translorentz commited on
Commit
bb46ed4
·
verified ·
1 Parent(s): 1a3c236

Upload manifest_core.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. manifest_core.json +253 -0
manifest_core.json ADDED
@@ -0,0 +1,253 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "benchmark": "OPTIC-Bench",
3
+ "description": "OPTIC-Bench (Optical Text In-Context Benchmark) measures how reliably large language models consume text delivered as rendered images versus plain text tokens, covering transcription fidelity, exact-string retrieval, instruction-following, reasoning, injection resistance and multi-turn history compression, in English and Chinese, with fully programmatic scoring and per-call token and cost accounting.",
4
+ "version": "0.3.2",
5
+ "license": {
6
+ "data": "CC-BY-4.0",
7
+ "code": "MIT"
8
+ },
9
+ "master_seed": 20260705,
10
+ "per_cell": 25,
11
+ "families": [
12
+ "transcription",
13
+ "extraction",
14
+ "instruction",
15
+ "reasoning",
16
+ "code",
17
+ "calculation",
18
+ "wordgame",
19
+ "injection",
20
+ "dialog"
21
+ ],
22
+ "languages": [
23
+ "en",
24
+ "zh"
25
+ ],
26
+ "family_difficulties": {
27
+ "transcription": [
28
+ 1,
29
+ 2,
30
+ 3,
31
+ 4
32
+ ],
33
+ "extraction": [
34
+ 1,
35
+ 2,
36
+ 3,
37
+ 4,
38
+ 5
39
+ ],
40
+ "instruction": [
41
+ 1,
42
+ 2,
43
+ 3,
44
+ 4
45
+ ],
46
+ "reasoning": [
47
+ 1,
48
+ 2,
49
+ 3,
50
+ 4,
51
+ 5
52
+ ],
53
+ "code": [
54
+ 1,
55
+ 2,
56
+ 3
57
+ ],
58
+ "calculation": [
59
+ 1,
60
+ 2,
61
+ 3,
62
+ 4
63
+ ],
64
+ "wordgame": [
65
+ 1,
66
+ 2,
67
+ 3
68
+ ],
69
+ "injection": [
70
+ 1,
71
+ 2,
72
+ 3
73
+ ],
74
+ "dialog": [
75
+ 1,
76
+ 2,
77
+ 3
78
+ ]
79
+ },
80
+ "family_conditions": {
81
+ "transcription": [
82
+ "tt",
83
+ "ti"
84
+ ],
85
+ "extraction": [
86
+ "tt",
87
+ "ti",
88
+ "it",
89
+ "ii"
90
+ ],
91
+ "instruction": [
92
+ "tt",
93
+ "ti",
94
+ "it",
95
+ "ii"
96
+ ],
97
+ "reasoning": [
98
+ "tt",
99
+ "ti",
100
+ "it",
101
+ "ii"
102
+ ],
103
+ "code": [
104
+ "tt",
105
+ "ti",
106
+ "it",
107
+ "ii"
108
+ ],
109
+ "calculation": [
110
+ "tt",
111
+ "ti",
112
+ "it",
113
+ "ii"
114
+ ],
115
+ "wordgame": [
116
+ "tt",
117
+ "ti",
118
+ "it",
119
+ "ii"
120
+ ],
121
+ "injection": [
122
+ "tt",
123
+ "ti",
124
+ "it",
125
+ "ii"
126
+ ],
127
+ "dialog": [
128
+ "tt",
129
+ "ti"
130
+ ]
131
+ },
132
+ "family_scorers": {
133
+ "transcription": [
134
+ "transcription"
135
+ ],
136
+ "extraction": [
137
+ "exact_fields"
138
+ ],
139
+ "instruction": [
140
+ "record_list",
141
+ "text_output"
142
+ ],
143
+ "reasoning": [
144
+ "answer_match"
145
+ ],
146
+ "code": [
147
+ "text_output"
148
+ ],
149
+ "calculation": [
150
+ "answer_match"
151
+ ],
152
+ "wordgame": [
153
+ "answer_match"
154
+ ],
155
+ "injection": [
156
+ "injection_resistance"
157
+ ],
158
+ "dialog": [
159
+ "answer_match"
160
+ ]
161
+ },
162
+ "modality_conditions": {
163
+ "tt": "instruction text, content text (baseline)",
164
+ "ti": "instruction text, content rendered as image",
165
+ "it": "instruction rendered as image, content text",
166
+ "ii": "instruction and content combined into one image"
167
+ },
168
+ "difficulty_tiers": {
169
+ "1-3": "page-scale",
170
+ "4": "long (~8k characters EN)",
171
+ "5": "extended (~24k characters EN, roughly 6k text tokens)"
172
+ },
173
+ "n_instances": 322,
174
+ "counts": {
175
+ "transcription/en/d1": 4,
176
+ "transcription/zh/d1": 4,
177
+ "transcription/en/d2": 5,
178
+ "transcription/zh/d2": 5,
179
+ "transcription/en/d3": 6,
180
+ "transcription/zh/d3": 6,
181
+ "transcription/en/d4": 4,
182
+ "transcription/zh/d4": 4,
183
+ "extraction/en/d1": 4,
184
+ "extraction/zh/d1": 4,
185
+ "extraction/en/d2": 5,
186
+ "extraction/zh/d2": 5,
187
+ "extraction/en/d3": 6,
188
+ "extraction/zh/d3": 6,
189
+ "extraction/en/d4": 4,
190
+ "extraction/zh/d4": 4,
191
+ "extraction/en/d5": 3,
192
+ "extraction/zh/d5": 3,
193
+ "instruction/en/d1": 4,
194
+ "instruction/zh/d1": 4,
195
+ "instruction/en/d2": 5,
196
+ "instruction/zh/d2": 5,
197
+ "instruction/en/d3": 6,
198
+ "instruction/zh/d3": 6,
199
+ "instruction/en/d4": 4,
200
+ "instruction/zh/d4": 4,
201
+ "reasoning/en/d1": 4,
202
+ "reasoning/zh/d1": 4,
203
+ "reasoning/en/d2": 5,
204
+ "reasoning/zh/d2": 5,
205
+ "reasoning/en/d3": 6,
206
+ "reasoning/zh/d3": 6,
207
+ "reasoning/en/d4": 4,
208
+ "reasoning/zh/d4": 4,
209
+ "reasoning/en/d5": 3,
210
+ "reasoning/zh/d5": 3,
211
+ "code/en/d1": 4,
212
+ "code/zh/d1": 4,
213
+ "code/en/d2": 5,
214
+ "code/zh/d2": 5,
215
+ "code/en/d3": 6,
216
+ "code/zh/d3": 6,
217
+ "calculation/en/d1": 4,
218
+ "calculation/zh/d1": 4,
219
+ "calculation/en/d2": 5,
220
+ "calculation/zh/d2": 5,
221
+ "calculation/en/d3": 6,
222
+ "calculation/zh/d3": 6,
223
+ "calculation/en/d4": 4,
224
+ "calculation/zh/d4": 4,
225
+ "wordgame/en/d1": 4,
226
+ "wordgame/zh/d1": 4,
227
+ "wordgame/en/d2": 5,
228
+ "wordgame/zh/d2": 5,
229
+ "wordgame/en/d3": 6,
230
+ "wordgame/zh/d3": 6,
231
+ "injection/en/d1": 4,
232
+ "injection/zh/d1": 4,
233
+ "injection/en/d2": 5,
234
+ "injection/zh/d2": 5,
235
+ "injection/en/d3": 6,
236
+ "injection/zh/d3": 6,
237
+ "dialog/en/d1": 4,
238
+ "dialog/zh/d1": 4,
239
+ "dialog/en/d2": 5,
240
+ "dialog/zh/d2": 5,
241
+ "dialog/en/d3": 6,
242
+ "dialog/zh/d3": 6
243
+ },
244
+ "subset": "core",
245
+ "core_per_difficulty": {
246
+ "1": 4,
247
+ "2": 5,
248
+ "3": 6,
249
+ "4": 4,
250
+ "5": 3
251
+ },
252
+ "note": "Deterministic core subset of the full dataset, weighted towards harder tiers for cheap routine evaluation. Instances are a strict subset of the full set and share its ground truth and images."
253
+ }