description": "Reflex 4B LoRA decision logits with two-order aggregation",
|
|
854
|
+
"support_status": "supported",
|
|
855
|
+
"enabled": true,
|
|
856
|
+
"config": {
|
|
857
|
+
"decision": {
|
|
858
|
+
"readout": "causal_options",
|
|
859
|
+
"profile": "reflex",
|
|
860
|
+
"question_types": ["choice", "score", "noul"],
|
|
861
|
+
"temperature": 1.0,
|
|
862
|
+
"two_order_aggregation": true,
|
|
863
|
+
"loader": {
|
|
864
|
+
"model": "Qwen/Qwen3.5-4B",
|
|
865
|
+
"revision": "851bf6e806efd8d0a36b00ddf55e13ccb7b8cd0a",
|
|
866
|
+
"adapter": "kshetrajna12/reflex-qwen3.5-4b-lora",
|
|
867
|
+
"adapter_revision": "9df1cbbbc1837ba494530a59d3d9d345688ed250",
|
|
868
|
+
"dtype": "bf16",
|
|
869
|
+
"device_map": "auto",
|
|
870
|
+
"attn_implementation": "sdpa"
|
|
871
|
+
}
|
|
872
|
+
}
|
|
873
|
+
}
|
|
874
|
+
},
|
|
875
|
+
"reflex-27b": {
|
|
876
|
+
"backend": "transformers",
|
|
877
|
+
"model": "Qwen/Qwen3.8-27B",
|
|
878
|
+
"description": "Reflex 27B direct decision logits with two-order aggregation",
|
|
879
|
+
"support_status": "supported",
|
|
880
|
+
"enabled": true,
|
|
881
|
+
"config": {
|
|
882
|
+
"decision": {
|
|
883
|
+
"readout": "causal_options",
|
|
884
|
+
"profile": "reflex",
|
|
885
|
+
"question_types": ["choice", "score", "noul"],
|
|
886
|
+
"temperature": 1.0,
|
|
887
|
+
"two_order_aggregation": true,
|
|
888
|
+
"loader": {
|
|
889
|
+
"model": "Qwen/Qwen3.8-27B",
|
|
890
|
+
"revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0",
|
|
891
|
+
"dtype": "bf16",
|
|
892
|
+
"device_map": "auto",
|
|
893
|
+
"attn_implementation": "sdpa"
|
|
894
|
+
}
|
|
895
|
+
}
|
|
896
|
+
}
|
|
897
|
+
},
|
|
898
|
+
"simplejev-qwen3.6-35b-a3b": {
|
|
899
|
+
"backend": "transformers",
|
|
900
|
+
"model": "Qwen/Qwen3.6-35B-A3B",
|
|
901
|
+
"description": "SimpleJev Qwen3.6 direct next-token decision logits",
|
|
902
|
+
"support_status": "supported",
|
|
903
|
+
"enabled": true,
|
|
904
|
+
"config": {
|
|
905
|
+
"decision": {
|
|
906
|
+
"readout": "causal_options",
|
|
907
|
+
"profile": "simplejev_v1",
|
|
908
|
+
"question_types": ["choice", "score", "noul"],
|
|
909
|
+
"temperature": 1.0,
|
|
910
|
+
"loader": {
|
|
911
|
+
"model": "Qwen/Qwen3.6-35B-A3B",
|
|
912
|
+
"revision": "995ad96eacd98c81ed38be0c5b274b04031597b0",
|
|
913
|
+
"dtype": "bf16",
|
|
914
|
+
"device_map": "auto",
|
|
915
|
+
"attn_implementation": "sdpa"
|
|
916
|
+
}
|
|
917
|
+
}
|
|
918
|
+
}
|
|
919
|
+
},
|
|
920
|
+
"simplejev-qwen3.8-27b": {
|
|
921
|
+
"backend": "transformers",
|
|
922
|
+
"model": "Qwen/Qwen3.8-27B",
|
|
923
|
+
"description": "SimpleJev Qwen3.8 direct next-token decision logits",
|
|
924
|
+
"support_status": "supported",
|
|
925
|
+
"enabled": true,
|
|
926
|
+
"config": {
|
|
927
|
+
"decision": {
|
|
928
|
+
"readout": "causal_options",
|
|
929
|
+
"profile": "simplejev_v1",
|
|
930
|
+
"question_types": ["choice", "score", "noul"],
|
|
931
|
+
"temperature": 1.0,
|
|
932
|
+
"loader": {
|
|
933
|
+
"model": "Qwen/Qwen3.8-27B",
|
|
934
|
+
"revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0",
|
|
935
|
+
"dtype": "bf16",
|
|
936
|
+
"device_map": "auto",
|
|
937
|
+
"attn_implementation": "sdpa"
|
|
938
|
+
}
|
|
939
|
+
}
|
|
940
|
+
}
|
|
941
|
+
},
|
|
942
|
+
"smalljev": {
|
|
943
|
+
"backend": "transformers",
|
|
944
|
+
"model": "isHeSatoshi/smalljev-semantic-v9",
|
|
945
|
+
"description": "SmallJev semantic-v9 LoRA plus option-span scorer",
|
|
946
|
+
"support_status": "supported",
|
|
947
|
+
"enabled": true,
|
|
948
|
+
"config": {
|
|
949
|
+
"decision": {
|
|
950
|
+
"readout": "semantic_option_head",
|
|
951
|
+
"question_types": ["choice"],
|
|
952
|
+
"loader": {
|
|
953
|
+
"base_model": "openbmb/MiniCPM5-2B-Base",
|
|
954
|
+
"revision": "96a57cd572a02506b4500f54427dca24970c1bac",
|
|
955
|
+
"tokenizer": "openbmb/MiniCPM5-2B-Base",
|
|
956
|
+
"adapter": {
|
|
957
|
+
"repo": "isHeSatoshi/smalljev-semantic-v9",
|
|
958
|
+
"revision": "8e2bac7c4e922371c05899f4f000233e8096fee6",
|
|
959
|
+
"subfolder": "semantic-v9-lora"
|
|
960
|
+
}
|
|
961
|
+
},
|
|
962
|
+
"artifacts": {
|
|
963
|
+
"head": {
|
|
964
|
+
"repo": "isHeSatoshi/smalljev-semantic-v9",
|
|
965
|
+
"revision": "8e2bac7c4e922371c05899f4f000233e8096fee6",
|
|
966
|
+
"file": "semantic-v9-scorer.pt"
|
|
967
|
+
}
|
|
968
|
+
}
|
|
969
|
+
}
|
|
970
|
+
}
|
|
971
|
+
},
|
|
972
|
+
"winnow-12b": {
|
|
973
|
+
"backend": "llama",
|
|
974
|
+
"model": "EldanRing/Winnow-12B-Q8_0.gguf",
|
|
975
|
+
"description": "Winnow shared-branch llama.cpp decision server",
|
|
976
|
+
"support_status": "pending",
|
|
977
|
+
"enabled": false,
|
|
978
|
+
"config": {"decision.readout": "winnow_shared_branch"}
|
|
979
|
+
}
|
|
980
|
+
}
|
|
981
|
+
}
|