Spaces:
Running
Running
add gemini-2.0 SI subset results
Browse files
static/eval_results/SI/Gemini-2.0-thinking/summary_and_keyword_stats.json
ADDED
|
@@ -0,0 +1,213 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_summary": {
|
| 3 |
+
"core": {
|
| 4 |
+
"num_eval_tasks": 273,
|
| 5 |
+
"num_eval_samples": 4108,
|
| 6 |
+
"macro_mean_score": 0.5343557337812288
|
| 7 |
+
},
|
| 8 |
+
"open": {
|
| 9 |
+
"num_eval_tasks": 42,
|
| 10 |
+
"num_eval_samples": 808,
|
| 11 |
+
"macro_mean_score": 0.648456858748277
|
| 12 |
+
},
|
| 13 |
+
"overall_score": 0.5495692171101685
|
| 14 |
+
},
|
| 15 |
+
"keyword_stats": {
|
| 16 |
+
"skills": {
|
| 17 |
+
"Object Recognition and Classification": {
|
| 18 |
+
"count": 172,
|
| 19 |
+
"num_samples": 2704,
|
| 20 |
+
"tasks": [],
|
| 21 |
+
"average_score": 0.5642112150199124
|
| 22 |
+
},
|
| 23 |
+
"Scene and Event Understanding": {
|
| 24 |
+
"count": 60,
|
| 25 |
+
"num_samples": 1004,
|
| 26 |
+
"tasks": [],
|
| 27 |
+
"average_score": 0.6916143699384771
|
| 28 |
+
},
|
| 29 |
+
"Text Recognition (OCR)": {
|
| 30 |
+
"count": 101,
|
| 31 |
+
"num_samples": 1680,
|
| 32 |
+
"tasks": [],
|
| 33 |
+
"average_score": 0.5916850214382456
|
| 34 |
+
},
|
| 35 |
+
"Domain-Specific Knowledge and Skills": {
|
| 36 |
+
"count": 46,
|
| 37 |
+
"num_samples": 896,
|
| 38 |
+
"tasks": [],
|
| 39 |
+
"average_score": 0.5670929640047282
|
| 40 |
+
},
|
| 41 |
+
"Language Understanding and Generation": {
|
| 42 |
+
"count": 102,
|
| 43 |
+
"num_samples": 1707,
|
| 44 |
+
"tasks": [],
|
| 45 |
+
"average_score": 0.5832985237106808
|
| 46 |
+
},
|
| 47 |
+
"Mathematical and Logical Reasoning": {
|
| 48 |
+
"count": 91,
|
| 49 |
+
"num_samples": 1628,
|
| 50 |
+
"tasks": [],
|
| 51 |
+
"average_score": 0.5117575336849055
|
| 52 |
+
},
|
| 53 |
+
"Spatial and Temporal Reasoning": {
|
| 54 |
+
"count": 78,
|
| 55 |
+
"num_samples": 1270,
|
| 56 |
+
"tasks": [],
|
| 57 |
+
"average_score": 0.42642077000042744
|
| 58 |
+
},
|
| 59 |
+
"Planning and Decision Making": {
|
| 60 |
+
"count": 23,
|
| 61 |
+
"num_samples": 355,
|
| 62 |
+
"tasks": [],
|
| 63 |
+
"average_score": 0.18765788298411018
|
| 64 |
+
},
|
| 65 |
+
"Commonsense and Social Reasoning": {
|
| 66 |
+
"count": 38,
|
| 67 |
+
"num_samples": 652,
|
| 68 |
+
"tasks": [],
|
| 69 |
+
"average_score": 0.6543178496880195
|
| 70 |
+
},
|
| 71 |
+
"Ethical and Safety Reasoning": {
|
| 72 |
+
"count": 10,
|
| 73 |
+
"num_samples": 170,
|
| 74 |
+
"tasks": [],
|
| 75 |
+
"average_score": 0.7155827067669173
|
| 76 |
+
}
|
| 77 |
+
},
|
| 78 |
+
"input_format": {
|
| 79 |
+
"Photographs": {
|
| 80 |
+
"count": 83,
|
| 81 |
+
"num_samples": 1310,
|
| 82 |
+
"tasks": [],
|
| 83 |
+
"average_score": 0.5632377506511956
|
| 84 |
+
},
|
| 85 |
+
"User Interface Screenshots": {
|
| 86 |
+
"count": 67,
|
| 87 |
+
"num_samples": 1117,
|
| 88 |
+
"tasks": [],
|
| 89 |
+
"average_score": 0.5656524791411764
|
| 90 |
+
},
|
| 91 |
+
"Artistic and Creative Content": {
|
| 92 |
+
"count": 22,
|
| 93 |
+
"num_samples": 388,
|
| 94 |
+
"tasks": [],
|
| 95 |
+
"average_score": 0.6235072262694881
|
| 96 |
+
},
|
| 97 |
+
"Diagrams and Data Visualizations": {
|
| 98 |
+
"count": 88,
|
| 99 |
+
"num_samples": 1523,
|
| 100 |
+
"tasks": [],
|
| 101 |
+
"average_score": 0.5613562780665254
|
| 102 |
+
},
|
| 103 |
+
"Text-Based Images and Documents": {
|
| 104 |
+
"count": 53,
|
| 105 |
+
"num_samples": 847,
|
| 106 |
+
"tasks": [],
|
| 107 |
+
"average_score": 0.4684193193457211
|
| 108 |
+
},
|
| 109 |
+
"3D Models and Aerial Imagery": {
|
| 110 |
+
"count": 2,
|
| 111 |
+
"num_samples": 30,
|
| 112 |
+
"tasks": [],
|
| 113 |
+
"average_score": 0.26205930504439073
|
| 114 |
+
}
|
| 115 |
+
},
|
| 116 |
+
"output_format": {
|
| 117 |
+
"exact_text": {
|
| 118 |
+
"count": 57,
|
| 119 |
+
"num_samples": 876,
|
| 120 |
+
"tasks": [],
|
| 121 |
+
"average_score": 0.5635910639054841
|
| 122 |
+
},
|
| 123 |
+
"structured_output": {
|
| 124 |
+
"count": 72,
|
| 125 |
+
"num_samples": 1120,
|
| 126 |
+
"tasks": [],
|
| 127 |
+
"average_score": 0.5553206502944843
|
| 128 |
+
},
|
| 129 |
+
"numerical_data": {
|
| 130 |
+
"count": 39,
|
| 131 |
+
"num_samples": 694,
|
| 132 |
+
"tasks": [],
|
| 133 |
+
"average_score": 0.5091728623704704
|
| 134 |
+
},
|
| 135 |
+
"contextual_formatted_text": {
|
| 136 |
+
"count": 63,
|
| 137 |
+
"num_samples": 972,
|
| 138 |
+
"tasks": [],
|
| 139 |
+
"average_score": 0.5141636818210693
|
| 140 |
+
},
|
| 141 |
+
"multiple_choice": {
|
| 142 |
+
"count": 33,
|
| 143 |
+
"num_samples": 567,
|
| 144 |
+
"tasks": [],
|
| 145 |
+
"average_score": 0.6457868730596004
|
| 146 |
+
},
|
| 147 |
+
"open_ended_output": {
|
| 148 |
+
"count": 51,
|
| 149 |
+
"num_samples": 986,
|
| 150 |
+
"tasks": [],
|
| 151 |
+
"average_score": 0.5381471672106882
|
| 152 |
+
}
|
| 153 |
+
},
|
| 154 |
+
"input_num": {
|
| 155 |
+
"1-image": {
|
| 156 |
+
"count": 315,
|
| 157 |
+
"num_samples": 5215,
|
| 158 |
+
"tasks": [],
|
| 159 |
+
"average_score": 0.5495692171101688
|
| 160 |
+
}
|
| 161 |
+
},
|
| 162 |
+
"app": {
|
| 163 |
+
"Perception": {
|
| 164 |
+
"count": 82,
|
| 165 |
+
"num_samples": 1318,
|
| 166 |
+
"tasks": [],
|
| 167 |
+
"average_score": 0.6122266550210048
|
| 168 |
+
},
|
| 169 |
+
"Information_Extraction": {
|
| 170 |
+
"count": 41,
|
| 171 |
+
"num_samples": 639,
|
| 172 |
+
"tasks": [],
|
| 173 |
+
"average_score": 0.6692014953534847
|
| 174 |
+
},
|
| 175 |
+
"Knowledge": {
|
| 176 |
+
"count": 77,
|
| 177 |
+
"num_samples": 1291,
|
| 178 |
+
"tasks": [],
|
| 179 |
+
"average_score": 0.5672022990024259
|
| 180 |
+
},
|
| 181 |
+
"Mathematics": {
|
| 182 |
+
"count": 30,
|
| 183 |
+
"num_samples": 497,
|
| 184 |
+
"tasks": [],
|
| 185 |
+
"average_score": 0.4805461436715291
|
| 186 |
+
},
|
| 187 |
+
"Planning": {
|
| 188 |
+
"count": 44,
|
| 189 |
+
"num_samples": 712,
|
| 190 |
+
"tasks": [],
|
| 191 |
+
"average_score": 0.3076764038690534
|
| 192 |
+
},
|
| 193 |
+
"Coding": {
|
| 194 |
+
"count": 16,
|
| 195 |
+
"num_samples": 244,
|
| 196 |
+
"tasks": [],
|
| 197 |
+
"average_score": 0.5742413632119515
|
| 198 |
+
},
|
| 199 |
+
"Science": {
|
| 200 |
+
"count": 22,
|
| 201 |
+
"num_samples": 469,
|
| 202 |
+
"tasks": [],
|
| 203 |
+
"average_score": 0.5807690487533278
|
| 204 |
+
},
|
| 205 |
+
"Metrics": {
|
| 206 |
+
"count": 3,
|
| 207 |
+
"num_samples": 45,
|
| 208 |
+
"tasks": [],
|
| 209 |
+
"average_score": 0.626984126984127
|
| 210 |
+
}
|
| 211 |
+
}
|
| 212 |
+
}
|
| 213 |
+
}
|
static/eval_results/SI/Gemini-2.0-thinking/task_results.json
ADDED
|
@@ -0,0 +1,2207 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"name": "places365_scene_type_classification",
|
| 4 |
+
"score": 1.0,
|
| 5 |
+
"eval_type": "rule",
|
| 6 |
+
"num_demo": 1,
|
| 7 |
+
"num_query": 14
|
| 8 |
+
},
|
| 9 |
+
{
|
| 10 |
+
"name": "music_info_parsing",
|
| 11 |
+
"score": 0.5982142857142857,
|
| 12 |
+
"eval_type": "rule",
|
| 13 |
+
"num_demo": 1,
|
| 14 |
+
"num_query": 14
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"name": "location_vqa",
|
| 18 |
+
"score": 0.5,
|
| 19 |
+
"eval_type": "rule",
|
| 20 |
+
"num_demo": 1,
|
| 21 |
+
"num_query": 14
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"name": "mindmap_elements_parsing",
|
| 25 |
+
"score": 0.42857142857142855,
|
| 26 |
+
"eval_type": "rule",
|
| 27 |
+
"num_demo": 1,
|
| 28 |
+
"num_query": 14
|
| 29 |
+
},
|
| 30 |
+
{
|
| 31 |
+
"name": "multilingual_movie_info_parsing",
|
| 32 |
+
"score": 0.7346938775510202,
|
| 33 |
+
"eval_type": "rule",
|
| 34 |
+
"num_demo": 1,
|
| 35 |
+
"num_query": 14
|
| 36 |
+
},
|
| 37 |
+
{
|
| 38 |
+
"name": "iconqa_count_and_reasoning",
|
| 39 |
+
"score": 0.3684210526315789,
|
| 40 |
+
"eval_type": "rule",
|
| 41 |
+
"num_demo": 1,
|
| 42 |
+
"num_query": 19
|
| 43 |
+
},
|
| 44 |
+
{
|
| 45 |
+
"name": "graph_connectivity",
|
| 46 |
+
"score": 0.8333333333333334,
|
| 47 |
+
"eval_type": "rule",
|
| 48 |
+
"num_demo": 1,
|
| 49 |
+
"num_query": 15
|
| 50 |
+
},
|
| 51 |
+
{
|
| 52 |
+
"name": "graph_shortest_path_planar",
|
| 53 |
+
"score": 0.21428571428571427,
|
| 54 |
+
"eval_type": "rule",
|
| 55 |
+
"num_demo": 1,
|
| 56 |
+
"num_query": 14
|
| 57 |
+
},
|
| 58 |
+
{
|
| 59 |
+
"name": "famous_building_recognition",
|
| 60 |
+
"score": 0.8125,
|
| 61 |
+
"eval_type": "rule",
|
| 62 |
+
"num_demo": 1,
|
| 63 |
+
"num_query": 16
|
| 64 |
+
},
|
| 65 |
+
{
|
| 66 |
+
"name": "geometry_transformation",
|
| 67 |
+
"score": 0.21428571428571427,
|
| 68 |
+
"eval_type": "rule",
|
| 69 |
+
"num_demo": 1,
|
| 70 |
+
"num_query": 14
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"name": "long_string_letter_recognition",
|
| 74 |
+
"score": 0.0,
|
| 75 |
+
"eval_type": "rule",
|
| 76 |
+
"num_demo": 1,
|
| 77 |
+
"num_query": 14
|
| 78 |
+
},
|
| 79 |
+
{
|
| 80 |
+
"name": "handwritten_math_expression_extraction",
|
| 81 |
+
"score": 0.6428571428571429,
|
| 82 |
+
"eval_type": "rule",
|
| 83 |
+
"num_demo": 1,
|
| 84 |
+
"num_query": 14
|
| 85 |
+
},
|
| 86 |
+
{
|
| 87 |
+
"name": "mensa_iq_test",
|
| 88 |
+
"score": 0.3499245852187029,
|
| 89 |
+
"eval_type": "rule",
|
| 90 |
+
"num_demo": 1,
|
| 91 |
+
"num_query": 17
|
| 92 |
+
},
|
| 93 |
+
{
|
| 94 |
+
"name": "geometry_solid",
|
| 95 |
+
"score": 0.35714285714285715,
|
| 96 |
+
"eval_type": "rule",
|
| 97 |
+
"num_demo": 1,
|
| 98 |
+
"num_query": 14
|
| 99 |
+
},
|
| 100 |
+
{
|
| 101 |
+
"name": "flowchart_code_generation",
|
| 102 |
+
"score": 0.6666666666666666,
|
| 103 |
+
"eval_type": "rule",
|
| 104 |
+
"num_demo": 1,
|
| 105 |
+
"num_query": 9
|
| 106 |
+
},
|
| 107 |
+
{
|
| 108 |
+
"name": "animal_pose_estimation",
|
| 109 |
+
"score": 0.29655560921886226,
|
| 110 |
+
"eval_type": "rule",
|
| 111 |
+
"num_demo": 1,
|
| 112 |
+
"num_query": 14
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"name": "stackoverflow_debug_QA",
|
| 116 |
+
"score": 0.6666666666666666,
|
| 117 |
+
"eval_type": "rule",
|
| 118 |
+
"num_demo": 1,
|
| 119 |
+
"num_query": 14
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"name": "single_person_pose_estimation",
|
| 123 |
+
"score": 0.27073549247901185,
|
| 124 |
+
"eval_type": "rule",
|
| 125 |
+
"num_demo": 1,
|
| 126 |
+
"num_query": 14
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"name": "signage_navigation",
|
| 130 |
+
"score": 0.7142857142857143,
|
| 131 |
+
"eval_type": "rule",
|
| 132 |
+
"num_demo": 1,
|
| 133 |
+
"num_query": 14
|
| 134 |
+
},
|
| 135 |
+
{
|
| 136 |
+
"name": "geometry_area",
|
| 137 |
+
"score": 0.42857142857142855,
|
| 138 |
+
"eval_type": "rule",
|
| 139 |
+
"num_demo": 1,
|
| 140 |
+
"num_query": 14
|
| 141 |
+
},
|
| 142 |
+
{
|
| 143 |
+
"name": "hotel_booking_confirmation_parsing",
|
| 144 |
+
"score": 0.6857142857142857,
|
| 145 |
+
"eval_type": "rule",
|
| 146 |
+
"num_demo": 1,
|
| 147 |
+
"num_query": 14
|
| 148 |
+
},
|
| 149 |
+
{
|
| 150 |
+
"name": "ili_ratio_future_prediction",
|
| 151 |
+
"score": 0.15385714285714278,
|
| 152 |
+
"eval_type": "rule",
|
| 153 |
+
"num_demo": 1,
|
| 154 |
+
"num_query": 14
|
| 155 |
+
},
|
| 156 |
+
{
|
| 157 |
+
"name": "webpage_code_understanding",
|
| 158 |
+
"score": 0.6666666666666666,
|
| 159 |
+
"eval_type": "rule",
|
| 160 |
+
"num_demo": 1,
|
| 161 |
+
"num_query": 9
|
| 162 |
+
},
|
| 163 |
+
{
|
| 164 |
+
"name": "electricity_load_estimate_plot",
|
| 165 |
+
"score": 0.6818571428571429,
|
| 166 |
+
"eval_type": "rule",
|
| 167 |
+
"num_demo": 1,
|
| 168 |
+
"num_query": 14
|
| 169 |
+
},
|
| 170 |
+
{
|
| 171 |
+
"name": "tqa_textbook_qa",
|
| 172 |
+
"score": 0.9285714285714286,
|
| 173 |
+
"eval_type": "rule",
|
| 174 |
+
"num_demo": 1,
|
| 175 |
+
"num_query": 14
|
| 176 |
+
},
|
| 177 |
+
{
|
| 178 |
+
"name": "stock_info_parsing",
|
| 179 |
+
"score": 0.9663865546218489,
|
| 180 |
+
"eval_type": "rule",
|
| 181 |
+
"num_demo": 1,
|
| 182 |
+
"num_query": 14
|
| 183 |
+
},
|
| 184 |
+
{
|
| 185 |
+
"name": "medical_counting_lymphocytes",
|
| 186 |
+
"score": 0.14285714285714285,
|
| 187 |
+
"eval_type": "rule",
|
| 188 |
+
"num_demo": 1,
|
| 189 |
+
"num_query": 14
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"name": "quizlet_question_solving",
|
| 193 |
+
"score": 0.7142857142857143,
|
| 194 |
+
"eval_type": "rule",
|
| 195 |
+
"num_demo": 1,
|
| 196 |
+
"num_query": 14
|
| 197 |
+
},
|
| 198 |
+
{
|
| 199 |
+
"name": "icon_arithmetic_puzzle",
|
| 200 |
+
"score": 0.8928571428571429,
|
| 201 |
+
"eval_type": "rule",
|
| 202 |
+
"num_demo": 1,
|
| 203 |
+
"num_query": 14
|
| 204 |
+
},
|
| 205 |
+
{
|
| 206 |
+
"name": "planning_screenshot_grippers",
|
| 207 |
+
"score": 0.8,
|
| 208 |
+
"eval_type": "rule",
|
| 209 |
+
"num_demo": 1,
|
| 210 |
+
"num_query": 15
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"name": "stock_price_future_prediction",
|
| 214 |
+
"score": 0.8618571428571427,
|
| 215 |
+
"eval_type": "rule",
|
| 216 |
+
"num_demo": 1,
|
| 217 |
+
"num_query": 14
|
| 218 |
+
},
|
| 219 |
+
{
|
| 220 |
+
"name": "Ad_count_detection",
|
| 221 |
+
"score": 0.2857142857142857,
|
| 222 |
+
"eval_type": "rule",
|
| 223 |
+
"num_demo": 1,
|
| 224 |
+
"num_query": 14
|
| 225 |
+
},
|
| 226 |
+
{
|
| 227 |
+
"name": "MMMU_pro_exam_screenshot",
|
| 228 |
+
"score": 0.41414141414141414,
|
| 229 |
+
"eval_type": "rule",
|
| 230 |
+
"num_demo": 1,
|
| 231 |
+
"num_query": 99
|
| 232 |
+
},
|
| 233 |
+
{
|
| 234 |
+
"name": "recover_masked_word_in_figure",
|
| 235 |
+
"score": 0.2857142857142857,
|
| 236 |
+
"eval_type": "rule",
|
| 237 |
+
"num_demo": 1,
|
| 238 |
+
"num_query": 14
|
| 239 |
+
},
|
| 240 |
+
{
|
| 241 |
+
"name": "clevrer_physics",
|
| 242 |
+
"score": 0.5,
|
| 243 |
+
"eval_type": "rule",
|
| 244 |
+
"num_demo": 1,
|
| 245 |
+
"num_query": 20
|
| 246 |
+
},
|
| 247 |
+
{
|
| 248 |
+
"name": "polygon_interior_angles",
|
| 249 |
+
"score": 0.0,
|
| 250 |
+
"eval_type": "rule",
|
| 251 |
+
"num_demo": 1,
|
| 252 |
+
"num_query": 14
|
| 253 |
+
},
|
| 254 |
+
{
|
| 255 |
+
"name": "web_action_grounding",
|
| 256 |
+
"score": 0.6428571428571429,
|
| 257 |
+
"eval_type": "rule",
|
| 258 |
+
"num_demo": 1,
|
| 259 |
+
"num_query": 14
|
| 260 |
+
},
|
| 261 |
+
{
|
| 262 |
+
"name": "MMMU_physics_chemistry_selected",
|
| 263 |
+
"score": 0.7142857142857143,
|
| 264 |
+
"eval_type": "rule",
|
| 265 |
+
"num_demo": 1,
|
| 266 |
+
"num_query": 14
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"name": "table_understanding_complex_question_answering",
|
| 270 |
+
"score": 0.6428571428571429,
|
| 271 |
+
"eval_type": "rule",
|
| 272 |
+
"num_demo": 1,
|
| 273 |
+
"num_query": 14
|
| 274 |
+
},
|
| 275 |
+
{
|
| 276 |
+
"name": "table_understanding_fact_verification",
|
| 277 |
+
"score": 0.8333333333333333,
|
| 278 |
+
"eval_type": "rule",
|
| 279 |
+
"num_demo": 1,
|
| 280 |
+
"num_query": 14
|
| 281 |
+
},
|
| 282 |
+
{
|
| 283 |
+
"name": "panel_images_multi_question",
|
| 284 |
+
"score": 0.8571428571428571,
|
| 285 |
+
"eval_type": "rule",
|
| 286 |
+
"num_demo": 1,
|
| 287 |
+
"num_query": 14
|
| 288 |
+
},
|
| 289 |
+
{
|
| 290 |
+
"name": "panel_images_single_question",
|
| 291 |
+
"score": 0.8571428571428571,
|
| 292 |
+
"eval_type": "rule",
|
| 293 |
+
"num_demo": 1,
|
| 294 |
+
"num_query": 14
|
| 295 |
+
},
|
| 296 |
+
{
|
| 297 |
+
"name": "MMSoc_Misinformation_GossipCop",
|
| 298 |
+
"score": 0.6428571428571429,
|
| 299 |
+
"eval_type": "rule",
|
| 300 |
+
"num_demo": 1,
|
| 301 |
+
"num_query": 14
|
| 302 |
+
},
|
| 303 |
+
{
|
| 304 |
+
"name": "MMSoc_HatefulMemes",
|
| 305 |
+
"score": 0.7142857142857143,
|
| 306 |
+
"eval_type": "rule",
|
| 307 |
+
"num_demo": 1,
|
| 308 |
+
"num_query": 14
|
| 309 |
+
},
|
| 310 |
+
{
|
| 311 |
+
"name": "planning_screenshot_tyreworld",
|
| 312 |
+
"score": 0.6,
|
| 313 |
+
"eval_type": "rule",
|
| 314 |
+
"num_demo": 1,
|
| 315 |
+
"num_query": 15
|
| 316 |
+
},
|
| 317 |
+
{
|
| 318 |
+
"name": "MMSoc_Memotion",
|
| 319 |
+
"score": 0.5764705882352942,
|
| 320 |
+
"eval_type": "rule",
|
| 321 |
+
"num_demo": 1,
|
| 322 |
+
"num_query": 17
|
| 323 |
+
},
|
| 324 |
+
{
|
| 325 |
+
"name": "MMSoc_Misinformation_PolitiFact",
|
| 326 |
+
"score": 0.7857142857142857,
|
| 327 |
+
"eval_type": "rule",
|
| 328 |
+
"num_demo": 1,
|
| 329 |
+
"num_query": 14
|
| 330 |
+
},
|
| 331 |
+
{
|
| 332 |
+
"name": "chess_puzzles_checkmate",
|
| 333 |
+
"score": 0.0,
|
| 334 |
+
"eval_type": "rule",
|
| 335 |
+
"num_demo": 1,
|
| 336 |
+
"num_query": 14
|
| 337 |
+
},
|
| 338 |
+
{
|
| 339 |
+
"name": "poetry_acrostic_alliteration",
|
| 340 |
+
"score": 0.0,
|
| 341 |
+
"eval_type": "rule",
|
| 342 |
+
"num_demo": 0,
|
| 343 |
+
"num_query": 15
|
| 344 |
+
},
|
| 345 |
+
{
|
| 346 |
+
"name": "poetry_acrostic",
|
| 347 |
+
"score": 0.0,
|
| 348 |
+
"eval_type": "rule",
|
| 349 |
+
"num_demo": 0,
|
| 350 |
+
"num_query": 15
|
| 351 |
+
},
|
| 352 |
+
{
|
| 353 |
+
"name": "chess_puzzles_equality",
|
| 354 |
+
"score": 0.0,
|
| 355 |
+
"eval_type": "rule",
|
| 356 |
+
"num_demo": 1,
|
| 357 |
+
"num_query": 15
|
| 358 |
+
},
|
| 359 |
+
{
|
| 360 |
+
"name": "poetry_limerick",
|
| 361 |
+
"score": 0.0,
|
| 362 |
+
"eval_type": "rule",
|
| 363 |
+
"num_demo": 0,
|
| 364 |
+
"num_query": 15
|
| 365 |
+
},
|
| 366 |
+
{
|
| 367 |
+
"name": "app_layout_understanding_notes",
|
| 368 |
+
"score": 0.6428571428571429,
|
| 369 |
+
"eval_type": "rule",
|
| 370 |
+
"num_demo": 1,
|
| 371 |
+
"num_query": 14
|
| 372 |
+
},
|
| 373 |
+
{
|
| 374 |
+
"name": "app_layout_understanding_twitter",
|
| 375 |
+
"score": 0.7142857142857143,
|
| 376 |
+
"eval_type": "rule",
|
| 377 |
+
"num_demo": 1,
|
| 378 |
+
"num_query": 14
|
| 379 |
+
},
|
| 380 |
+
{
|
| 381 |
+
"name": "app_layout_understanding_youtube",
|
| 382 |
+
"score": 0.7857142857142857,
|
| 383 |
+
"eval_type": "rule",
|
| 384 |
+
"num_demo": 1,
|
| 385 |
+
"num_query": 14
|
| 386 |
+
},
|
| 387 |
+
{
|
| 388 |
+
"name": "app_layout_understanding_tiktok",
|
| 389 |
+
"score": 0.6428571428571429,
|
| 390 |
+
"eval_type": "rule",
|
| 391 |
+
"num_demo": 1,
|
| 392 |
+
"num_query": 14
|
| 393 |
+
},
|
| 394 |
+
{
|
| 395 |
+
"name": "app_layout_understanding_excel",
|
| 396 |
+
"score": 0.7857142857142857,
|
| 397 |
+
"eval_type": "rule",
|
| 398 |
+
"num_demo": 1,
|
| 399 |
+
"num_query": 14
|
| 400 |
+
},
|
| 401 |
+
{
|
| 402 |
+
"name": "app_layout_understanding_amazon",
|
| 403 |
+
"score": 0.6428571428571429,
|
| 404 |
+
"eval_type": "rule",
|
| 405 |
+
"num_demo": 1,
|
| 406 |
+
"num_query": 14
|
| 407 |
+
},
|
| 408 |
+
{
|
| 409 |
+
"name": "app_layout_understanding_instagram",
|
| 410 |
+
"score": 0.8571428571428571,
|
| 411 |
+
"eval_type": "rule",
|
| 412 |
+
"num_demo": 1,
|
| 413 |
+
"num_query": 14
|
| 414 |
+
},
|
| 415 |
+
{
|
| 416 |
+
"name": "app_layout_understanding_zoom",
|
| 417 |
+
"score": 0.6666666666666666,
|
| 418 |
+
"eval_type": "rule",
|
| 419 |
+
"num_demo": 1,
|
| 420 |
+
"num_query": 15
|
| 421 |
+
},
|
| 422 |
+
{
|
| 423 |
+
"name": "app_layout_understanding_word",
|
| 424 |
+
"score": 0.5714285714285714,
|
| 425 |
+
"eval_type": "rule",
|
| 426 |
+
"num_demo": 1,
|
| 427 |
+
"num_query": 14
|
| 428 |
+
},
|
| 429 |
+
{
|
| 430 |
+
"name": "app_layout_understanding_iphone_settings",
|
| 431 |
+
"score": 0.8571428571428571,
|
| 432 |
+
"eval_type": "rule",
|
| 433 |
+
"num_demo": 1,
|
| 434 |
+
"num_query": 14
|
| 435 |
+
},
|
| 436 |
+
{
|
| 437 |
+
"name": "app_layout_understanding_leetcode",
|
| 438 |
+
"score": 0.7857142857142857,
|
| 439 |
+
"eval_type": "rule",
|
| 440 |
+
"num_demo": 1,
|
| 441 |
+
"num_query": 14
|
| 442 |
+
},
|
| 443 |
+
{
|
| 444 |
+
"name": "app_layout_understanding_ppt",
|
| 445 |
+
"score": 0.7142857142857143,
|
| 446 |
+
"eval_type": "rule",
|
| 447 |
+
"num_demo": 1,
|
| 448 |
+
"num_query": 14
|
| 449 |
+
},
|
| 450 |
+
{
|
| 451 |
+
"name": "app_layout_understanding_alipay",
|
| 452 |
+
"score": 0.7647058823529411,
|
| 453 |
+
"eval_type": "rule",
|
| 454 |
+
"num_demo": 1,
|
| 455 |
+
"num_query": 17
|
| 456 |
+
},
|
| 457 |
+
{
|
| 458 |
+
"name": "ocr_table_to_markdown",
|
| 459 |
+
"score": 0.8571428571428571,
|
| 460 |
+
"eval_type": "rule",
|
| 461 |
+
"num_demo": 1,
|
| 462 |
+
"num_query": 14
|
| 463 |
+
},
|
| 464 |
+
{
|
| 465 |
+
"name": "ocr_table_to_latex",
|
| 466 |
+
"score": 0.6428571428571429,
|
| 467 |
+
"eval_type": "rule",
|
| 468 |
+
"num_demo": 1,
|
| 469 |
+
"num_query": 14
|
| 470 |
+
},
|
| 471 |
+
{
|
| 472 |
+
"name": "ocr_resume_employer_plain",
|
| 473 |
+
"score": 0.5714285714285714,
|
| 474 |
+
"eval_type": "rule",
|
| 475 |
+
"num_demo": 1,
|
| 476 |
+
"num_query": 14
|
| 477 |
+
},
|
| 478 |
+
{
|
| 479 |
+
"name": "ocr_article_journal",
|
| 480 |
+
"score": 0.5714285714285714,
|
| 481 |
+
"eval_type": "rule",
|
| 482 |
+
"num_demo": 1,
|
| 483 |
+
"num_query": 14
|
| 484 |
+
},
|
| 485 |
+
{
|
| 486 |
+
"name": "ocr_resume_experience_plain",
|
| 487 |
+
"score": 0.5,
|
| 488 |
+
"eval_type": "rule",
|
| 489 |
+
"num_demo": 1,
|
| 490 |
+
"num_query": 14
|
| 491 |
+
},
|
| 492 |
+
{
|
| 493 |
+
"name": "ocr_math_text_latex",
|
| 494 |
+
"score": 0.5,
|
| 495 |
+
"eval_type": "rule",
|
| 496 |
+
"num_demo": 1,
|
| 497 |
+
"num_query": 14
|
| 498 |
+
},
|
| 499 |
+
{
|
| 500 |
+
"name": "ocr_article_authors",
|
| 501 |
+
"score": 0.9285714285714286,
|
| 502 |
+
"eval_type": "rule",
|
| 503 |
+
"num_demo": 1,
|
| 504 |
+
"num_query": 14
|
| 505 |
+
},
|
| 506 |
+
{
|
| 507 |
+
"name": "ocr_table_to_csv",
|
| 508 |
+
"score": 0.5714285714285714,
|
| 509 |
+
"eval_type": "rule",
|
| 510 |
+
"num_demo": 1,
|
| 511 |
+
"num_query": 14
|
| 512 |
+
},
|
| 513 |
+
{
|
| 514 |
+
"name": "ocr_math_equation",
|
| 515 |
+
"score": 0.6428571428571429,
|
| 516 |
+
"eval_type": "rule",
|
| 517 |
+
"num_demo": 1,
|
| 518 |
+
"num_query": 14
|
| 519 |
+
},
|
| 520 |
+
{
|
| 521 |
+
"name": "ocr_resume_school_plain",
|
| 522 |
+
"score": 0.6428571428571429,
|
| 523 |
+
"eval_type": "rule",
|
| 524 |
+
"num_demo": 1,
|
| 525 |
+
"num_query": 14
|
| 526 |
+
},
|
| 527 |
+
{
|
| 528 |
+
"name": "ocr_table_to_html",
|
| 529 |
+
"score": 0.7142857142857143,
|
| 530 |
+
"eval_type": "rule",
|
| 531 |
+
"num_demo": 1,
|
| 532 |
+
"num_query": 14
|
| 533 |
+
},
|
| 534 |
+
{
|
| 535 |
+
"name": "ocr_resume_skill_plain",
|
| 536 |
+
"score": 0.6428571428571429,
|
| 537 |
+
"eval_type": "rule",
|
| 538 |
+
"num_demo": 1,
|
| 539 |
+
"num_query": 14
|
| 540 |
+
},
|
| 541 |
+
{
|
| 542 |
+
"name": "crossword_mini_5x5",
|
| 543 |
+
"score": 0.65,
|
| 544 |
+
"eval_type": "rule",
|
| 545 |
+
"num_demo": 1,
|
| 546 |
+
"num_query": 14
|
| 547 |
+
},
|
| 548 |
+
{
|
| 549 |
+
"name": "contain_position_length",
|
| 550 |
+
"score": 0.2,
|
| 551 |
+
"eval_type": "rule",
|
| 552 |
+
"num_demo": 0,
|
| 553 |
+
"num_query": 15
|
| 554 |
+
},
|
| 555 |
+
{
|
| 556 |
+
"name": "contain_repeat_length",
|
| 557 |
+
"score": 0.0,
|
| 558 |
+
"eval_type": "rule",
|
| 559 |
+
"num_demo": 0,
|
| 560 |
+
"num_query": 15
|
| 561 |
+
},
|
| 562 |
+
{
|
| 563 |
+
"name": "contain_length",
|
| 564 |
+
"score": 0.26666666666666666,
|
| 565 |
+
"eval_type": "rule",
|
| 566 |
+
"num_demo": 0,
|
| 567 |
+
"num_query": 15
|
| 568 |
+
},
|
| 569 |
+
{
|
| 570 |
+
"name": "contain_contain_length",
|
| 571 |
+
"score": 0.7333333333333333,
|
| 572 |
+
"eval_type": "rule",
|
| 573 |
+
"num_demo": 0,
|
| 574 |
+
"num_query": 15
|
| 575 |
+
},
|
| 576 |
+
{
|
| 577 |
+
"name": "pictionary_skribbl_io",
|
| 578 |
+
"score": 0.35,
|
| 579 |
+
"eval_type": "rule",
|
| 580 |
+
"num_demo": 1,
|
| 581 |
+
"num_query": 20
|
| 582 |
+
},
|
| 583 |
+
{
|
| 584 |
+
"name": "pictionary_doodle_guess",
|
| 585 |
+
"score": 0.7333333333333333,
|
| 586 |
+
"eval_type": "rule",
|
| 587 |
+
"num_demo": 1,
|
| 588 |
+
"num_query": 15
|
| 589 |
+
},
|
| 590 |
+
{
|
| 591 |
+
"name": "pictionary_genai_output_chinese",
|
| 592 |
+
"score": 0.35714285714285715,
|
| 593 |
+
"eval_type": "rule",
|
| 594 |
+
"num_demo": 1,
|
| 595 |
+
"num_query": 14
|
| 596 |
+
},
|
| 597 |
+
{
|
| 598 |
+
"name": "pictionary_cartoon_drawing_guess",
|
| 599 |
+
"score": 0.8571428571428571,
|
| 600 |
+
"eval_type": "rule",
|
| 601 |
+
"num_demo": 1,
|
| 602 |
+
"num_query": 14
|
| 603 |
+
},
|
| 604 |
+
{
|
| 605 |
+
"name": "pictionary_chinese_food_img2en",
|
| 606 |
+
"score": 0.7142857142857143,
|
| 607 |
+
"eval_type": "rule",
|
| 608 |
+
"num_demo": 1,
|
| 609 |
+
"num_query": 14
|
| 610 |
+
},
|
| 611 |
+
{
|
| 612 |
+
"name": "brand_logo_recognition_and_elaboration",
|
| 613 |
+
"score": 0.86,
|
| 614 |
+
"eval_type": "rule",
|
| 615 |
+
"num_demo": 1,
|
| 616 |
+
"num_query": 25
|
| 617 |
+
},
|
| 618 |
+
{
|
| 619 |
+
"name": "exchange_rate_estimate_plot",
|
| 620 |
+
"score": 0.9678642857142855,
|
| 621 |
+
"eval_type": "rule",
|
| 622 |
+
"num_demo": 1,
|
| 623 |
+
"num_query": 14
|
| 624 |
+
},
|
| 625 |
+
{
|
| 626 |
+
"name": "geometry_analytic",
|
| 627 |
+
"score": 0.2857142857142857,
|
| 628 |
+
"eval_type": "rule",
|
| 629 |
+
"num_demo": 1,
|
| 630 |
+
"num_query": 14
|
| 631 |
+
},
|
| 632 |
+
{
|
| 633 |
+
"name": "math_parity",
|
| 634 |
+
"score": 0.8,
|
| 635 |
+
"eval_type": "rule",
|
| 636 |
+
"num_demo": 1,
|
| 637 |
+
"num_query": 15
|
| 638 |
+
},
|
| 639 |
+
{
|
| 640 |
+
"name": "nlvr2_two_image_compare_qa",
|
| 641 |
+
"score": 0.7857142857142857,
|
| 642 |
+
"eval_type": "rule",
|
| 643 |
+
"num_demo": 1,
|
| 644 |
+
"num_query": 14
|
| 645 |
+
},
|
| 646 |
+
{
|
| 647 |
+
"name": "traffic_future_prediction_from_line_plot",
|
| 648 |
+
"score": 0.5982105263157894,
|
| 649 |
+
"eval_type": "rule",
|
| 650 |
+
"num_demo": 1,
|
| 651 |
+
"num_query": 19
|
| 652 |
+
},
|
| 653 |
+
{
|
| 654 |
+
"name": "math_exams_v",
|
| 655 |
+
"score": 0.5714285714285714,
|
| 656 |
+
"eval_type": "rule",
|
| 657 |
+
"num_demo": 1,
|
| 658 |
+
"num_query": 14
|
| 659 |
+
},
|
| 660 |
+
{
|
| 661 |
+
"name": "newspaper_ocr_in_query_box",
|
| 662 |
+
"score": 0.13333333333333333,
|
| 663 |
+
"eval_type": "rule",
|
| 664 |
+
"num_demo": 1,
|
| 665 |
+
"num_query": 15
|
| 666 |
+
},
|
| 667 |
+
{
|
| 668 |
+
"name": "mvsa_sentiment_classification",
|
| 669 |
+
"score": 0.7142857142857143,
|
| 670 |
+
"eval_type": "rule",
|
| 671 |
+
"num_demo": 1,
|
| 672 |
+
"num_query": 14
|
| 673 |
+
},
|
| 674 |
+
{
|
| 675 |
+
"name": "graph_chordless_cycle",
|
| 676 |
+
"score": 0.14285714285714285,
|
| 677 |
+
"eval_type": "rule",
|
| 678 |
+
"num_demo": 1,
|
| 679 |
+
"num_query": 14
|
| 680 |
+
},
|
| 681 |
+
{
|
| 682 |
+
"name": "youtube_video_info_parsing",
|
| 683 |
+
"score": 0.773809523809524,
|
| 684 |
+
"eval_type": "rule",
|
| 685 |
+
"num_demo": 1,
|
| 686 |
+
"num_query": 14
|
| 687 |
+
},
|
| 688 |
+
{
|
| 689 |
+
"name": "super_clevr_scene_understanding",
|
| 690 |
+
"score": 0.5,
|
| 691 |
+
"eval_type": "rule",
|
| 692 |
+
"num_demo": 1,
|
| 693 |
+
"num_query": 14
|
| 694 |
+
},
|
| 695 |
+
{
|
| 696 |
+
"name": "monthly_weather_days_count",
|
| 697 |
+
"score": 0.380952380952381,
|
| 698 |
+
"eval_type": "rule",
|
| 699 |
+
"num_demo": 1,
|
| 700 |
+
"num_query": 14
|
| 701 |
+
},
|
| 702 |
+
{
|
| 703 |
+
"name": "weather_map_climate_type_temperature_parsing",
|
| 704 |
+
"score": 0.6785714285714286,
|
| 705 |
+
"eval_type": "rule",
|
| 706 |
+
"num_demo": 1,
|
| 707 |
+
"num_query": 14
|
| 708 |
+
},
|
| 709 |
+
{
|
| 710 |
+
"name": "top_video_creator_identification",
|
| 711 |
+
"score": 0.5714285714285714,
|
| 712 |
+
"eval_type": "rule",
|
| 713 |
+
"num_demo": 1,
|
| 714 |
+
"num_query": 14
|
| 715 |
+
},
|
| 716 |
+
{
|
| 717 |
+
"name": "actor_recognition_in_Movie",
|
| 718 |
+
"score": 0.8571428571428571,
|
| 719 |
+
"eval_type": "rule",
|
| 720 |
+
"num_demo": 1,
|
| 721 |
+
"num_query": 14
|
| 722 |
+
},
|
| 723 |
+
{
|
| 724 |
+
"name": "rebus",
|
| 725 |
+
"score": 0.30434782608695654,
|
| 726 |
+
"eval_type": "rule",
|
| 727 |
+
"num_demo": 1,
|
| 728 |
+
"num_query": 23
|
| 729 |
+
},
|
| 730 |
+
{
|
| 731 |
+
"name": "extract_webpage_headline",
|
| 732 |
+
"score": 0.5,
|
| 733 |
+
"eval_type": "rule",
|
| 734 |
+
"num_demo": 1,
|
| 735 |
+
"num_query": 14
|
| 736 |
+
},
|
| 737 |
+
{
|
| 738 |
+
"name": "ishihara_test",
|
| 739 |
+
"score": 0.09999999999999999,
|
| 740 |
+
"eval_type": "rule",
|
| 741 |
+
"num_demo": 1,
|
| 742 |
+
"num_query": 14
|
| 743 |
+
},
|
| 744 |
+
{
|
| 745 |
+
"name": "paper_vqa",
|
| 746 |
+
"score": 0.35714285714285715,
|
| 747 |
+
"eval_type": "rule",
|
| 748 |
+
"num_demo": 1,
|
| 749 |
+
"num_query": 14
|
| 750 |
+
},
|
| 751 |
+
{
|
| 752 |
+
"name": "planning_screenshot_storage",
|
| 753 |
+
"score": 0.0,
|
| 754 |
+
"eval_type": "rule",
|
| 755 |
+
"num_demo": 1,
|
| 756 |
+
"num_query": 15
|
| 757 |
+
},
|
| 758 |
+
{
|
| 759 |
+
"name": "soccer_offside",
|
| 760 |
+
"score": 0.4444444444444444,
|
| 761 |
+
"eval_type": "rule",
|
| 762 |
+
"num_demo": 1,
|
| 763 |
+
"num_query": 9
|
| 764 |
+
},
|
| 765 |
+
{
|
| 766 |
+
"name": "geometry_reasoning_grid",
|
| 767 |
+
"score": 0.7142857142857143,
|
| 768 |
+
"eval_type": "rule",
|
| 769 |
+
"num_demo": 1,
|
| 770 |
+
"num_query": 14
|
| 771 |
+
},
|
| 772 |
+
{
|
| 773 |
+
"name": "relative_reflectance_of_different_regions",
|
| 774 |
+
"score": 0.5,
|
| 775 |
+
"eval_type": "rule",
|
| 776 |
+
"num_demo": 1,
|
| 777 |
+
"num_query": 14
|
| 778 |
+
},
|
| 779 |
+
{
|
| 780 |
+
"name": "entertainment_web_game_style",
|
| 781 |
+
"score": 0.8214285714285714,
|
| 782 |
+
"eval_type": "rule",
|
| 783 |
+
"num_demo": 1,
|
| 784 |
+
"num_query": 14
|
| 785 |
+
},
|
| 786 |
+
{
|
| 787 |
+
"name": "orchestra_score_recognition",
|
| 788 |
+
"score": 0.07142857142857142,
|
| 789 |
+
"eval_type": "rule",
|
| 790 |
+
"num_demo": 1,
|
| 791 |
+
"num_query": 14
|
| 792 |
+
},
|
| 793 |
+
{
|
| 794 |
+
"name": "music_sheet_note_count",
|
| 795 |
+
"score": 0.058823529411764705,
|
| 796 |
+
"eval_type": "rule",
|
| 797 |
+
"num_demo": 1,
|
| 798 |
+
"num_query": 17
|
| 799 |
+
},
|
| 800 |
+
{
|
| 801 |
+
"name": "game_platform_support_identification",
|
| 802 |
+
"score": 0.9642857142857143,
|
| 803 |
+
"eval_type": "rule",
|
| 804 |
+
"num_demo": 1,
|
| 805 |
+
"num_query": 14
|
| 806 |
+
},
|
| 807 |
+
{
|
| 808 |
+
"name": "GUI_Act_Mobile_swipe",
|
| 809 |
+
"score": 0.4205673337417017,
|
| 810 |
+
"eval_type": "rule",
|
| 811 |
+
"num_demo": 1,
|
| 812 |
+
"num_query": 13
|
| 813 |
+
},
|
| 814 |
+
{
|
| 815 |
+
"name": "mahjong",
|
| 816 |
+
"score": 0.14285714285714285,
|
| 817 |
+
"eval_type": "rule",
|
| 818 |
+
"num_demo": 1,
|
| 819 |
+
"num_query": 14
|
| 820 |
+
},
|
| 821 |
+
{
|
| 822 |
+
"name": "scibench_calculus_wo_solution",
|
| 823 |
+
"score": 0.32653061224489793,
|
| 824 |
+
"eval_type": "rule",
|
| 825 |
+
"num_demo": 1,
|
| 826 |
+
"num_query": 49
|
| 827 |
+
},
|
| 828 |
+
{
|
| 829 |
+
"name": "knowledge_graph_understanding",
|
| 830 |
+
"score": 0.6,
|
| 831 |
+
"eval_type": "rule",
|
| 832 |
+
"num_demo": 1,
|
| 833 |
+
"num_query": 15
|
| 834 |
+
},
|
| 835 |
+
{
|
| 836 |
+
"name": "image_translation_en2cn",
|
| 837 |
+
"score": 0.39500840600047465,
|
| 838 |
+
"eval_type": "rule",
|
| 839 |
+
"num_demo": 1,
|
| 840 |
+
"num_query": 9
|
| 841 |
+
},
|
| 842 |
+
{
|
| 843 |
+
"name": "realworld_qa_en2cn",
|
| 844 |
+
"score": 0.8571428571428571,
|
| 845 |
+
"eval_type": "rule",
|
| 846 |
+
"num_demo": 1,
|
| 847 |
+
"num_query": 14
|
| 848 |
+
},
|
| 849 |
+
{
|
| 850 |
+
"name": "planning_visual_storage",
|
| 851 |
+
"score": 0.0,
|
| 852 |
+
"eval_type": "rule",
|
| 853 |
+
"num_demo": 1,
|
| 854 |
+
"num_query": 15
|
| 855 |
+
},
|
| 856 |
+
{
|
| 857 |
+
"name": "GUI_Act_Web_Multi",
|
| 858 |
+
"score": 0.4683992228805717,
|
| 859 |
+
"eval_type": "rule",
|
| 860 |
+
"num_demo": 1,
|
| 861 |
+
"num_query": 14
|
| 862 |
+
},
|
| 863 |
+
{
|
| 864 |
+
"name": "chinese_idiom_recognition",
|
| 865 |
+
"score": 0.7142857142857143,
|
| 866 |
+
"eval_type": "rule",
|
| 867 |
+
"num_demo": 1,
|
| 868 |
+
"num_query": 14
|
| 869 |
+
},
|
| 870 |
+
{
|
| 871 |
+
"name": "number_comparison",
|
| 872 |
+
"score": 1.0,
|
| 873 |
+
"eval_type": "rule",
|
| 874 |
+
"num_demo": 1,
|
| 875 |
+
"num_query": 14
|
| 876 |
+
},
|
| 877 |
+
{
|
| 878 |
+
"name": "planning_screenshot_blocksworld",
|
| 879 |
+
"score": 0.13333333333333333,
|
| 880 |
+
"eval_type": "rule",
|
| 881 |
+
"num_demo": 1,
|
| 882 |
+
"num_query": 15
|
| 883 |
+
},
|
| 884 |
+
{
|
| 885 |
+
"name": "product_ocr_qa",
|
| 886 |
+
"score": 0.5714285714285714,
|
| 887 |
+
"eval_type": "rule",
|
| 888 |
+
"num_demo": 1,
|
| 889 |
+
"num_query": 14
|
| 890 |
+
},
|
| 891 |
+
{
|
| 892 |
+
"name": "geometry_reasoning_circled_letter",
|
| 893 |
+
"score": 0.6785714285714286,
|
| 894 |
+
"eval_type": "rule",
|
| 895 |
+
"num_demo": 1,
|
| 896 |
+
"num_query": 14
|
| 897 |
+
},
|
| 898 |
+
{
|
| 899 |
+
"name": "GUI_Act_Web_Single",
|
| 900 |
+
"score": 0.038819875776397506,
|
| 901 |
+
"eval_type": "rule",
|
| 902 |
+
"num_demo": 1,
|
| 903 |
+
"num_query": 14
|
| 904 |
+
},
|
| 905 |
+
{
|
| 906 |
+
"name": "latex_complex_formula_convertion",
|
| 907 |
+
"score": 0.35294117647058826,
|
| 908 |
+
"eval_type": "rule",
|
| 909 |
+
"num_demo": 1,
|
| 910 |
+
"num_query": 17
|
| 911 |
+
},
|
| 912 |
+
{
|
| 913 |
+
"name": "movie_info_parsing",
|
| 914 |
+
"score": 0.7142857142857143,
|
| 915 |
+
"eval_type": "rule",
|
| 916 |
+
"num_demo": 1,
|
| 917 |
+
"num_query": 14
|
| 918 |
+
},
|
| 919 |
+
{
|
| 920 |
+
"name": "transit_map_intersection_points",
|
| 921 |
+
"score": 0.4107142857142857,
|
| 922 |
+
"eval_type": "rule",
|
| 923 |
+
"num_demo": 1,
|
| 924 |
+
"num_query": 14
|
| 925 |
+
},
|
| 926 |
+
{
|
| 927 |
+
"name": "arxiv_vqa",
|
| 928 |
+
"score": 0.8571428571428571,
|
| 929 |
+
"eval_type": "rule",
|
| 930 |
+
"num_demo": 1,
|
| 931 |
+
"num_query": 14
|
| 932 |
+
},
|
| 933 |
+
{
|
| 934 |
+
"name": "human_relationship_reasoning",
|
| 935 |
+
"score": 0.9285714285714286,
|
| 936 |
+
"eval_type": "rule",
|
| 937 |
+
"num_demo": 1,
|
| 938 |
+
"num_query": 14
|
| 939 |
+
},
|
| 940 |
+
{
|
| 941 |
+
"name": "graph_shortest_path_kamada_kawai",
|
| 942 |
+
"score": 0.14285714285714285,
|
| 943 |
+
"eval_type": "rule",
|
| 944 |
+
"num_demo": 1,
|
| 945 |
+
"num_query": 14
|
| 946 |
+
},
|
| 947 |
+
{
|
| 948 |
+
"name": "coco_person_detection",
|
| 949 |
+
"score": 0.586989127272571,
|
| 950 |
+
"eval_type": "rule",
|
| 951 |
+
"num_demo": 1,
|
| 952 |
+
"num_query": 14
|
| 953 |
+
},
|
| 954 |
+
{
|
| 955 |
+
"name": "medical_image_artifacts_indentification",
|
| 956 |
+
"score": 0.14285714285714285,
|
| 957 |
+
"eval_type": "rule",
|
| 958 |
+
"num_demo": 1,
|
| 959 |
+
"num_query": 14
|
| 960 |
+
},
|
| 961 |
+
{
|
| 962 |
+
"name": "song_title_identification_from_lyrics",
|
| 963 |
+
"score": 0.6428571428571429,
|
| 964 |
+
"eval_type": "rule",
|
| 965 |
+
"num_demo": 1,
|
| 966 |
+
"num_query": 14
|
| 967 |
+
},
|
| 968 |
+
{
|
| 969 |
+
"name": "chart_vqa",
|
| 970 |
+
"score": 0.7857142857142857,
|
| 971 |
+
"eval_type": "rule",
|
| 972 |
+
"num_demo": 1,
|
| 973 |
+
"num_query": 14
|
| 974 |
+
},
|
| 975 |
+
{
|
| 976 |
+
"name": "egocentric_spatial_reasoning",
|
| 977 |
+
"score": 0.5555555555555556,
|
| 978 |
+
"eval_type": "rule",
|
| 979 |
+
"num_demo": 1,
|
| 980 |
+
"num_query": 9
|
| 981 |
+
},
|
| 982 |
+
{
|
| 983 |
+
"name": "bongard_problem",
|
| 984 |
+
"score": 0.42105263157894735,
|
| 985 |
+
"eval_type": "rule",
|
| 986 |
+
"num_demo": 1,
|
| 987 |
+
"num_query": 19
|
| 988 |
+
},
|
| 989 |
+
{
|
| 990 |
+
"name": "ascii_art_understanding",
|
| 991 |
+
"score": 0.42857142857142855,
|
| 992 |
+
"eval_type": "rule",
|
| 993 |
+
"num_demo": 1,
|
| 994 |
+
"num_query": 14
|
| 995 |
+
},
|
| 996 |
+
{
|
| 997 |
+
"name": "calendar_schedule_suggestion",
|
| 998 |
+
"score": 0.35714285714285715,
|
| 999 |
+
"eval_type": "rule",
|
| 1000 |
+
"num_demo": 1,
|
| 1001 |
+
"num_query": 14
|
| 1002 |
+
},
|
| 1003 |
+
{
|
| 1004 |
+
"name": "geometry_reasoning_overlapped_circle",
|
| 1005 |
+
"score": 0.7857142857142857,
|
| 1006 |
+
"eval_type": "rule",
|
| 1007 |
+
"num_demo": 1,
|
| 1008 |
+
"num_query": 14
|
| 1009 |
+
},
|
| 1010 |
+
{
|
| 1011 |
+
"name": "planning_screenshot_barman",
|
| 1012 |
+
"score": 0.13333333333333333,
|
| 1013 |
+
"eval_type": "rule",
|
| 1014 |
+
"num_demo": 1,
|
| 1015 |
+
"num_query": 15
|
| 1016 |
+
},
|
| 1017 |
+
{
|
| 1018 |
+
"name": "planning_screenshot_floortile",
|
| 1019 |
+
"score": 0.0,
|
| 1020 |
+
"eval_type": "rule",
|
| 1021 |
+
"num_demo": 1,
|
| 1022 |
+
"num_query": 15
|
| 1023 |
+
},
|
| 1024 |
+
{
|
| 1025 |
+
"name": "medical_blood_vessels_recognition",
|
| 1026 |
+
"score": 0.75,
|
| 1027 |
+
"eval_type": "rule",
|
| 1028 |
+
"num_demo": 1,
|
| 1029 |
+
"num_query": 14
|
| 1030 |
+
},
|
| 1031 |
+
{
|
| 1032 |
+
"name": "logical_reasoning_find_odd_one_out",
|
| 1033 |
+
"score": 0.17857142857142858,
|
| 1034 |
+
"eval_type": "rule",
|
| 1035 |
+
"num_demo": 1,
|
| 1036 |
+
"num_query": 14
|
| 1037 |
+
},
|
| 1038 |
+
{
|
| 1039 |
+
"name": "graph_isomorphism",
|
| 1040 |
+
"score": 0.5333333333333333,
|
| 1041 |
+
"eval_type": "rule",
|
| 1042 |
+
"num_demo": 1,
|
| 1043 |
+
"num_query": 15
|
| 1044 |
+
},
|
| 1045 |
+
{
|
| 1046 |
+
"name": "web_action_prediction",
|
| 1047 |
+
"score": 0.8571428571428571,
|
| 1048 |
+
"eval_type": "rule",
|
| 1049 |
+
"num_demo": 1,
|
| 1050 |
+
"num_query": 14
|
| 1051 |
+
},
|
| 1052 |
+
{
|
| 1053 |
+
"name": "code_programming_test_easy",
|
| 1054 |
+
"score": 0.5694444444444444,
|
| 1055 |
+
"eval_type": "rule",
|
| 1056 |
+
"num_demo": 1,
|
| 1057 |
+
"num_query": 24
|
| 1058 |
+
},
|
| 1059 |
+
{
|
| 1060 |
+
"name": "code_execution",
|
| 1061 |
+
"score": 0.625,
|
| 1062 |
+
"eval_type": "rule",
|
| 1063 |
+
"num_demo": 1,
|
| 1064 |
+
"num_query": 16
|
| 1065 |
+
},
|
| 1066 |
+
{
|
| 1067 |
+
"name": "remaining_playback_time_calculation",
|
| 1068 |
+
"score": 0.0,
|
| 1069 |
+
"eval_type": "rule",
|
| 1070 |
+
"num_demo": 1,
|
| 1071 |
+
"num_query": 14
|
| 1072 |
+
},
|
| 1073 |
+
{
|
| 1074 |
+
"name": "medical_cell_recognition",
|
| 1075 |
+
"score": 0.5,
|
| 1076 |
+
"eval_type": "rule",
|
| 1077 |
+
"num_demo": 1,
|
| 1078 |
+
"num_query": 14
|
| 1079 |
+
},
|
| 1080 |
+
{
|
| 1081 |
+
"name": "chess_find_legal_moves",
|
| 1082 |
+
"score": 0.2134015655782666,
|
| 1083 |
+
"eval_type": "rule",
|
| 1084 |
+
"num_demo": 1,
|
| 1085 |
+
"num_query": 14
|
| 1086 |
+
},
|
| 1087 |
+
{
|
| 1088 |
+
"name": "relative_depth_of_different_points",
|
| 1089 |
+
"score": 0.7142857142857143,
|
| 1090 |
+
"eval_type": "rule",
|
| 1091 |
+
"num_demo": 1,
|
| 1092 |
+
"num_query": 14
|
| 1093 |
+
},
|
| 1094 |
+
{
|
| 1095 |
+
"name": "distinguish_ai_generated_image",
|
| 1096 |
+
"score": 0.7894736842105263,
|
| 1097 |
+
"eval_type": "rule",
|
| 1098 |
+
"num_demo": 1,
|
| 1099 |
+
"num_query": 19
|
| 1100 |
+
},
|
| 1101 |
+
{
|
| 1102 |
+
"name": "autonomous_driving_scene_analysis",
|
| 1103 |
+
"score": 0.9285714285714286,
|
| 1104 |
+
"eval_type": "rule",
|
| 1105 |
+
"num_demo": 1,
|
| 1106 |
+
"num_query": 14
|
| 1107 |
+
},
|
| 1108 |
+
{
|
| 1109 |
+
"name": "counting_single_image",
|
| 1110 |
+
"score": 0.6428571428571429,
|
| 1111 |
+
"eval_type": "rule",
|
| 1112 |
+
"num_demo": 1,
|
| 1113 |
+
"num_query": 14
|
| 1114 |
+
},
|
| 1115 |
+
{
|
| 1116 |
+
"name": "GUI_Act_Mobile_tap",
|
| 1117 |
+
"score": 0.14285714285714285,
|
| 1118 |
+
"eval_type": "rule",
|
| 1119 |
+
"num_demo": 1,
|
| 1120 |
+
"num_query": 14
|
| 1121 |
+
},
|
| 1122 |
+
{
|
| 1123 |
+
"name": "road_map_find_highway_between_two_place",
|
| 1124 |
+
"score": 0.7647058823529411,
|
| 1125 |
+
"eval_type": "rule",
|
| 1126 |
+
"num_demo": 1,
|
| 1127 |
+
"num_query": 17
|
| 1128 |
+
},
|
| 1129 |
+
{
|
| 1130 |
+
"name": "chess_sygyzy_endgames",
|
| 1131 |
+
"score": 0.14875283446712015,
|
| 1132 |
+
"eval_type": "rule",
|
| 1133 |
+
"num_demo": 1,
|
| 1134 |
+
"num_query": 14
|
| 1135 |
+
},
|
| 1136 |
+
{
|
| 1137 |
+
"name": "planning_screenshot_termes",
|
| 1138 |
+
"score": 0.0,
|
| 1139 |
+
"eval_type": "rule",
|
| 1140 |
+
"num_demo": 1,
|
| 1141 |
+
"num_query": 15
|
| 1142 |
+
},
|
| 1143 |
+
{
|
| 1144 |
+
"name": "topological_sort",
|
| 1145 |
+
"score": 0.0,
|
| 1146 |
+
"eval_type": "rule",
|
| 1147 |
+
"num_demo": 1,
|
| 1148 |
+
"num_query": 14
|
| 1149 |
+
},
|
| 1150 |
+
{
|
| 1151 |
+
"name": "multiple_states_identify_asia",
|
| 1152 |
+
"score": 0.857142857142857,
|
| 1153 |
+
"eval_type": "rule",
|
| 1154 |
+
"num_demo": 1,
|
| 1155 |
+
"num_query": 14
|
| 1156 |
+
},
|
| 1157 |
+
{
|
| 1158 |
+
"name": "multiple_states_identify_africa",
|
| 1159 |
+
"score": 0.7285714285714284,
|
| 1160 |
+
"eval_type": "rule",
|
| 1161 |
+
"num_demo": 1,
|
| 1162 |
+
"num_query": 14
|
| 1163 |
+
},
|
| 1164 |
+
{
|
| 1165 |
+
"name": "multiple_states_identify_europe",
|
| 1166 |
+
"score": 0.7857142857142856,
|
| 1167 |
+
"eval_type": "rule",
|
| 1168 |
+
"num_demo": 1,
|
| 1169 |
+
"num_query": 14
|
| 1170 |
+
},
|
| 1171 |
+
{
|
| 1172 |
+
"name": "multiple_states_identify_americas",
|
| 1173 |
+
"score": 0.6571428571428571,
|
| 1174 |
+
"eval_type": "rule",
|
| 1175 |
+
"num_demo": 1,
|
| 1176 |
+
"num_query": 14
|
| 1177 |
+
},
|
| 1178 |
+
{
|
| 1179 |
+
"name": "adapted_cvbench_distance",
|
| 1180 |
+
"score": 0.7857142857142857,
|
| 1181 |
+
"eval_type": "rule",
|
| 1182 |
+
"num_demo": 1,
|
| 1183 |
+
"num_query": 14
|
| 1184 |
+
},
|
| 1185 |
+
{
|
| 1186 |
+
"name": "adapted_cvbench_count",
|
| 1187 |
+
"score": 0.42857142857142855,
|
| 1188 |
+
"eval_type": "rule",
|
| 1189 |
+
"num_demo": 1,
|
| 1190 |
+
"num_query": 14
|
| 1191 |
+
},
|
| 1192 |
+
{
|
| 1193 |
+
"name": "adapted_cvbench_depth",
|
| 1194 |
+
"score": 0.7857142857142857,
|
| 1195 |
+
"eval_type": "rule",
|
| 1196 |
+
"num_demo": 1,
|
| 1197 |
+
"num_query": 14
|
| 1198 |
+
},
|
| 1199 |
+
{
|
| 1200 |
+
"name": "adapted_cvbench_relation",
|
| 1201 |
+
"score": 0.7142857142857143,
|
| 1202 |
+
"eval_type": "rule",
|
| 1203 |
+
"num_demo": 1,
|
| 1204 |
+
"num_query": 14
|
| 1205 |
+
},
|
| 1206 |
+
{
|
| 1207 |
+
"name": "symbolic_graphics_programs_computer_aided_design",
|
| 1208 |
+
"score": 0.21428571428571427,
|
| 1209 |
+
"eval_type": "rule",
|
| 1210 |
+
"num_demo": 1,
|
| 1211 |
+
"num_query": 14
|
| 1212 |
+
},
|
| 1213 |
+
{
|
| 1214 |
+
"name": "symbolic_graphics_programs_scalable_vector_graphics",
|
| 1215 |
+
"score": 0.05555555555555555,
|
| 1216 |
+
"eval_type": "rule",
|
| 1217 |
+
"num_demo": 1,
|
| 1218 |
+
"num_query": 18
|
| 1219 |
+
},
|
| 1220 |
+
{
|
| 1221 |
+
"name": "poetry_custom_rhyming_scheme",
|
| 1222 |
+
"score": 0.0,
|
| 1223 |
+
"eval_type": "rule",
|
| 1224 |
+
"num_demo": 0,
|
| 1225 |
+
"num_query": 15
|
| 1226 |
+
},
|
| 1227 |
+
{
|
| 1228 |
+
"name": "poetry_petrarchian_sonnet_optional_meter",
|
| 1229 |
+
"score": 0.0,
|
| 1230 |
+
"eval_type": "rule",
|
| 1231 |
+
"num_demo": 0,
|
| 1232 |
+
"num_query": 15
|
| 1233 |
+
},
|
| 1234 |
+
{
|
| 1235 |
+
"name": "poetry_haiku",
|
| 1236 |
+
"score": 0.0,
|
| 1237 |
+
"eval_type": "rule",
|
| 1238 |
+
"num_demo": 0,
|
| 1239 |
+
"num_query": 15
|
| 1240 |
+
},
|
| 1241 |
+
{
|
| 1242 |
+
"name": "poetry_shakespearean_sonnet",
|
| 1243 |
+
"score": 0.0,
|
| 1244 |
+
"eval_type": "rule",
|
| 1245 |
+
"num_demo": 0,
|
| 1246 |
+
"num_query": 15
|
| 1247 |
+
},
|
| 1248 |
+
{
|
| 1249 |
+
"name": "scibench_fundamental_wo_solution",
|
| 1250 |
+
"score": 0.5306122448979592,
|
| 1251 |
+
"eval_type": "rule",
|
| 1252 |
+
"num_demo": 1,
|
| 1253 |
+
"num_query": 49
|
| 1254 |
+
},
|
| 1255 |
+
{
|
| 1256 |
+
"name": "screenshot_lighteval_math",
|
| 1257 |
+
"score": 0.5333333333333333,
|
| 1258 |
+
"eval_type": "rule",
|
| 1259 |
+
"num_demo": 1,
|
| 1260 |
+
"num_query": 15
|
| 1261 |
+
},
|
| 1262 |
+
{
|
| 1263 |
+
"name": "screenshot_theoremqa",
|
| 1264 |
+
"score": 0.9285714285714286,
|
| 1265 |
+
"eval_type": "rule",
|
| 1266 |
+
"num_demo": 1,
|
| 1267 |
+
"num_query": 14
|
| 1268 |
+
},
|
| 1269 |
+
{
|
| 1270 |
+
"name": "geometry_reasoning_nested_squares",
|
| 1271 |
+
"score": 0.42857142857142855,
|
| 1272 |
+
"eval_type": "rule",
|
| 1273 |
+
"num_demo": 1,
|
| 1274 |
+
"num_query": 14
|
| 1275 |
+
},
|
| 1276 |
+
{
|
| 1277 |
+
"name": "font_recognition",
|
| 1278 |
+
"score": 0.14285714285714285,
|
| 1279 |
+
"eval_type": "rule",
|
| 1280 |
+
"num_demo": 1,
|
| 1281 |
+
"num_query": 14
|
| 1282 |
+
},
|
| 1283 |
+
{
|
| 1284 |
+
"name": "geometry_reasoning_count_line_intersections",
|
| 1285 |
+
"score": 0.5,
|
| 1286 |
+
"eval_type": "rule",
|
| 1287 |
+
"num_demo": 1,
|
| 1288 |
+
"num_query": 14
|
| 1289 |
+
},
|
| 1290 |
+
{
|
| 1291 |
+
"name": "number_puzzle_sudoku",
|
| 1292 |
+
"score": 0.0,
|
| 1293 |
+
"eval_type": "rule",
|
| 1294 |
+
"num_demo": 1,
|
| 1295 |
+
"num_query": 15
|
| 1296 |
+
},
|
| 1297 |
+
{
|
| 1298 |
+
"name": "circuit_diagram_understanding",
|
| 1299 |
+
"score": 0.5333333333333333,
|
| 1300 |
+
"eval_type": "rule",
|
| 1301 |
+
"num_demo": 1,
|
| 1302 |
+
"num_query": 15
|
| 1303 |
+
},
|
| 1304 |
+
{
|
| 1305 |
+
"name": "number_puzzle_kakuro_5x5",
|
| 1306 |
+
"score": 0.0,
|
| 1307 |
+
"eval_type": "rule",
|
| 1308 |
+
"num_demo": 1,
|
| 1309 |
+
"num_query": 15
|
| 1310 |
+
},
|
| 1311 |
+
{
|
| 1312 |
+
"name": "go_capture_stone",
|
| 1313 |
+
"score": 0.2,
|
| 1314 |
+
"eval_type": "rule",
|
| 1315 |
+
"num_demo": 1,
|
| 1316 |
+
"num_query": 15
|
| 1317 |
+
},
|
| 1318 |
+
{
|
| 1319 |
+
"name": "clip_stable_diffusion_generate",
|
| 1320 |
+
"score": 0.6428571428571429,
|
| 1321 |
+
"eval_type": "rule",
|
| 1322 |
+
"num_demo": 1,
|
| 1323 |
+
"num_query": 14
|
| 1324 |
+
},
|
| 1325 |
+
{
|
| 1326 |
+
"name": "veracity",
|
| 1327 |
+
"score": 0.8571428571428571,
|
| 1328 |
+
"eval_type": "rule",
|
| 1329 |
+
"num_demo": 1,
|
| 1330 |
+
"num_query": 14
|
| 1331 |
+
},
|
| 1332 |
+
{
|
| 1333 |
+
"name": "counterfactual_arithmetic",
|
| 1334 |
+
"score": 1.0,
|
| 1335 |
+
"eval_type": "rule",
|
| 1336 |
+
"num_demo": 1,
|
| 1337 |
+
"num_query": 14
|
| 1338 |
+
},
|
| 1339 |
+
{
|
| 1340 |
+
"name": "maze_2d_8x8",
|
| 1341 |
+
"score": 0.0,
|
| 1342 |
+
"eval_type": "rule",
|
| 1343 |
+
"num_demo": 1,
|
| 1344 |
+
"num_query": 14
|
| 1345 |
+
},
|
| 1346 |
+
{
|
| 1347 |
+
"name": "shape_composition_shapes",
|
| 1348 |
+
"score": 0.483673469387755,
|
| 1349 |
+
"eval_type": "rule",
|
| 1350 |
+
"num_demo": 1,
|
| 1351 |
+
"num_query": 14
|
| 1352 |
+
},
|
| 1353 |
+
{
|
| 1354 |
+
"name": "shape_composition_colours",
|
| 1355 |
+
"score": 0.4028061224489797,
|
| 1356 |
+
"eval_type": "rule",
|
| 1357 |
+
"num_demo": 1,
|
| 1358 |
+
"num_query": 14
|
| 1359 |
+
},
|
| 1360 |
+
{
|
| 1361 |
+
"name": "autorater_artifact",
|
| 1362 |
+
"score": 0.7142857142857143,
|
| 1363 |
+
"eval_type": "rule",
|
| 1364 |
+
"num_demo": 1,
|
| 1365 |
+
"num_query": 14
|
| 1366 |
+
},
|
| 1367 |
+
{
|
| 1368 |
+
"name": "autorater_artifact_reason",
|
| 1369 |
+
"score": 0.6666666666666666,
|
| 1370 |
+
"eval_type": "rule",
|
| 1371 |
+
"num_demo": 0,
|
| 1372 |
+
"num_query": 15
|
| 1373 |
+
},
|
| 1374 |
+
{
|
| 1375 |
+
"name": "chess_puzzles_crushing",
|
| 1376 |
+
"score": 0.0,
|
| 1377 |
+
"eval_type": "rule",
|
| 1378 |
+
"num_demo": 1,
|
| 1379 |
+
"num_query": 14
|
| 1380 |
+
},
|
| 1381 |
+
{
|
| 1382 |
+
"name": "reward_models_i2t_reward",
|
| 1383 |
+
"score": 0.5,
|
| 1384 |
+
"eval_type": "rule",
|
| 1385 |
+
"num_demo": 1,
|
| 1386 |
+
"num_query": 14
|
| 1387 |
+
},
|
| 1388 |
+
{
|
| 1389 |
+
"name": "memorization_chinese_celebrity",
|
| 1390 |
+
"score": 0.5357142857142857,
|
| 1391 |
+
"eval_type": "rule",
|
| 1392 |
+
"num_demo": 1,
|
| 1393 |
+
"num_query": 14
|
| 1394 |
+
},
|
| 1395 |
+
{
|
| 1396 |
+
"name": "memorization_papers",
|
| 1397 |
+
"score": 0.5666666666666667,
|
| 1398 |
+
"eval_type": "rule",
|
| 1399 |
+
"num_demo": 1,
|
| 1400 |
+
"num_query": 15
|
| 1401 |
+
},
|
| 1402 |
+
{
|
| 1403 |
+
"name": "memorization_famous_treaty",
|
| 1404 |
+
"score": 0.6071428571428571,
|
| 1405 |
+
"eval_type": "rule",
|
| 1406 |
+
"num_demo": 1,
|
| 1407 |
+
"num_query": 14
|
| 1408 |
+
},
|
| 1409 |
+
{
|
| 1410 |
+
"name": "memorization_indian_celebrity",
|
| 1411 |
+
"score": 0.75,
|
| 1412 |
+
"eval_type": "rule",
|
| 1413 |
+
"num_demo": 1,
|
| 1414 |
+
"num_query": 14
|
| 1415 |
+
},
|
| 1416 |
+
{
|
| 1417 |
+
"name": "figureqa",
|
| 1418 |
+
"score": 0.5,
|
| 1419 |
+
"eval_type": "rule",
|
| 1420 |
+
"num_demo": 1,
|
| 1421 |
+
"num_query": 14
|
| 1422 |
+
},
|
| 1423 |
+
{
|
| 1424 |
+
"name": "TV_show_info_parsing",
|
| 1425 |
+
"score": 0.8095238095238094,
|
| 1426 |
+
"eval_type": "rule",
|
| 1427 |
+
"num_demo": 1,
|
| 1428 |
+
"num_query": 14
|
| 1429 |
+
},
|
| 1430 |
+
{
|
| 1431 |
+
"name": "face_keypoint_detection",
|
| 1432 |
+
"score": 0.42415387699613927,
|
| 1433 |
+
"eval_type": "rule",
|
| 1434 |
+
"num_demo": 1,
|
| 1435 |
+
"num_query": 14
|
| 1436 |
+
},
|
| 1437 |
+
{
|
| 1438 |
+
"name": "widerface_face_count_and_event_classification",
|
| 1439 |
+
"score": 0.6428571428571429,
|
| 1440 |
+
"eval_type": "rule",
|
| 1441 |
+
"num_demo": 1,
|
| 1442 |
+
"num_query": 14
|
| 1443 |
+
},
|
| 1444 |
+
{
|
| 1445 |
+
"name": "average_humidity_estimate_plot",
|
| 1446 |
+
"score": 0.844,
|
| 1447 |
+
"eval_type": "rule",
|
| 1448 |
+
"num_demo": 1,
|
| 1449 |
+
"num_query": 15
|
| 1450 |
+
},
|
| 1451 |
+
{
|
| 1452 |
+
"name": "insect_order_classification",
|
| 1453 |
+
"score": 0.26666666666666666,
|
| 1454 |
+
"eval_type": "rule",
|
| 1455 |
+
"num_demo": 1,
|
| 1456 |
+
"num_query": 15
|
| 1457 |
+
},
|
| 1458 |
+
{
|
| 1459 |
+
"name": "weather_info_parsing",
|
| 1460 |
+
"score": 0.896825396825397,
|
| 1461 |
+
"eval_type": "rule",
|
| 1462 |
+
"num_demo": 1,
|
| 1463 |
+
"num_query": 14
|
| 1464 |
+
},
|
| 1465 |
+
{
|
| 1466 |
+
"name": "electricity_plot_future_prediction",
|
| 1467 |
+
"score": 0.827342105263158,
|
| 1468 |
+
"eval_type": "rule",
|
| 1469 |
+
"num_demo": 1,
|
| 1470 |
+
"num_query": 19
|
| 1471 |
+
},
|
| 1472 |
+
{
|
| 1473 |
+
"name": "egocentric_analysis_single_image",
|
| 1474 |
+
"score": 0.5555555555555556,
|
| 1475 |
+
"eval_type": "rule",
|
| 1476 |
+
"num_demo": 1,
|
| 1477 |
+
"num_query": 9
|
| 1478 |
+
},
|
| 1479 |
+
{
|
| 1480 |
+
"name": "waybill_number_sequence_extraction",
|
| 1481 |
+
"score": 0.7857142857142857,
|
| 1482 |
+
"eval_type": "rule",
|
| 1483 |
+
"num_demo": 1,
|
| 1484 |
+
"num_query": 14
|
| 1485 |
+
},
|
| 1486 |
+
{
|
| 1487 |
+
"name": "graph_maxflow",
|
| 1488 |
+
"score": 0.4666666666666667,
|
| 1489 |
+
"eval_type": "rule",
|
| 1490 |
+
"num_demo": 1,
|
| 1491 |
+
"num_query": 15
|
| 1492 |
+
},
|
| 1493 |
+
{
|
| 1494 |
+
"name": "chemistry_exams_v",
|
| 1495 |
+
"score": 0.42857142857142855,
|
| 1496 |
+
"eval_type": "rule",
|
| 1497 |
+
"num_demo": 1,
|
| 1498 |
+
"num_query": 14
|
| 1499 |
+
},
|
| 1500 |
+
{
|
| 1501 |
+
"name": "finance_table_understanding",
|
| 1502 |
+
"score": 0.7142857142857143,
|
| 1503 |
+
"eval_type": "rule",
|
| 1504 |
+
"num_demo": 1,
|
| 1505 |
+
"num_query": 14
|
| 1506 |
+
},
|
| 1507 |
+
{
|
| 1508 |
+
"name": "funsd_document_qa",
|
| 1509 |
+
"score": 0.7857142857142857,
|
| 1510 |
+
"eval_type": "rule",
|
| 1511 |
+
"num_demo": 1,
|
| 1512 |
+
"num_query": 14
|
| 1513 |
+
},
|
| 1514 |
+
{
|
| 1515 |
+
"name": "graph_theory",
|
| 1516 |
+
"score": 0.2857142857142857,
|
| 1517 |
+
"eval_type": "rule",
|
| 1518 |
+
"num_demo": 1,
|
| 1519 |
+
"num_query": 14
|
| 1520 |
+
},
|
| 1521 |
+
{
|
| 1522 |
+
"name": "geometry_length",
|
| 1523 |
+
"score": 0.2857142857142857,
|
| 1524 |
+
"eval_type": "rule",
|
| 1525 |
+
"num_demo": 1,
|
| 1526 |
+
"num_query": 14
|
| 1527 |
+
},
|
| 1528 |
+
{
|
| 1529 |
+
"name": "vibe_eval_open",
|
| 1530 |
+
"score": 0.0,
|
| 1531 |
+
"eval_type": "rule",
|
| 1532 |
+
"num_demo": 1,
|
| 1533 |
+
"num_query": 14
|
| 1534 |
+
},
|
| 1535 |
+
{
|
| 1536 |
+
"name": "algebra",
|
| 1537 |
+
"score": 0.35714285714285715,
|
| 1538 |
+
"eval_type": "rule",
|
| 1539 |
+
"num_demo": 1,
|
| 1540 |
+
"num_query": 14
|
| 1541 |
+
},
|
| 1542 |
+
{
|
| 1543 |
+
"name": "question_solution_solving",
|
| 1544 |
+
"score": 0.2857142857142857,
|
| 1545 |
+
"eval_type": "rule",
|
| 1546 |
+
"num_demo": 1,
|
| 1547 |
+
"num_query": 14
|
| 1548 |
+
},
|
| 1549 |
+
{
|
| 1550 |
+
"name": "chess_puzzle_single_step",
|
| 1551 |
+
"score": 0.06666666666666667,
|
| 1552 |
+
"eval_type": "rule",
|
| 1553 |
+
"num_demo": 1,
|
| 1554 |
+
"num_query": 15
|
| 1555 |
+
},
|
| 1556 |
+
{
|
| 1557 |
+
"name": "multilingual_game_info_parsing",
|
| 1558 |
+
"score": 0.8392857142857143,
|
| 1559 |
+
"eval_type": "rule",
|
| 1560 |
+
"num_demo": 1,
|
| 1561 |
+
"num_query": 14
|
| 1562 |
+
},
|
| 1563 |
+
{
|
| 1564 |
+
"name": "mnist_pattern",
|
| 1565 |
+
"score": 0.9285714285714286,
|
| 1566 |
+
"eval_type": "rule",
|
| 1567 |
+
"num_demo": 1,
|
| 1568 |
+
"num_query": 14
|
| 1569 |
+
},
|
| 1570 |
+
{
|
| 1571 |
+
"name": "dvqa",
|
| 1572 |
+
"score": 0.7894736842105263,
|
| 1573 |
+
"eval_type": "rule",
|
| 1574 |
+
"num_demo": 1,
|
| 1575 |
+
"num_query": 19
|
| 1576 |
+
},
|
| 1577 |
+
{
|
| 1578 |
+
"name": "chess_winner_identification",
|
| 1579 |
+
"score": 0.4666666666666667,
|
| 1580 |
+
"eval_type": "rule",
|
| 1581 |
+
"num_demo": 1,
|
| 1582 |
+
"num_query": 15
|
| 1583 |
+
},
|
| 1584 |
+
{
|
| 1585 |
+
"name": "physics_exams_v",
|
| 1586 |
+
"score": 0.5,
|
| 1587 |
+
"eval_type": "rule",
|
| 1588 |
+
"num_demo": 1,
|
| 1589 |
+
"num_query": 14
|
| 1590 |
+
},
|
| 1591 |
+
{
|
| 1592 |
+
"name": "snli_ve_visual_entailment",
|
| 1593 |
+
"score": 0.8,
|
| 1594 |
+
"eval_type": "rule",
|
| 1595 |
+
"num_demo": 1,
|
| 1596 |
+
"num_query": 15
|
| 1597 |
+
},
|
| 1598 |
+
{
|
| 1599 |
+
"name": "3d_indoor_scene_text_bbox_selection",
|
| 1600 |
+
"score": 0.42857142857142855,
|
| 1601 |
+
"eval_type": "rule",
|
| 1602 |
+
"num_demo": 1,
|
| 1603 |
+
"num_query": 14
|
| 1604 |
+
},
|
| 1605 |
+
{
|
| 1606 |
+
"name": "geometry_descriptive",
|
| 1607 |
+
"score": 0.2857142857142857,
|
| 1608 |
+
"eval_type": "rule",
|
| 1609 |
+
"num_demo": 1,
|
| 1610 |
+
"num_query": 14
|
| 1611 |
+
},
|
| 1612 |
+
{
|
| 1613 |
+
"name": "top_rated_hotel_identification",
|
| 1614 |
+
"score": 0.9285714285714286,
|
| 1615 |
+
"eval_type": "rule",
|
| 1616 |
+
"num_demo": 1,
|
| 1617 |
+
"num_query": 14
|
| 1618 |
+
},
|
| 1619 |
+
{
|
| 1620 |
+
"name": "science_molecule_chemistry",
|
| 1621 |
+
"score": 0.8,
|
| 1622 |
+
"eval_type": "rule",
|
| 1623 |
+
"num_demo": 1,
|
| 1624 |
+
"num_query": 15
|
| 1625 |
+
},
|
| 1626 |
+
{
|
| 1627 |
+
"name": "physical_property_reasoning",
|
| 1628 |
+
"score": 0.9285714285714286,
|
| 1629 |
+
"eval_type": "rule",
|
| 1630 |
+
"num_demo": 1,
|
| 1631 |
+
"num_query": 14
|
| 1632 |
+
},
|
| 1633 |
+
{
|
| 1634 |
+
"name": "game_info_parsing",
|
| 1635 |
+
"score": 0.9155844155844154,
|
| 1636 |
+
"eval_type": "rule",
|
| 1637 |
+
"num_demo": 1,
|
| 1638 |
+
"num_query": 14
|
| 1639 |
+
},
|
| 1640 |
+
{
|
| 1641 |
+
"name": "humor_understand_caption_match",
|
| 1642 |
+
"score": 0.6666666666666666,
|
| 1643 |
+
"eval_type": "rule",
|
| 1644 |
+
"num_demo": 1,
|
| 1645 |
+
"num_query": 15
|
| 1646 |
+
},
|
| 1647 |
+
{
|
| 1648 |
+
"name": "signboard_identification",
|
| 1649 |
+
"score": 0.5714285714285714,
|
| 1650 |
+
"eval_type": "rule",
|
| 1651 |
+
"num_demo": 1,
|
| 1652 |
+
"num_query": 14
|
| 1653 |
+
},
|
| 1654 |
+
{
|
| 1655 |
+
"name": "coco_object_detection_by_query_property",
|
| 1656 |
+
"score": 0.5713086902037786,
|
| 1657 |
+
"eval_type": "rule",
|
| 1658 |
+
"num_demo": 1,
|
| 1659 |
+
"num_query": 14
|
| 1660 |
+
},
|
| 1661 |
+
{
|
| 1662 |
+
"name": "hashtag_recommendation",
|
| 1663 |
+
"score": 0.9107142857142857,
|
| 1664 |
+
"eval_type": "rule",
|
| 1665 |
+
"num_demo": 1,
|
| 1666 |
+
"num_query": 14
|
| 1667 |
+
},
|
| 1668 |
+
{
|
| 1669 |
+
"name": "pmc_vqa_medical_image_qa",
|
| 1670 |
+
"score": 0.8947368421052632,
|
| 1671 |
+
"eval_type": "rule",
|
| 1672 |
+
"num_demo": 1,
|
| 1673 |
+
"num_query": 19
|
| 1674 |
+
},
|
| 1675 |
+
{
|
| 1676 |
+
"name": "newspaper_page_parse_and_count",
|
| 1677 |
+
"score": 0.46666666666666673,
|
| 1678 |
+
"eval_type": "rule",
|
| 1679 |
+
"num_demo": 1,
|
| 1680 |
+
"num_query": 15
|
| 1681 |
+
},
|
| 1682 |
+
{
|
| 1683 |
+
"name": "science_basic_physics",
|
| 1684 |
+
"score": 0.6666666666666666,
|
| 1685 |
+
"eval_type": "rule",
|
| 1686 |
+
"num_demo": 1,
|
| 1687 |
+
"num_query": 15
|
| 1688 |
+
},
|
| 1689 |
+
{
|
| 1690 |
+
"name": "electricity_future_prediction_from_table",
|
| 1691 |
+
"score": 0.7310526315789473,
|
| 1692 |
+
"eval_type": "rule",
|
| 1693 |
+
"num_demo": 1,
|
| 1694 |
+
"num_query": 19
|
| 1695 |
+
},
|
| 1696 |
+
{
|
| 1697 |
+
"name": "license_plate_recognition",
|
| 1698 |
+
"score": 0.7857142857142857,
|
| 1699 |
+
"eval_type": "rule",
|
| 1700 |
+
"num_demo": 1,
|
| 1701 |
+
"num_query": 14
|
| 1702 |
+
},
|
| 1703 |
+
{
|
| 1704 |
+
"name": "annoying_word_search",
|
| 1705 |
+
"score": 0.0,
|
| 1706 |
+
"eval_type": "rule",
|
| 1707 |
+
"num_demo": 1,
|
| 1708 |
+
"num_query": 14
|
| 1709 |
+
},
|
| 1710 |
+
{
|
| 1711 |
+
"name": "music_sheet_format_QA",
|
| 1712 |
+
"score": 0.35714285714285715,
|
| 1713 |
+
"eval_type": "rule",
|
| 1714 |
+
"num_demo": 1,
|
| 1715 |
+
"num_query": 14
|
| 1716 |
+
},
|
| 1717 |
+
{
|
| 1718 |
+
"name": "deciphering_oracle_bone",
|
| 1719 |
+
"score": 0.0,
|
| 1720 |
+
"eval_type": "rule",
|
| 1721 |
+
"num_demo": 1,
|
| 1722 |
+
"num_query": 14
|
| 1723 |
+
},
|
| 1724 |
+
{
|
| 1725 |
+
"name": "map_diagram_qa",
|
| 1726 |
+
"score": 0.35714285714285715,
|
| 1727 |
+
"eval_type": "rule",
|
| 1728 |
+
"num_demo": 1,
|
| 1729 |
+
"num_query": 14
|
| 1730 |
+
},
|
| 1731 |
+
{
|
| 1732 |
+
"name": "interpret_force_perspective_illusion",
|
| 1733 |
+
"score": 0.8666666666666667,
|
| 1734 |
+
"eval_type": "rule",
|
| 1735 |
+
"num_demo": 1,
|
| 1736 |
+
"num_query": 15
|
| 1737 |
+
},
|
| 1738 |
+
{
|
| 1739 |
+
"name": "healthcare_info_judgement",
|
| 1740 |
+
"score": 1.0,
|
| 1741 |
+
"eval_type": "rule",
|
| 1742 |
+
"num_demo": 1,
|
| 1743 |
+
"num_query": 14
|
| 1744 |
+
},
|
| 1745 |
+
{
|
| 1746 |
+
"name": "geometry_plot_position_relationship",
|
| 1747 |
+
"score": 0.7857142857142857,
|
| 1748 |
+
"eval_type": "rule",
|
| 1749 |
+
"num_demo": 1,
|
| 1750 |
+
"num_query": 14
|
| 1751 |
+
},
|
| 1752 |
+
{
|
| 1753 |
+
"name": "image_style_recognition",
|
| 1754 |
+
"score": 1.0,
|
| 1755 |
+
"eval_type": "rule",
|
| 1756 |
+
"num_demo": 1,
|
| 1757 |
+
"num_query": 14
|
| 1758 |
+
},
|
| 1759 |
+
{
|
| 1760 |
+
"name": "llavaguard",
|
| 1761 |
+
"score": 0.5714285714285714,
|
| 1762 |
+
"eval_type": "rule",
|
| 1763 |
+
"num_demo": 1,
|
| 1764 |
+
"num_query": 14
|
| 1765 |
+
},
|
| 1766 |
+
{
|
| 1767 |
+
"name": "math_convexity_value_estimation",
|
| 1768 |
+
"score": 0.6261945500312965,
|
| 1769 |
+
"eval_type": "rule",
|
| 1770 |
+
"num_demo": 1,
|
| 1771 |
+
"num_query": 15
|
| 1772 |
+
},
|
| 1773 |
+
{
|
| 1774 |
+
"name": "3d_indoor_scene_text_bbox_prediction",
|
| 1775 |
+
"score": 0.09554718151735289,
|
| 1776 |
+
"eval_type": "rule",
|
| 1777 |
+
"num_demo": 1,
|
| 1778 |
+
"num_query": 14
|
| 1779 |
+
},
|
| 1780 |
+
{
|
| 1781 |
+
"name": "biology_exams_v",
|
| 1782 |
+
"score": 0.7857142857142857,
|
| 1783 |
+
"eval_type": "rule",
|
| 1784 |
+
"num_demo": 1,
|
| 1785 |
+
"num_query": 14
|
| 1786 |
+
},
|
| 1787 |
+
{
|
| 1788 |
+
"name": "logical_reasoning_fit_pattern",
|
| 1789 |
+
"score": 0.2857142857142857,
|
| 1790 |
+
"eval_type": "rule",
|
| 1791 |
+
"num_demo": 1,
|
| 1792 |
+
"num_query": 14
|
| 1793 |
+
},
|
| 1794 |
+
{
|
| 1795 |
+
"name": "out_of_context",
|
| 1796 |
+
"score": 0.8571428571428571,
|
| 1797 |
+
"eval_type": "rule",
|
| 1798 |
+
"num_demo": 1,
|
| 1799 |
+
"num_query": 14
|
| 1800 |
+
},
|
| 1801 |
+
{
|
| 1802 |
+
"name": "research_website_parsing_blogpost",
|
| 1803 |
+
"score": 0.14285714285714285,
|
| 1804 |
+
"eval_type": "rule",
|
| 1805 |
+
"num_demo": 1,
|
| 1806 |
+
"num_query": 14
|
| 1807 |
+
},
|
| 1808 |
+
{
|
| 1809 |
+
"name": "long_string_number_recognition",
|
| 1810 |
+
"score": 1.0,
|
| 1811 |
+
"eval_type": "rule",
|
| 1812 |
+
"num_demo": 1,
|
| 1813 |
+
"num_query": 14
|
| 1814 |
+
},
|
| 1815 |
+
{
|
| 1816 |
+
"name": "kvqa_knowledge_aware_qa",
|
| 1817 |
+
"score": 0.631578947368421,
|
| 1818 |
+
"eval_type": "rule",
|
| 1819 |
+
"num_demo": 1,
|
| 1820 |
+
"num_query": 19
|
| 1821 |
+
},
|
| 1822 |
+
{
|
| 1823 |
+
"name": "math_breakpoint",
|
| 1824 |
+
"score": 1.0,
|
| 1825 |
+
"eval_type": "rule",
|
| 1826 |
+
"num_demo": 1,
|
| 1827 |
+
"num_query": 15
|
| 1828 |
+
},
|
| 1829 |
+
{
|
| 1830 |
+
"name": "landmark_recognition_and_qa",
|
| 1831 |
+
"score": 0.6222222222222221,
|
| 1832 |
+
"eval_type": "rule",
|
| 1833 |
+
"num_demo": 1,
|
| 1834 |
+
"num_query": 15
|
| 1835 |
+
},
|
| 1836 |
+
{
|
| 1837 |
+
"name": "character_recognition_in_TV_shows",
|
| 1838 |
+
"score": 0.5714285714285714,
|
| 1839 |
+
"eval_type": "rule",
|
| 1840 |
+
"num_demo": 1,
|
| 1841 |
+
"num_query": 14
|
| 1842 |
+
},
|
| 1843 |
+
{
|
| 1844 |
+
"name": "highest_discount_game_price_identification",
|
| 1845 |
+
"score": 1.0,
|
| 1846 |
+
"eval_type": "rule",
|
| 1847 |
+
"num_demo": 1,
|
| 1848 |
+
"num_query": 14
|
| 1849 |
+
},
|
| 1850 |
+
{
|
| 1851 |
+
"name": "text_style",
|
| 1852 |
+
"score": 0.7142857142857143,
|
| 1853 |
+
"eval_type": "rule",
|
| 1854 |
+
"num_demo": 1,
|
| 1855 |
+
"num_query": 14
|
| 1856 |
+
},
|
| 1857 |
+
{
|
| 1858 |
+
"name": "face_swap",
|
| 1859 |
+
"score": 0.5,
|
| 1860 |
+
"eval_type": "rule",
|
| 1861 |
+
"num_demo": 1,
|
| 1862 |
+
"num_query": 14
|
| 1863 |
+
},
|
| 1864 |
+
{
|
| 1865 |
+
"name": "background_change",
|
| 1866 |
+
"score": 0.8571428571428571,
|
| 1867 |
+
"eval_type": "rule",
|
| 1868 |
+
"num_demo": 1,
|
| 1869 |
+
"num_query": 14
|
| 1870 |
+
},
|
| 1871 |
+
{
|
| 1872 |
+
"name": "face_attribute_edit",
|
| 1873 |
+
"score": 0.5714285714285714,
|
| 1874 |
+
"eval_type": "rule",
|
| 1875 |
+
"num_demo": 1,
|
| 1876 |
+
"num_query": 14
|
| 1877 |
+
},
|
| 1878 |
+
{
|
| 1879 |
+
"name": "research_website_parsing_homepage",
|
| 1880 |
+
"score": 0.07142857142857142,
|
| 1881 |
+
"eval_type": "rule",
|
| 1882 |
+
"num_demo": 1,
|
| 1883 |
+
"num_query": 14
|
| 1884 |
+
},
|
| 1885 |
+
{
|
| 1886 |
+
"name": "cultural_vqa",
|
| 1887 |
+
"score": 0.5333333333333333,
|
| 1888 |
+
"eval_type": "rule",
|
| 1889 |
+
"num_demo": 1,
|
| 1890 |
+
"num_query": 15
|
| 1891 |
+
},
|
| 1892 |
+
{
|
| 1893 |
+
"name": "medical_multi_organ_segmentation_rater",
|
| 1894 |
+
"score": 0.2857142857142857,
|
| 1895 |
+
"eval_type": "rule",
|
| 1896 |
+
"num_demo": 1,
|
| 1897 |
+
"num_query": 14
|
| 1898 |
+
},
|
| 1899 |
+
{
|
| 1900 |
+
"name": "research_website_parsing_publication",
|
| 1901 |
+
"score": 0.14285714285714285,
|
| 1902 |
+
"eval_type": "rule",
|
| 1903 |
+
"num_demo": 1,
|
| 1904 |
+
"num_query": 14
|
| 1905 |
+
},
|
| 1906 |
+
{
|
| 1907 |
+
"name": "text_entity_replace",
|
| 1908 |
+
"score": 0.5714285714285714,
|
| 1909 |
+
"eval_type": "rule",
|
| 1910 |
+
"num_demo": 1,
|
| 1911 |
+
"num_query": 14
|
| 1912 |
+
},
|
| 1913 |
+
{
|
| 1914 |
+
"name": "table2latex_complex",
|
| 1915 |
+
"score": 0.7777777777777779,
|
| 1916 |
+
"eval_type": "llm",
|
| 1917 |
+
"num_demo": 1,
|
| 1918 |
+
"num_query": 9
|
| 1919 |
+
},
|
| 1920 |
+
{
|
| 1921 |
+
"name": "traffic_accident_analysis",
|
| 1922 |
+
"score": 0.642857142857143,
|
| 1923 |
+
"eval_type": "llm",
|
| 1924 |
+
"num_demo": 1,
|
| 1925 |
+
"num_query": 14
|
| 1926 |
+
},
|
| 1927 |
+
{
|
| 1928 |
+
"name": "ascii_art_30",
|
| 1929 |
+
"score": 0.21428571428571427,
|
| 1930 |
+
"eval_type": "llm",
|
| 1931 |
+
"num_demo": 1,
|
| 1932 |
+
"num_query": 14
|
| 1933 |
+
},
|
| 1934 |
+
{
|
| 1935 |
+
"name": "bar_chart_interpretation",
|
| 1936 |
+
"score": 0.6310344827586206,
|
| 1937 |
+
"eval_type": "llm",
|
| 1938 |
+
"num_demo": 1,
|
| 1939 |
+
"num_query": 29
|
| 1940 |
+
},
|
| 1941 |
+
{
|
| 1942 |
+
"name": "humor_explanation",
|
| 1943 |
+
"score": 0.8666666666666669,
|
| 1944 |
+
"eval_type": "llm",
|
| 1945 |
+
"num_demo": 1,
|
| 1946 |
+
"num_query": 15
|
| 1947 |
+
},
|
| 1948 |
+
{
|
| 1949 |
+
"name": "art_explanation",
|
| 1950 |
+
"score": 0.7551724137931034,
|
| 1951 |
+
"eval_type": "llm",
|
| 1952 |
+
"num_demo": 1,
|
| 1953 |
+
"num_query": 29
|
| 1954 |
+
},
|
| 1955 |
+
{
|
| 1956 |
+
"name": "image_humor_understanding",
|
| 1957 |
+
"score": 0.8655172413793104,
|
| 1958 |
+
"eval_type": "llm",
|
| 1959 |
+
"num_demo": 1,
|
| 1960 |
+
"num_query": 29
|
| 1961 |
+
},
|
| 1962 |
+
{
|
| 1963 |
+
"name": "science_figure_explanation",
|
| 1964 |
+
"score": 0.8517241379310342,
|
| 1965 |
+
"eval_type": "llm",
|
| 1966 |
+
"num_demo": 1,
|
| 1967 |
+
"num_query": 29
|
| 1968 |
+
},
|
| 1969 |
+
{
|
| 1970 |
+
"name": "figurative_speech_explanation",
|
| 1971 |
+
"score": 0.8620689655172411,
|
| 1972 |
+
"eval_type": "llm",
|
| 1973 |
+
"num_demo": 1,
|
| 1974 |
+
"num_query": 29
|
| 1975 |
+
},
|
| 1976 |
+
{
|
| 1977 |
+
"name": "unusual_images",
|
| 1978 |
+
"score": 0.8999999999999999,
|
| 1979 |
+
"eval_type": "llm",
|
| 1980 |
+
"num_demo": 1,
|
| 1981 |
+
"num_query": 29
|
| 1982 |
+
},
|
| 1983 |
+
{
|
| 1984 |
+
"name": "funny_image_title",
|
| 1985 |
+
"score": 0.5928571428571427,
|
| 1986 |
+
"eval_type": "llm",
|
| 1987 |
+
"num_demo": 1,
|
| 1988 |
+
"num_query": 14
|
| 1989 |
+
},
|
| 1990 |
+
{
|
| 1991 |
+
"name": "ocr_open_ended_qa",
|
| 1992 |
+
"score": 0.8310344827586207,
|
| 1993 |
+
"eval_type": "llm",
|
| 1994 |
+
"num_demo": 1,
|
| 1995 |
+
"num_query": 29
|
| 1996 |
+
},
|
| 1997 |
+
{
|
| 1998 |
+
"name": "GUI_Chat_Hard",
|
| 1999 |
+
"score": 0.3692307692307693,
|
| 2000 |
+
"eval_type": "llm",
|
| 2001 |
+
"num_demo": 1,
|
| 2002 |
+
"num_query": 26
|
| 2003 |
+
},
|
| 2004 |
+
{
|
| 2005 |
+
"name": "tweets_captioning",
|
| 2006 |
+
"score": 0.5714285714285714,
|
| 2007 |
+
"eval_type": "llm",
|
| 2008 |
+
"num_demo": 1,
|
| 2009 |
+
"num_query": 14
|
| 2010 |
+
},
|
| 2011 |
+
{
|
| 2012 |
+
"name": "meme_explain",
|
| 2013 |
+
"score": 0.8285714285714286,
|
| 2014 |
+
"eval_type": "llm",
|
| 2015 |
+
"num_demo": 1,
|
| 2016 |
+
"num_query": 14
|
| 2017 |
+
},
|
| 2018 |
+
{
|
| 2019 |
+
"name": "vibe_eval_phrase",
|
| 2020 |
+
"score": 0.5928571428571427,
|
| 2021 |
+
"eval_type": "llm",
|
| 2022 |
+
"num_demo": 1,
|
| 2023 |
+
"num_query": 14
|
| 2024 |
+
},
|
| 2025 |
+
{
|
| 2026 |
+
"name": "guess_image_generation_prompt",
|
| 2027 |
+
"score": 0.8210526315789476,
|
| 2028 |
+
"eval_type": "llm",
|
| 2029 |
+
"num_demo": 1,
|
| 2030 |
+
"num_query": 19
|
| 2031 |
+
},
|
| 2032 |
+
{
|
| 2033 |
+
"name": "electrocardiogram",
|
| 2034 |
+
"score": 0.33571428571428574,
|
| 2035 |
+
"eval_type": "llm",
|
| 2036 |
+
"num_demo": 1,
|
| 2037 |
+
"num_query": 14
|
| 2038 |
+
},
|
| 2039 |
+
{
|
| 2040 |
+
"name": "defeasible_reasoning",
|
| 2041 |
+
"score": 0.8827586206896553,
|
| 2042 |
+
"eval_type": "llm",
|
| 2043 |
+
"num_demo": 1,
|
| 2044 |
+
"num_query": 29
|
| 2045 |
+
},
|
| 2046 |
+
{
|
| 2047 |
+
"name": "graph_interpretation",
|
| 2048 |
+
"score": 0.9034482758620688,
|
| 2049 |
+
"eval_type": "llm",
|
| 2050 |
+
"num_demo": 1,
|
| 2051 |
+
"num_query": 29
|
| 2052 |
+
},
|
| 2053 |
+
{
|
| 2054 |
+
"name": "visualization_with_code",
|
| 2055 |
+
"score": 0.6642857142857143,
|
| 2056 |
+
"eval_type": "llm",
|
| 2057 |
+
"num_demo": 1,
|
| 2058 |
+
"num_query": 14
|
| 2059 |
+
},
|
| 2060 |
+
{
|
| 2061 |
+
"name": "image_captioning_with_additional_requirements",
|
| 2062 |
+
"score": 0.9357142857142859,
|
| 2063 |
+
"eval_type": "llm",
|
| 2064 |
+
"num_demo": 1,
|
| 2065 |
+
"num_query": 14
|
| 2066 |
+
},
|
| 2067 |
+
{
|
| 2068 |
+
"name": "docci_image_description_long",
|
| 2069 |
+
"score": 0.8357142857142861,
|
| 2070 |
+
"eval_type": "llm",
|
| 2071 |
+
"num_demo": 1,
|
| 2072 |
+
"num_query": 14
|
| 2073 |
+
},
|
| 2074 |
+
{
|
| 2075 |
+
"name": "bridge_strategies_advanced",
|
| 2076 |
+
"score": 0.22857142857142856,
|
| 2077 |
+
"eval_type": "llm",
|
| 2078 |
+
"num_demo": 1,
|
| 2079 |
+
"num_query": 14
|
| 2080 |
+
},
|
| 2081 |
+
{
|
| 2082 |
+
"name": "scibench_w_solution_open_ended",
|
| 2083 |
+
"score": 0.5660000000000001,
|
| 2084 |
+
"eval_type": "llm",
|
| 2085 |
+
"num_demo": 1,
|
| 2086 |
+
"num_query": 25
|
| 2087 |
+
},
|
| 2088 |
+
{
|
| 2089 |
+
"name": "multi_lingual_Ruozhiba_expalnation_Japanese",
|
| 2090 |
+
"score": 0.6642857142857143,
|
| 2091 |
+
"eval_type": "llm",
|
| 2092 |
+
"num_demo": 1,
|
| 2093 |
+
"num_query": 14
|
| 2094 |
+
},
|
| 2095 |
+
{
|
| 2096 |
+
"name": "bridge_strategies_worldclass",
|
| 2097 |
+
"score": 0.1642857142857143,
|
| 2098 |
+
"eval_type": "llm",
|
| 2099 |
+
"num_demo": 1,
|
| 2100 |
+
"num_query": 14
|
| 2101 |
+
},
|
| 2102 |
+
{
|
| 2103 |
+
"name": "iq_test_open_ended",
|
| 2104 |
+
"score": 0.803448275862069,
|
| 2105 |
+
"eval_type": "llm",
|
| 2106 |
+
"num_demo": 1,
|
| 2107 |
+
"num_query": 29
|
| 2108 |
+
},
|
| 2109 |
+
{
|
| 2110 |
+
"name": "multi_lingual_Ruozhiba_expalnation_Spanish",
|
| 2111 |
+
"score": 0.6571428571428573,
|
| 2112 |
+
"eval_type": "llm",
|
| 2113 |
+
"num_demo": 1,
|
| 2114 |
+
"num_query": 14
|
| 2115 |
+
},
|
| 2116 |
+
{
|
| 2117 |
+
"name": "multi_lingual_Ruozhiba_expalnation_French",
|
| 2118 |
+
"score": 0.5428571428571428,
|
| 2119 |
+
"eval_type": "llm",
|
| 2120 |
+
"num_demo": 1,
|
| 2121 |
+
"num_query": 14
|
| 2122 |
+
},
|
| 2123 |
+
{
|
| 2124 |
+
"name": "multi_lingual_Ruozhiba_expalnation_Russian",
|
| 2125 |
+
"score": 0.6214285714285713,
|
| 2126 |
+
"eval_type": "llm",
|
| 2127 |
+
"num_demo": 1,
|
| 2128 |
+
"num_query": 14
|
| 2129 |
+
},
|
| 2130 |
+
{
|
| 2131 |
+
"name": "red_teaming_celebrity",
|
| 2132 |
+
"score": 0.7950000000000002,
|
| 2133 |
+
"eval_type": "llm",
|
| 2134 |
+
"num_demo": 0,
|
| 2135 |
+
"num_query": 20
|
| 2136 |
+
},
|
| 2137 |
+
{
|
| 2138 |
+
"name": "red_teaming_visualmisleading",
|
| 2139 |
+
"score": 0.773684210526316,
|
| 2140 |
+
"eval_type": "llm",
|
| 2141 |
+
"num_demo": 1,
|
| 2142 |
+
"num_query": 19
|
| 2143 |
+
},
|
| 2144 |
+
{
|
| 2145 |
+
"name": "bridge_strategies_expert",
|
| 2146 |
+
"score": 0.34285714285714286,
|
| 2147 |
+
"eval_type": "llm",
|
| 2148 |
+
"num_demo": 1,
|
| 2149 |
+
"num_query": 14
|
| 2150 |
+
},
|
| 2151 |
+
{
|
| 2152 |
+
"name": "multi_lingual_Ruozhiba_expalnation_Arabic",
|
| 2153 |
+
"score": 0.5999999999999999,
|
| 2154 |
+
"eval_type": "llm",
|
| 2155 |
+
"num_demo": 1,
|
| 2156 |
+
"num_query": 14
|
| 2157 |
+
},
|
| 2158 |
+
{
|
| 2159 |
+
"name": "red_teaming_captcha",
|
| 2160 |
+
"score": 0.12105263157894741,
|
| 2161 |
+
"eval_type": "llm",
|
| 2162 |
+
"num_demo": 1,
|
| 2163 |
+
"num_query": 19
|
| 2164 |
+
},
|
| 2165 |
+
{
|
| 2166 |
+
"name": "multi_lingual_Ruozhiba_expalnation_English",
|
| 2167 |
+
"score": 0.692857142857143,
|
| 2168 |
+
"eval_type": "llm",
|
| 2169 |
+
"num_demo": 1,
|
| 2170 |
+
"num_query": 14
|
| 2171 |
+
},
|
| 2172 |
+
{
|
| 2173 |
+
"name": "GUI_Chat_Easy",
|
| 2174 |
+
"score": 0.5192307692307693,
|
| 2175 |
+
"eval_type": "llm",
|
| 2176 |
+
"num_demo": 1,
|
| 2177 |
+
"num_query": 26
|
| 2178 |
+
},
|
| 2179 |
+
{
|
| 2180 |
+
"name": "table_understanding_fetaqa",
|
| 2181 |
+
"score": 0.5357142857142857,
|
| 2182 |
+
"eval_type": "llm",
|
| 2183 |
+
"num_demo": 1,
|
| 2184 |
+
"num_query": 14
|
| 2185 |
+
},
|
| 2186 |
+
{
|
| 2187 |
+
"name": "red_teaming_politics",
|
| 2188 |
+
"score": 0.7550000000000002,
|
| 2189 |
+
"eval_type": "llm",
|
| 2190 |
+
"num_demo": 0,
|
| 2191 |
+
"num_query": 20
|
| 2192 |
+
},
|
| 2193 |
+
{
|
| 2194 |
+
"name": "red_teaming_jailbreak",
|
| 2195 |
+
"score": 0.5399999999999999,
|
| 2196 |
+
"eval_type": "llm",
|
| 2197 |
+
"num_demo": 0,
|
| 2198 |
+
"num_query": 20
|
| 2199 |
+
},
|
| 2200 |
+
{
|
| 2201 |
+
"name": "red_teaming_racial",
|
| 2202 |
+
"score": 0.7800000000000001,
|
| 2203 |
+
"eval_type": "llm",
|
| 2204 |
+
"num_demo": 0,
|
| 2205 |
+
"num_query": 20
|
| 2206 |
+
}
|
| 2207 |
+
]
|
static/eval_results/SI/Gemini-exp-1206/summary_and_keyword_stats.json
ADDED
|
@@ -0,0 +1,213 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_summary": {
|
| 3 |
+
"core": {
|
| 4 |
+
"num_eval_tasks": 273,
|
| 5 |
+
"num_eval_samples": 4108,
|
| 6 |
+
"macro_mean_score": 0.5910051631121387
|
| 7 |
+
},
|
| 8 |
+
"open": {
|
| 9 |
+
"num_eval_tasks": 42,
|
| 10 |
+
"num_eval_samples": 808,
|
| 11 |
+
"macro_mean_score": 0.6587380703875388
|
| 12 |
+
},
|
| 13 |
+
"overall_score": 0.6000362174155255
|
| 14 |
+
},
|
| 15 |
+
"keyword_stats": {
|
| 16 |
+
"skills": {
|
| 17 |
+
"Object Recognition and Classification": {
|
| 18 |
+
"count": 172,
|
| 19 |
+
"num_samples": 2704,
|
| 20 |
+
"tasks": [],
|
| 21 |
+
"average_score": 0.6201167201282601
|
| 22 |
+
},
|
| 23 |
+
"Domain-Specific Knowledge and Skills": {
|
| 24 |
+
"count": 46,
|
| 25 |
+
"num_samples": 896,
|
| 26 |
+
"tasks": [],
|
| 27 |
+
"average_score": 0.6086832235774449
|
| 28 |
+
},
|
| 29 |
+
"Mathematical and Logical Reasoning": {
|
| 30 |
+
"count": 91,
|
| 31 |
+
"num_samples": 1628,
|
| 32 |
+
"tasks": [],
|
| 33 |
+
"average_score": 0.5432816766625482
|
| 34 |
+
},
|
| 35 |
+
"Spatial and Temporal Reasoning": {
|
| 36 |
+
"count": 78,
|
| 37 |
+
"num_samples": 1270,
|
| 38 |
+
"tasks": [],
|
| 39 |
+
"average_score": 0.45216224164053703
|
| 40 |
+
},
|
| 41 |
+
"Text Recognition (OCR)": {
|
| 42 |
+
"count": 101,
|
| 43 |
+
"num_samples": 1680,
|
| 44 |
+
"tasks": [],
|
| 45 |
+
"average_score": 0.6294881872244182
|
| 46 |
+
},
|
| 47 |
+
"Scene and Event Understanding": {
|
| 48 |
+
"count": 60,
|
| 49 |
+
"num_samples": 1004,
|
| 50 |
+
"tasks": [],
|
| 51 |
+
"average_score": 0.7132374515291743
|
| 52 |
+
},
|
| 53 |
+
"Language Understanding and Generation": {
|
| 54 |
+
"count": 102,
|
| 55 |
+
"num_samples": 1707,
|
| 56 |
+
"tasks": [],
|
| 57 |
+
"average_score": 0.6735586026846926
|
| 58 |
+
},
|
| 59 |
+
"Commonsense and Social Reasoning": {
|
| 60 |
+
"count": 38,
|
| 61 |
+
"num_samples": 652,
|
| 62 |
+
"tasks": [],
|
| 63 |
+
"average_score": 0.7029764742096468
|
| 64 |
+
},
|
| 65 |
+
"Planning and Decision Making": {
|
| 66 |
+
"count": 23,
|
| 67 |
+
"num_samples": 355,
|
| 68 |
+
"tasks": [],
|
| 69 |
+
"average_score": 0.21165289602888632
|
| 70 |
+
},
|
| 71 |
+
"Ethical and Safety Reasoning": {
|
| 72 |
+
"count": 10,
|
| 73 |
+
"num_samples": 170,
|
| 74 |
+
"tasks": [],
|
| 75 |
+
"average_score": 0.7103609022556392
|
| 76 |
+
}
|
| 77 |
+
},
|
| 78 |
+
"input_format": {
|
| 79 |
+
"Photographs": {
|
| 80 |
+
"count": 83,
|
| 81 |
+
"num_samples": 1310,
|
| 82 |
+
"tasks": [],
|
| 83 |
+
"average_score": 0.6513225278038967
|
| 84 |
+
},
|
| 85 |
+
"Diagrams and Data Visualizations": {
|
| 86 |
+
"count": 88,
|
| 87 |
+
"num_samples": 1523,
|
| 88 |
+
"tasks": [],
|
| 89 |
+
"average_score": 0.5872084252022427
|
| 90 |
+
},
|
| 91 |
+
"User Interface Screenshots": {
|
| 92 |
+
"count": 67,
|
| 93 |
+
"num_samples": 1117,
|
| 94 |
+
"tasks": [],
|
| 95 |
+
"average_score": 0.5916991181006717
|
| 96 |
+
},
|
| 97 |
+
"Text-Based Images and Documents": {
|
| 98 |
+
"count": 53,
|
| 99 |
+
"num_samples": 847,
|
| 100 |
+
"tasks": [],
|
| 101 |
+
"average_score": 0.5323189527101717
|
| 102 |
+
},
|
| 103 |
+
"Artistic and Creative Content": {
|
| 104 |
+
"count": 22,
|
| 105 |
+
"num_samples": 388,
|
| 106 |
+
"tasks": [],
|
| 107 |
+
"average_score": 0.6795067312521446
|
| 108 |
+
},
|
| 109 |
+
"3D Models and Aerial Imagery": {
|
| 110 |
+
"count": 2,
|
| 111 |
+
"num_samples": 30,
|
| 112 |
+
"tasks": [],
|
| 113 |
+
"average_score": 0.23570188321921526
|
| 114 |
+
}
|
| 115 |
+
},
|
| 116 |
+
"output_format": {
|
| 117 |
+
"structured_output": {
|
| 118 |
+
"count": 72,
|
| 119 |
+
"num_samples": 1120,
|
| 120 |
+
"tasks": [],
|
| 121 |
+
"average_score": 0.5908800551967335
|
| 122 |
+
},
|
| 123 |
+
"numerical_data": {
|
| 124 |
+
"count": 39,
|
| 125 |
+
"num_samples": 694,
|
| 126 |
+
"tasks": [],
|
| 127 |
+
"average_score": 0.5339185914731945
|
| 128 |
+
},
|
| 129 |
+
"multiple_choice": {
|
| 130 |
+
"count": 33,
|
| 131 |
+
"num_samples": 567,
|
| 132 |
+
"tasks": [],
|
| 133 |
+
"average_score": 0.6710437710437711
|
| 134 |
+
},
|
| 135 |
+
"contextual_formatted_text": {
|
| 136 |
+
"count": 63,
|
| 137 |
+
"num_samples": 972,
|
| 138 |
+
"tasks": [],
|
| 139 |
+
"average_score": 0.5546071692625641
|
| 140 |
+
},
|
| 141 |
+
"exact_text": {
|
| 142 |
+
"count": 57,
|
| 143 |
+
"num_samples": 876,
|
| 144 |
+
"tasks": [],
|
| 145 |
+
"average_score": 0.6089924912115107
|
| 146 |
+
},
|
| 147 |
+
"open_ended_output": {
|
| 148 |
+
"count": 51,
|
| 149 |
+
"num_samples": 986,
|
| 150 |
+
"tasks": [],
|
| 151 |
+
"average_score": 0.6636853203378238
|
| 152 |
+
}
|
| 153 |
+
},
|
| 154 |
+
"input_num": {
|
| 155 |
+
"1-image": {
|
| 156 |
+
"count": 315,
|
| 157 |
+
"num_samples": 5215,
|
| 158 |
+
"tasks": [],
|
| 159 |
+
"average_score": 0.600036217415525
|
| 160 |
+
}
|
| 161 |
+
},
|
| 162 |
+
"app": {
|
| 163 |
+
"Knowledge": {
|
| 164 |
+
"count": 77,
|
| 165 |
+
"num_samples": 1291,
|
| 166 |
+
"tasks": [],
|
| 167 |
+
"average_score": 0.6512422690024146
|
| 168 |
+
},
|
| 169 |
+
"Perception": {
|
| 170 |
+
"count": 82,
|
| 171 |
+
"num_samples": 1318,
|
| 172 |
+
"tasks": [],
|
| 173 |
+
"average_score": 0.6756355092787587
|
| 174 |
+
},
|
| 175 |
+
"Mathematics": {
|
| 176 |
+
"count": 30,
|
| 177 |
+
"num_samples": 497,
|
| 178 |
+
"tasks": [],
|
| 179 |
+
"average_score": 0.5337108357437262
|
| 180 |
+
},
|
| 181 |
+
"Information_Extraction": {
|
| 182 |
+
"count": 41,
|
| 183 |
+
"num_samples": 639,
|
| 184 |
+
"tasks": [],
|
| 185 |
+
"average_score": 0.6897380121518055
|
| 186 |
+
},
|
| 187 |
+
"Science": {
|
| 188 |
+
"count": 22,
|
| 189 |
+
"num_samples": 469,
|
| 190 |
+
"tasks": [],
|
| 191 |
+
"average_score": 0.5753973787282057
|
| 192 |
+
},
|
| 193 |
+
"Planning": {
|
| 194 |
+
"count": 44,
|
| 195 |
+
"num_samples": 712,
|
| 196 |
+
"tasks": [],
|
| 197 |
+
"average_score": 0.31789284765829734
|
| 198 |
+
},
|
| 199 |
+
"Coding": {
|
| 200 |
+
"count": 16,
|
| 201 |
+
"num_samples": 244,
|
| 202 |
+
"tasks": [],
|
| 203 |
+
"average_score": 0.6487132352941177
|
| 204 |
+
},
|
| 205 |
+
"Metrics": {
|
| 206 |
+
"count": 3,
|
| 207 |
+
"num_samples": 45,
|
| 208 |
+
"tasks": [],
|
| 209 |
+
"average_score": 0.7158730158730159
|
| 210 |
+
}
|
| 211 |
+
}
|
| 212 |
+
}
|
| 213 |
+
}
|
static/eval_results/SI/Gemini-exp-1206/task_results.json
ADDED
|
@@ -0,0 +1,2207 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"name": "brand_logo_recognition_and_elaboration",
|
| 4 |
+
"score": 0.86,
|
| 5 |
+
"eval_type": "rule",
|
| 6 |
+
"num_demo": 1,
|
| 7 |
+
"num_query": 25
|
| 8 |
+
},
|
| 9 |
+
{
|
| 10 |
+
"name": "exchange_rate_estimate_plot",
|
| 11 |
+
"score": 0.9776357142857142,
|
| 12 |
+
"eval_type": "rule",
|
| 13 |
+
"num_demo": 1,
|
| 14 |
+
"num_query": 14
|
| 15 |
+
},
|
| 16 |
+
{
|
| 17 |
+
"name": "math_parity",
|
| 18 |
+
"score": 0.8666666666666667,
|
| 19 |
+
"eval_type": "rule",
|
| 20 |
+
"num_demo": 1,
|
| 21 |
+
"num_query": 15
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"name": "traffic_future_prediction_from_line_plot",
|
| 25 |
+
"score": 0.8233684210526315,
|
| 26 |
+
"eval_type": "rule",
|
| 27 |
+
"num_demo": 1,
|
| 28 |
+
"num_query": 19
|
| 29 |
+
},
|
| 30 |
+
{
|
| 31 |
+
"name": "graph_chordless_cycle",
|
| 32 |
+
"score": 0.21428571428571427,
|
| 33 |
+
"eval_type": "rule",
|
| 34 |
+
"num_demo": 1,
|
| 35 |
+
"num_query": 14
|
| 36 |
+
},
|
| 37 |
+
{
|
| 38 |
+
"name": "youtube_video_info_parsing",
|
| 39 |
+
"score": 0.7976190476190477,
|
| 40 |
+
"eval_type": "rule",
|
| 41 |
+
"num_demo": 1,
|
| 42 |
+
"num_query": 14
|
| 43 |
+
},
|
| 44 |
+
{
|
| 45 |
+
"name": "super_clevr_scene_understanding",
|
| 46 |
+
"score": 0.5,
|
| 47 |
+
"eval_type": "rule",
|
| 48 |
+
"num_demo": 1,
|
| 49 |
+
"num_query": 14
|
| 50 |
+
},
|
| 51 |
+
{
|
| 52 |
+
"name": "figureqa",
|
| 53 |
+
"score": 0.6428571428571429,
|
| 54 |
+
"eval_type": "rule",
|
| 55 |
+
"num_demo": 1,
|
| 56 |
+
"num_query": 14
|
| 57 |
+
},
|
| 58 |
+
{
|
| 59 |
+
"name": "face_keypoint_detection",
|
| 60 |
+
"score": 0.6994215520827866,
|
| 61 |
+
"eval_type": "rule",
|
| 62 |
+
"num_demo": 1,
|
| 63 |
+
"num_query": 14
|
| 64 |
+
},
|
| 65 |
+
{
|
| 66 |
+
"name": "widerface_face_count_and_event_classification",
|
| 67 |
+
"score": 0.7142857142857143,
|
| 68 |
+
"eval_type": "rule",
|
| 69 |
+
"num_demo": 1,
|
| 70 |
+
"num_query": 14
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"name": "average_humidity_estimate_plot",
|
| 74 |
+
"score": 0.8480000000000002,
|
| 75 |
+
"eval_type": "rule",
|
| 76 |
+
"num_demo": 1,
|
| 77 |
+
"num_query": 15
|
| 78 |
+
},
|
| 79 |
+
{
|
| 80 |
+
"name": "weather_info_parsing",
|
| 81 |
+
"score": 0.9206349206349208,
|
| 82 |
+
"eval_type": "rule",
|
| 83 |
+
"num_demo": 1,
|
| 84 |
+
"num_query": 14
|
| 85 |
+
},
|
| 86 |
+
{
|
| 87 |
+
"name": "egocentric_analysis_single_image",
|
| 88 |
+
"score": 0.5555555555555556,
|
| 89 |
+
"eval_type": "rule",
|
| 90 |
+
"num_demo": 1,
|
| 91 |
+
"num_query": 9
|
| 92 |
+
},
|
| 93 |
+
{
|
| 94 |
+
"name": "waybill_number_sequence_extraction",
|
| 95 |
+
"score": 0.7142857142857143,
|
| 96 |
+
"eval_type": "rule",
|
| 97 |
+
"num_demo": 1,
|
| 98 |
+
"num_query": 14
|
| 99 |
+
},
|
| 100 |
+
{
|
| 101 |
+
"name": "graph_maxflow",
|
| 102 |
+
"score": 0.4666666666666667,
|
| 103 |
+
"eval_type": "rule",
|
| 104 |
+
"num_demo": 1,
|
| 105 |
+
"num_query": 15
|
| 106 |
+
},
|
| 107 |
+
{
|
| 108 |
+
"name": "TV_show_info_parsing",
|
| 109 |
+
"score": 0.896825396825397,
|
| 110 |
+
"eval_type": "rule",
|
| 111 |
+
"num_demo": 1,
|
| 112 |
+
"num_query": 14
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"name": "insect_order_classification",
|
| 116 |
+
"score": 0.4,
|
| 117 |
+
"eval_type": "rule",
|
| 118 |
+
"num_demo": 1,
|
| 119 |
+
"num_query": 15
|
| 120 |
+
},
|
| 121 |
+
{
|
| 122 |
+
"name": "electricity_plot_future_prediction",
|
| 123 |
+
"score": 0.8896842105263157,
|
| 124 |
+
"eval_type": "rule",
|
| 125 |
+
"num_demo": 1,
|
| 126 |
+
"num_query": 19
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"name": "chemistry_exams_v",
|
| 130 |
+
"score": 0.42857142857142855,
|
| 131 |
+
"eval_type": "rule",
|
| 132 |
+
"num_demo": 1,
|
| 133 |
+
"num_query": 14
|
| 134 |
+
},
|
| 135 |
+
{
|
| 136 |
+
"name": "finance_table_understanding",
|
| 137 |
+
"score": 0.7142857142857143,
|
| 138 |
+
"eval_type": "rule",
|
| 139 |
+
"num_demo": 1,
|
| 140 |
+
"num_query": 14
|
| 141 |
+
},
|
| 142 |
+
{
|
| 143 |
+
"name": "funsd_document_qa",
|
| 144 |
+
"score": 0.7857142857142857,
|
| 145 |
+
"eval_type": "rule",
|
| 146 |
+
"num_demo": 1,
|
| 147 |
+
"num_query": 14
|
| 148 |
+
},
|
| 149 |
+
{
|
| 150 |
+
"name": "vibe_eval_open",
|
| 151 |
+
"score": 0.21428571428571427,
|
| 152 |
+
"eval_type": "rule",
|
| 153 |
+
"num_demo": 1,
|
| 154 |
+
"num_query": 14
|
| 155 |
+
},
|
| 156 |
+
{
|
| 157 |
+
"name": "question_solution_solving",
|
| 158 |
+
"score": 0.42857142857142855,
|
| 159 |
+
"eval_type": "rule",
|
| 160 |
+
"num_demo": 1,
|
| 161 |
+
"num_query": 14
|
| 162 |
+
},
|
| 163 |
+
{
|
| 164 |
+
"name": "graph_theory",
|
| 165 |
+
"score": 0.21428571428571427,
|
| 166 |
+
"eval_type": "rule",
|
| 167 |
+
"num_demo": 1,
|
| 168 |
+
"num_query": 14
|
| 169 |
+
},
|
| 170 |
+
{
|
| 171 |
+
"name": "geometry_analytic",
|
| 172 |
+
"score": 0.2857142857142857,
|
| 173 |
+
"eval_type": "rule",
|
| 174 |
+
"num_demo": 1,
|
| 175 |
+
"num_query": 14
|
| 176 |
+
},
|
| 177 |
+
{
|
| 178 |
+
"name": "geometry_length",
|
| 179 |
+
"score": 0.5,
|
| 180 |
+
"eval_type": "rule",
|
| 181 |
+
"num_demo": 1,
|
| 182 |
+
"num_query": 14
|
| 183 |
+
},
|
| 184 |
+
{
|
| 185 |
+
"name": "algebra",
|
| 186 |
+
"score": 0.35714285714285715,
|
| 187 |
+
"eval_type": "rule",
|
| 188 |
+
"num_demo": 1,
|
| 189 |
+
"num_query": 14
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"name": "chess_puzzle_single_step",
|
| 193 |
+
"score": 0.06666666666666667,
|
| 194 |
+
"eval_type": "rule",
|
| 195 |
+
"num_demo": 1,
|
| 196 |
+
"num_query": 15
|
| 197 |
+
},
|
| 198 |
+
{
|
| 199 |
+
"name": "chess_winner_identification",
|
| 200 |
+
"score": 0.4,
|
| 201 |
+
"eval_type": "rule",
|
| 202 |
+
"num_demo": 1,
|
| 203 |
+
"num_query": 15
|
| 204 |
+
},
|
| 205 |
+
{
|
| 206 |
+
"name": "physical_property_reasoning",
|
| 207 |
+
"score": 1.0,
|
| 208 |
+
"eval_type": "rule",
|
| 209 |
+
"num_demo": 1,
|
| 210 |
+
"num_query": 14
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"name": "humor_understand_caption_match",
|
| 214 |
+
"score": 0.9333333333333333,
|
| 215 |
+
"eval_type": "rule",
|
| 216 |
+
"num_demo": 1,
|
| 217 |
+
"num_query": 15
|
| 218 |
+
},
|
| 219 |
+
{
|
| 220 |
+
"name": "coco_object_detection_by_query_property",
|
| 221 |
+
"score": 0.6989737113485902,
|
| 222 |
+
"eval_type": "rule",
|
| 223 |
+
"num_demo": 1,
|
| 224 |
+
"num_query": 14
|
| 225 |
+
},
|
| 226 |
+
{
|
| 227 |
+
"name": "multilingual_game_info_parsing",
|
| 228 |
+
"score": 0.5803571428571429,
|
| 229 |
+
"eval_type": "rule",
|
| 230 |
+
"num_demo": 1,
|
| 231 |
+
"num_query": 14
|
| 232 |
+
},
|
| 233 |
+
{
|
| 234 |
+
"name": "mnist_pattern",
|
| 235 |
+
"score": 0.6428571428571429,
|
| 236 |
+
"eval_type": "rule",
|
| 237 |
+
"num_demo": 1,
|
| 238 |
+
"num_query": 14
|
| 239 |
+
},
|
| 240 |
+
{
|
| 241 |
+
"name": "dvqa",
|
| 242 |
+
"score": 0.8947368421052632,
|
| 243 |
+
"eval_type": "rule",
|
| 244 |
+
"num_demo": 1,
|
| 245 |
+
"num_query": 19
|
| 246 |
+
},
|
| 247 |
+
{
|
| 248 |
+
"name": "physics_exams_v",
|
| 249 |
+
"score": 0.5714285714285714,
|
| 250 |
+
"eval_type": "rule",
|
| 251 |
+
"num_demo": 1,
|
| 252 |
+
"num_query": 14
|
| 253 |
+
},
|
| 254 |
+
{
|
| 255 |
+
"name": "snli_ve_visual_entailment",
|
| 256 |
+
"score": 0.8666666666666667,
|
| 257 |
+
"eval_type": "rule",
|
| 258 |
+
"num_demo": 1,
|
| 259 |
+
"num_query": 15
|
| 260 |
+
},
|
| 261 |
+
{
|
| 262 |
+
"name": "3d_indoor_scene_text_bbox_selection",
|
| 263 |
+
"score": 0.42857142857142855,
|
| 264 |
+
"eval_type": "rule",
|
| 265 |
+
"num_demo": 1,
|
| 266 |
+
"num_query": 14
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"name": "geometry_descriptive",
|
| 270 |
+
"score": 0.21428571428571427,
|
| 271 |
+
"eval_type": "rule",
|
| 272 |
+
"num_demo": 1,
|
| 273 |
+
"num_query": 14
|
| 274 |
+
},
|
| 275 |
+
{
|
| 276 |
+
"name": "top_rated_hotel_identification",
|
| 277 |
+
"score": 0.7142857142857143,
|
| 278 |
+
"eval_type": "rule",
|
| 279 |
+
"num_demo": 1,
|
| 280 |
+
"num_query": 14
|
| 281 |
+
},
|
| 282 |
+
{
|
| 283 |
+
"name": "science_molecule_chemistry",
|
| 284 |
+
"score": 1.0,
|
| 285 |
+
"eval_type": "rule",
|
| 286 |
+
"num_demo": 1,
|
| 287 |
+
"num_query": 15
|
| 288 |
+
},
|
| 289 |
+
{
|
| 290 |
+
"name": "game_info_parsing",
|
| 291 |
+
"score": 0.9285714285714287,
|
| 292 |
+
"eval_type": "rule",
|
| 293 |
+
"num_demo": 1,
|
| 294 |
+
"num_query": 14
|
| 295 |
+
},
|
| 296 |
+
{
|
| 297 |
+
"name": "deciphering_oracle_bone",
|
| 298 |
+
"score": 0.07142857142857142,
|
| 299 |
+
"eval_type": "rule",
|
| 300 |
+
"num_demo": 1,
|
| 301 |
+
"num_query": 14
|
| 302 |
+
},
|
| 303 |
+
{
|
| 304 |
+
"name": "signboard_identification",
|
| 305 |
+
"score": 0.5714285714285714,
|
| 306 |
+
"eval_type": "rule",
|
| 307 |
+
"num_demo": 1,
|
| 308 |
+
"num_query": 14
|
| 309 |
+
},
|
| 310 |
+
{
|
| 311 |
+
"name": "image_style_recognition",
|
| 312 |
+
"score": 1.0,
|
| 313 |
+
"eval_type": "rule",
|
| 314 |
+
"num_demo": 1,
|
| 315 |
+
"num_query": 14
|
| 316 |
+
},
|
| 317 |
+
{
|
| 318 |
+
"name": "math_convexity_value_estimation",
|
| 319 |
+
"score": 0.6381261822294386,
|
| 320 |
+
"eval_type": "rule",
|
| 321 |
+
"num_demo": 1,
|
| 322 |
+
"num_query": 15
|
| 323 |
+
},
|
| 324 |
+
{
|
| 325 |
+
"name": "3d_indoor_scene_text_bbox_prediction",
|
| 326 |
+
"score": 0.04283233786700199,
|
| 327 |
+
"eval_type": "rule",
|
| 328 |
+
"num_demo": 1,
|
| 329 |
+
"num_query": 14
|
| 330 |
+
},
|
| 331 |
+
{
|
| 332 |
+
"name": "movie_info_parsing",
|
| 333 |
+
"score": 0.7053571428571429,
|
| 334 |
+
"eval_type": "rule",
|
| 335 |
+
"num_demo": 1,
|
| 336 |
+
"num_query": 14
|
| 337 |
+
},
|
| 338 |
+
{
|
| 339 |
+
"name": "human_relationship_reasoning",
|
| 340 |
+
"score": 1.0,
|
| 341 |
+
"eval_type": "rule",
|
| 342 |
+
"num_demo": 1,
|
| 343 |
+
"num_query": 14
|
| 344 |
+
},
|
| 345 |
+
{
|
| 346 |
+
"name": "graph_shortest_path_kamada_kawai",
|
| 347 |
+
"score": 0.5,
|
| 348 |
+
"eval_type": "rule",
|
| 349 |
+
"num_demo": 1,
|
| 350 |
+
"num_query": 14
|
| 351 |
+
},
|
| 352 |
+
{
|
| 353 |
+
"name": "coco_person_detection",
|
| 354 |
+
"score": 0.643609425401072,
|
| 355 |
+
"eval_type": "rule",
|
| 356 |
+
"num_demo": 1,
|
| 357 |
+
"num_query": 14
|
| 358 |
+
},
|
| 359 |
+
{
|
| 360 |
+
"name": "chart_vqa",
|
| 361 |
+
"score": 0.7857142857142857,
|
| 362 |
+
"eval_type": "rule",
|
| 363 |
+
"num_demo": 1,
|
| 364 |
+
"num_query": 14
|
| 365 |
+
},
|
| 366 |
+
{
|
| 367 |
+
"name": "nlvr2_two_image_compare_qa",
|
| 368 |
+
"score": 0.7142857142857143,
|
| 369 |
+
"eval_type": "rule",
|
| 370 |
+
"num_demo": 1,
|
| 371 |
+
"num_query": 14
|
| 372 |
+
},
|
| 373 |
+
{
|
| 374 |
+
"name": "math_exams_v",
|
| 375 |
+
"score": 0.5714285714285714,
|
| 376 |
+
"eval_type": "rule",
|
| 377 |
+
"num_demo": 1,
|
| 378 |
+
"num_query": 14
|
| 379 |
+
},
|
| 380 |
+
{
|
| 381 |
+
"name": "newspaper_ocr_in_query_box",
|
| 382 |
+
"score": 0.4,
|
| 383 |
+
"eval_type": "rule",
|
| 384 |
+
"num_demo": 1,
|
| 385 |
+
"num_query": 15
|
| 386 |
+
},
|
| 387 |
+
{
|
| 388 |
+
"name": "mvsa_sentiment_classification",
|
| 389 |
+
"score": 0.7142857142857143,
|
| 390 |
+
"eval_type": "rule",
|
| 391 |
+
"num_demo": 1,
|
| 392 |
+
"num_query": 14
|
| 393 |
+
},
|
| 394 |
+
{
|
| 395 |
+
"name": "egocentric_spatial_reasoning",
|
| 396 |
+
"score": 0.5555555555555556,
|
| 397 |
+
"eval_type": "rule",
|
| 398 |
+
"num_demo": 1,
|
| 399 |
+
"num_query": 9
|
| 400 |
+
},
|
| 401 |
+
{
|
| 402 |
+
"name": "graph_isomorphism",
|
| 403 |
+
"score": 0.6666666666666666,
|
| 404 |
+
"eval_type": "rule",
|
| 405 |
+
"num_demo": 1,
|
| 406 |
+
"num_query": 15
|
| 407 |
+
},
|
| 408 |
+
{
|
| 409 |
+
"name": "code_programming_test_easy",
|
| 410 |
+
"score": 0.5902777777777778,
|
| 411 |
+
"eval_type": "rule",
|
| 412 |
+
"num_demo": 1,
|
| 413 |
+
"num_query": 24
|
| 414 |
+
},
|
| 415 |
+
{
|
| 416 |
+
"name": "biology_exams_v",
|
| 417 |
+
"score": 0.5,
|
| 418 |
+
"eval_type": "rule",
|
| 419 |
+
"num_demo": 1,
|
| 420 |
+
"num_query": 14
|
| 421 |
+
},
|
| 422 |
+
{
|
| 423 |
+
"name": "long_string_number_recognition",
|
| 424 |
+
"score": 1.0,
|
| 425 |
+
"eval_type": "rule",
|
| 426 |
+
"num_demo": 1,
|
| 427 |
+
"num_query": 14
|
| 428 |
+
},
|
| 429 |
+
{
|
| 430 |
+
"name": "kvqa_knowledge_aware_qa",
|
| 431 |
+
"score": 0.47368421052631576,
|
| 432 |
+
"eval_type": "rule",
|
| 433 |
+
"num_demo": 1,
|
| 434 |
+
"num_query": 19
|
| 435 |
+
},
|
| 436 |
+
{
|
| 437 |
+
"name": "math_breakpoint",
|
| 438 |
+
"score": 0.6666666666666666,
|
| 439 |
+
"eval_type": "rule",
|
| 440 |
+
"num_demo": 1,
|
| 441 |
+
"num_query": 15
|
| 442 |
+
},
|
| 443 |
+
{
|
| 444 |
+
"name": "landmark_recognition_and_qa",
|
| 445 |
+
"score": 0.6666666666666666,
|
| 446 |
+
"eval_type": "rule",
|
| 447 |
+
"num_demo": 1,
|
| 448 |
+
"num_query": 15
|
| 449 |
+
},
|
| 450 |
+
{
|
| 451 |
+
"name": "map_diagram_qa",
|
| 452 |
+
"score": 0.35714285714285715,
|
| 453 |
+
"eval_type": "rule",
|
| 454 |
+
"num_demo": 1,
|
| 455 |
+
"num_query": 14
|
| 456 |
+
},
|
| 457 |
+
{
|
| 458 |
+
"name": "pmc_vqa_medical_image_qa",
|
| 459 |
+
"score": 0.7368421052631579,
|
| 460 |
+
"eval_type": "rule",
|
| 461 |
+
"num_demo": 1,
|
| 462 |
+
"num_query": 19
|
| 463 |
+
},
|
| 464 |
+
{
|
| 465 |
+
"name": "newspaper_page_parse_and_count",
|
| 466 |
+
"score": 0.4444444444444444,
|
| 467 |
+
"eval_type": "rule",
|
| 468 |
+
"num_demo": 1,
|
| 469 |
+
"num_query": 15
|
| 470 |
+
},
|
| 471 |
+
{
|
| 472 |
+
"name": "science_basic_physics",
|
| 473 |
+
"score": 0.8666666666666667,
|
| 474 |
+
"eval_type": "rule",
|
| 475 |
+
"num_demo": 1,
|
| 476 |
+
"num_query": 15
|
| 477 |
+
},
|
| 478 |
+
{
|
| 479 |
+
"name": "electricity_future_prediction_from_table",
|
| 480 |
+
"score": 0.7357894736842104,
|
| 481 |
+
"eval_type": "rule",
|
| 482 |
+
"num_demo": 1,
|
| 483 |
+
"num_query": 19
|
| 484 |
+
},
|
| 485 |
+
{
|
| 486 |
+
"name": "license_plate_recognition",
|
| 487 |
+
"score": 0.8571428571428571,
|
| 488 |
+
"eval_type": "rule",
|
| 489 |
+
"num_demo": 1,
|
| 490 |
+
"num_query": 14
|
| 491 |
+
},
|
| 492 |
+
{
|
| 493 |
+
"name": "places365_scene_type_classification",
|
| 494 |
+
"score": 0.9285714285714286,
|
| 495 |
+
"eval_type": "rule",
|
| 496 |
+
"num_demo": 1,
|
| 497 |
+
"num_query": 14
|
| 498 |
+
},
|
| 499 |
+
{
|
| 500 |
+
"name": "music_info_parsing",
|
| 501 |
+
"score": 0.6696428571428571,
|
| 502 |
+
"eval_type": "rule",
|
| 503 |
+
"num_demo": 1,
|
| 504 |
+
"num_query": 14
|
| 505 |
+
},
|
| 506 |
+
{
|
| 507 |
+
"name": "multilingual_movie_info_parsing",
|
| 508 |
+
"score": 0.7448979591836734,
|
| 509 |
+
"eval_type": "rule",
|
| 510 |
+
"num_demo": 1,
|
| 511 |
+
"num_query": 14
|
| 512 |
+
},
|
| 513 |
+
{
|
| 514 |
+
"name": "iconqa_count_and_reasoning",
|
| 515 |
+
"score": 0.42105263157894735,
|
| 516 |
+
"eval_type": "rule",
|
| 517 |
+
"num_demo": 1,
|
| 518 |
+
"num_query": 19
|
| 519 |
+
},
|
| 520 |
+
{
|
| 521 |
+
"name": "graph_connectivity",
|
| 522 |
+
"score": 0.8,
|
| 523 |
+
"eval_type": "rule",
|
| 524 |
+
"num_demo": 1,
|
| 525 |
+
"num_query": 15
|
| 526 |
+
},
|
| 527 |
+
{
|
| 528 |
+
"name": "graph_shortest_path_planar",
|
| 529 |
+
"score": 0.42857142857142855,
|
| 530 |
+
"eval_type": "rule",
|
| 531 |
+
"num_demo": 1,
|
| 532 |
+
"num_query": 14
|
| 533 |
+
},
|
| 534 |
+
{
|
| 535 |
+
"name": "famous_building_recognition",
|
| 536 |
+
"score": 0.875,
|
| 537 |
+
"eval_type": "rule",
|
| 538 |
+
"num_demo": 1,
|
| 539 |
+
"num_query": 16
|
| 540 |
+
},
|
| 541 |
+
{
|
| 542 |
+
"name": "geometry_transformation",
|
| 543 |
+
"score": 0.35714285714285715,
|
| 544 |
+
"eval_type": "rule",
|
| 545 |
+
"num_demo": 1,
|
| 546 |
+
"num_query": 14
|
| 547 |
+
},
|
| 548 |
+
{
|
| 549 |
+
"name": "long_string_letter_recognition",
|
| 550 |
+
"score": 0.2857142857142857,
|
| 551 |
+
"eval_type": "rule",
|
| 552 |
+
"num_demo": 1,
|
| 553 |
+
"num_query": 14
|
| 554 |
+
},
|
| 555 |
+
{
|
| 556 |
+
"name": "handwritten_math_expression_extraction",
|
| 557 |
+
"score": 0.7857142857142857,
|
| 558 |
+
"eval_type": "rule",
|
| 559 |
+
"num_demo": 1,
|
| 560 |
+
"num_query": 14
|
| 561 |
+
},
|
| 562 |
+
{
|
| 563 |
+
"name": "geometry_solid",
|
| 564 |
+
"score": 0.2857142857142857,
|
| 565 |
+
"eval_type": "rule",
|
| 566 |
+
"num_demo": 1,
|
| 567 |
+
"num_query": 14
|
| 568 |
+
},
|
| 569 |
+
{
|
| 570 |
+
"name": "animal_pose_estimation",
|
| 571 |
+
"score": 0.388821771393555,
|
| 572 |
+
"eval_type": "rule",
|
| 573 |
+
"num_demo": 1,
|
| 574 |
+
"num_query": 14
|
| 575 |
+
},
|
| 576 |
+
{
|
| 577 |
+
"name": "single_person_pose_estimation",
|
| 578 |
+
"score": 0.2837376608670773,
|
| 579 |
+
"eval_type": "rule",
|
| 580 |
+
"num_demo": 1,
|
| 581 |
+
"num_query": 14
|
| 582 |
+
},
|
| 583 |
+
{
|
| 584 |
+
"name": "geometry_area",
|
| 585 |
+
"score": 0.5714285714285714,
|
| 586 |
+
"eval_type": "rule",
|
| 587 |
+
"num_demo": 1,
|
| 588 |
+
"num_query": 14
|
| 589 |
+
},
|
| 590 |
+
{
|
| 591 |
+
"name": "hotel_booking_confirmation_parsing",
|
| 592 |
+
"score": 0.6999999999999998,
|
| 593 |
+
"eval_type": "rule",
|
| 594 |
+
"num_demo": 1,
|
| 595 |
+
"num_query": 14
|
| 596 |
+
},
|
| 597 |
+
{
|
| 598 |
+
"name": "ili_ratio_future_prediction",
|
| 599 |
+
"score": 0.1757142857142859,
|
| 600 |
+
"eval_type": "rule",
|
| 601 |
+
"num_demo": 1,
|
| 602 |
+
"num_query": 14
|
| 603 |
+
},
|
| 604 |
+
{
|
| 605 |
+
"name": "electricity_load_estimate_plot",
|
| 606 |
+
"score": 0.673357142857143,
|
| 607 |
+
"eval_type": "rule",
|
| 608 |
+
"num_demo": 1,
|
| 609 |
+
"num_query": 14
|
| 610 |
+
},
|
| 611 |
+
{
|
| 612 |
+
"name": "tqa_textbook_qa",
|
| 613 |
+
"score": 0.8571428571428571,
|
| 614 |
+
"eval_type": "rule",
|
| 615 |
+
"num_demo": 1,
|
| 616 |
+
"num_query": 14
|
| 617 |
+
},
|
| 618 |
+
{
|
| 619 |
+
"name": "stock_info_parsing",
|
| 620 |
+
"score": 0.9663865546218489,
|
| 621 |
+
"eval_type": "rule",
|
| 622 |
+
"num_demo": 1,
|
| 623 |
+
"num_query": 14
|
| 624 |
+
},
|
| 625 |
+
{
|
| 626 |
+
"name": "quizlet_question_solving",
|
| 627 |
+
"score": 0.5714285714285714,
|
| 628 |
+
"eval_type": "rule",
|
| 629 |
+
"num_demo": 1,
|
| 630 |
+
"num_query": 14
|
| 631 |
+
},
|
| 632 |
+
{
|
| 633 |
+
"name": "stock_price_future_prediction",
|
| 634 |
+
"score": 0.8460357142857141,
|
| 635 |
+
"eval_type": "rule",
|
| 636 |
+
"num_demo": 1,
|
| 637 |
+
"num_query": 14
|
| 638 |
+
},
|
| 639 |
+
{
|
| 640 |
+
"name": "Ad_count_detection",
|
| 641 |
+
"score": 0.35714285714285715,
|
| 642 |
+
"eval_type": "rule",
|
| 643 |
+
"num_demo": 1,
|
| 644 |
+
"num_query": 14
|
| 645 |
+
},
|
| 646 |
+
{
|
| 647 |
+
"name": "recover_masked_word_in_figure",
|
| 648 |
+
"score": 0.2857142857142857,
|
| 649 |
+
"eval_type": "rule",
|
| 650 |
+
"num_demo": 1,
|
| 651 |
+
"num_query": 14
|
| 652 |
+
},
|
| 653 |
+
{
|
| 654 |
+
"name": "polygon_interior_angles",
|
| 655 |
+
"score": 0.057928571428571406,
|
| 656 |
+
"eval_type": "rule",
|
| 657 |
+
"num_demo": 1,
|
| 658 |
+
"num_query": 14
|
| 659 |
+
},
|
| 660 |
+
{
|
| 661 |
+
"name": "web_action_grounding",
|
| 662 |
+
"score": 0.7857142857142857,
|
| 663 |
+
"eval_type": "rule",
|
| 664 |
+
"num_demo": 1,
|
| 665 |
+
"num_query": 14
|
| 666 |
+
},
|
| 667 |
+
{
|
| 668 |
+
"name": "latex_complex_formula_convertion",
|
| 669 |
+
"score": 0.5294117647058824,
|
| 670 |
+
"eval_type": "rule",
|
| 671 |
+
"num_demo": 1,
|
| 672 |
+
"num_query": 17
|
| 673 |
+
},
|
| 674 |
+
{
|
| 675 |
+
"name": "transit_map_intersection_points",
|
| 676 |
+
"score": 0.4672619047619047,
|
| 677 |
+
"eval_type": "rule",
|
| 678 |
+
"num_demo": 1,
|
| 679 |
+
"num_query": 14
|
| 680 |
+
},
|
| 681 |
+
{
|
| 682 |
+
"name": "arxiv_vqa",
|
| 683 |
+
"score": 0.7857142857142857,
|
| 684 |
+
"eval_type": "rule",
|
| 685 |
+
"num_demo": 1,
|
| 686 |
+
"num_query": 14
|
| 687 |
+
},
|
| 688 |
+
{
|
| 689 |
+
"name": "medical_image_artifacts_indentification",
|
| 690 |
+
"score": 0.14285714285714285,
|
| 691 |
+
"eval_type": "rule",
|
| 692 |
+
"num_demo": 1,
|
| 693 |
+
"num_query": 14
|
| 694 |
+
},
|
| 695 |
+
{
|
| 696 |
+
"name": "song_title_identification_from_lyrics",
|
| 697 |
+
"score": 0.75,
|
| 698 |
+
"eval_type": "rule",
|
| 699 |
+
"num_demo": 1,
|
| 700 |
+
"num_query": 14
|
| 701 |
+
},
|
| 702 |
+
{
|
| 703 |
+
"name": "actor_recognition_in_Movie",
|
| 704 |
+
"score": 0.9285714285714286,
|
| 705 |
+
"eval_type": "rule",
|
| 706 |
+
"num_demo": 1,
|
| 707 |
+
"num_query": 14
|
| 708 |
+
},
|
| 709 |
+
{
|
| 710 |
+
"name": "bongard_problem",
|
| 711 |
+
"score": 0.21052631578947367,
|
| 712 |
+
"eval_type": "rule",
|
| 713 |
+
"num_demo": 1,
|
| 714 |
+
"num_query": 19
|
| 715 |
+
},
|
| 716 |
+
{
|
| 717 |
+
"name": "ascii_art_understanding",
|
| 718 |
+
"score": 0.7142857142857143,
|
| 719 |
+
"eval_type": "rule",
|
| 720 |
+
"num_demo": 1,
|
| 721 |
+
"num_query": 14
|
| 722 |
+
},
|
| 723 |
+
{
|
| 724 |
+
"name": "calendar_schedule_suggestion",
|
| 725 |
+
"score": 0.6428571428571429,
|
| 726 |
+
"eval_type": "rule",
|
| 727 |
+
"num_demo": 1,
|
| 728 |
+
"num_query": 14
|
| 729 |
+
},
|
| 730 |
+
{
|
| 731 |
+
"name": "geometry_reasoning_overlapped_circle",
|
| 732 |
+
"score": 0.75,
|
| 733 |
+
"eval_type": "rule",
|
| 734 |
+
"num_demo": 1,
|
| 735 |
+
"num_query": 14
|
| 736 |
+
},
|
| 737 |
+
{
|
| 738 |
+
"name": "planning_screenshot_barman",
|
| 739 |
+
"score": 0.0,
|
| 740 |
+
"eval_type": "rule",
|
| 741 |
+
"num_demo": 1,
|
| 742 |
+
"num_query": 15
|
| 743 |
+
},
|
| 744 |
+
{
|
| 745 |
+
"name": "planning_screenshot_floortile",
|
| 746 |
+
"score": 0.0,
|
| 747 |
+
"eval_type": "rule",
|
| 748 |
+
"num_demo": 1,
|
| 749 |
+
"num_query": 15
|
| 750 |
+
},
|
| 751 |
+
{
|
| 752 |
+
"name": "medical_blood_vessels_recognition",
|
| 753 |
+
"score": 0.7142857142857143,
|
| 754 |
+
"eval_type": "rule",
|
| 755 |
+
"num_demo": 1,
|
| 756 |
+
"num_query": 14
|
| 757 |
+
},
|
| 758 |
+
{
|
| 759 |
+
"name": "location_vqa",
|
| 760 |
+
"score": 0.5714285714285714,
|
| 761 |
+
"eval_type": "rule",
|
| 762 |
+
"num_demo": 1,
|
| 763 |
+
"num_query": 14
|
| 764 |
+
},
|
| 765 |
+
{
|
| 766 |
+
"name": "mindmap_elements_parsing",
|
| 767 |
+
"score": 0.42857142857142855,
|
| 768 |
+
"eval_type": "rule",
|
| 769 |
+
"num_demo": 1,
|
| 770 |
+
"num_query": 14
|
| 771 |
+
},
|
| 772 |
+
{
|
| 773 |
+
"name": "mensa_iq_test",
|
| 774 |
+
"score": 0.328921568627451,
|
| 775 |
+
"eval_type": "rule",
|
| 776 |
+
"num_demo": 1,
|
| 777 |
+
"num_query": 17
|
| 778 |
+
},
|
| 779 |
+
{
|
| 780 |
+
"name": "flowchart_code_generation",
|
| 781 |
+
"score": 0.6666666666666666,
|
| 782 |
+
"eval_type": "rule",
|
| 783 |
+
"num_demo": 1,
|
| 784 |
+
"num_query": 9
|
| 785 |
+
},
|
| 786 |
+
{
|
| 787 |
+
"name": "stackoverflow_debug_QA",
|
| 788 |
+
"score": 0.642857142857143,
|
| 789 |
+
"eval_type": "rule",
|
| 790 |
+
"num_demo": 1,
|
| 791 |
+
"num_query": 14
|
| 792 |
+
},
|
| 793 |
+
{
|
| 794 |
+
"name": "logical_reasoning_find_odd_one_out",
|
| 795 |
+
"score": 0.14285714285714285,
|
| 796 |
+
"eval_type": "rule",
|
| 797 |
+
"num_demo": 1,
|
| 798 |
+
"num_query": 14
|
| 799 |
+
},
|
| 800 |
+
{
|
| 801 |
+
"name": "web_action_prediction",
|
| 802 |
+
"score": 0.7857142857142857,
|
| 803 |
+
"eval_type": "rule",
|
| 804 |
+
"num_demo": 1,
|
| 805 |
+
"num_query": 14
|
| 806 |
+
},
|
| 807 |
+
{
|
| 808 |
+
"name": "code_execution",
|
| 809 |
+
"score": 0.9375,
|
| 810 |
+
"eval_type": "rule",
|
| 811 |
+
"num_demo": 1,
|
| 812 |
+
"num_query": 16
|
| 813 |
+
},
|
| 814 |
+
{
|
| 815 |
+
"name": "music_sheet_format_QA",
|
| 816 |
+
"score": 0.5714285714285714,
|
| 817 |
+
"eval_type": "rule",
|
| 818 |
+
"num_demo": 1,
|
| 819 |
+
"num_query": 14
|
| 820 |
+
},
|
| 821 |
+
{
|
| 822 |
+
"name": "annoying_word_search",
|
| 823 |
+
"score": 0.0,
|
| 824 |
+
"eval_type": "rule",
|
| 825 |
+
"num_demo": 1,
|
| 826 |
+
"num_query": 14
|
| 827 |
+
},
|
| 828 |
+
{
|
| 829 |
+
"name": "interpret_force_perspective_illusion",
|
| 830 |
+
"score": 1.0,
|
| 831 |
+
"eval_type": "rule",
|
| 832 |
+
"num_demo": 1,
|
| 833 |
+
"num_query": 15
|
| 834 |
+
},
|
| 835 |
+
{
|
| 836 |
+
"name": "healthcare_info_judgement",
|
| 837 |
+
"score": 0.7142857142857143,
|
| 838 |
+
"eval_type": "rule",
|
| 839 |
+
"num_demo": 1,
|
| 840 |
+
"num_query": 14
|
| 841 |
+
},
|
| 842 |
+
{
|
| 843 |
+
"name": "geometry_plot_position_relationship",
|
| 844 |
+
"score": 0.7142857142857143,
|
| 845 |
+
"eval_type": "rule",
|
| 846 |
+
"num_demo": 1,
|
| 847 |
+
"num_query": 14
|
| 848 |
+
},
|
| 849 |
+
{
|
| 850 |
+
"name": "relative_depth_of_different_points",
|
| 851 |
+
"score": 0.7142857142857143,
|
| 852 |
+
"eval_type": "rule",
|
| 853 |
+
"num_demo": 1,
|
| 854 |
+
"num_query": 14
|
| 855 |
+
},
|
| 856 |
+
{
|
| 857 |
+
"name": "topological_sort",
|
| 858 |
+
"score": 0.2857142857142857,
|
| 859 |
+
"eval_type": "rule",
|
| 860 |
+
"num_demo": 1,
|
| 861 |
+
"num_query": 14
|
| 862 |
+
},
|
| 863 |
+
{
|
| 864 |
+
"name": "scibench_fundamental_wo_solution",
|
| 865 |
+
"score": 0.4489795918367347,
|
| 866 |
+
"eval_type": "rule",
|
| 867 |
+
"num_demo": 1,
|
| 868 |
+
"num_query": 49
|
| 869 |
+
},
|
| 870 |
+
{
|
| 871 |
+
"name": "geometry_reasoning_nested_squares",
|
| 872 |
+
"score": 0.5714285714285714,
|
| 873 |
+
"eval_type": "rule",
|
| 874 |
+
"num_demo": 1,
|
| 875 |
+
"num_query": 14
|
| 876 |
+
},
|
| 877 |
+
{
|
| 878 |
+
"name": "font_recognition",
|
| 879 |
+
"score": 0.21428571428571427,
|
| 880 |
+
"eval_type": "rule",
|
| 881 |
+
"num_demo": 1,
|
| 882 |
+
"num_query": 14
|
| 883 |
+
},
|
| 884 |
+
{
|
| 885 |
+
"name": "geometry_reasoning_count_line_intersections",
|
| 886 |
+
"score": 0.5714285714285714,
|
| 887 |
+
"eval_type": "rule",
|
| 888 |
+
"num_demo": 1,
|
| 889 |
+
"num_query": 14
|
| 890 |
+
},
|
| 891 |
+
{
|
| 892 |
+
"name": "circuit_diagram_understanding",
|
| 893 |
+
"score": 0.4666666666666667,
|
| 894 |
+
"eval_type": "rule",
|
| 895 |
+
"num_demo": 1,
|
| 896 |
+
"num_query": 15
|
| 897 |
+
},
|
| 898 |
+
{
|
| 899 |
+
"name": "go_capture_stone",
|
| 900 |
+
"score": 0.26666666666666666,
|
| 901 |
+
"eval_type": "rule",
|
| 902 |
+
"num_demo": 1,
|
| 903 |
+
"num_query": 15
|
| 904 |
+
},
|
| 905 |
+
{
|
| 906 |
+
"name": "monthly_weather_days_count",
|
| 907 |
+
"score": 0.42857142857142855,
|
| 908 |
+
"eval_type": "rule",
|
| 909 |
+
"num_demo": 1,
|
| 910 |
+
"num_query": 14
|
| 911 |
+
},
|
| 912 |
+
{
|
| 913 |
+
"name": "weather_map_climate_type_temperature_parsing",
|
| 914 |
+
"score": 0.7142857142857143,
|
| 915 |
+
"eval_type": "rule",
|
| 916 |
+
"num_demo": 1,
|
| 917 |
+
"num_query": 14
|
| 918 |
+
},
|
| 919 |
+
{
|
| 920 |
+
"name": "top_video_creator_identification",
|
| 921 |
+
"score": 0.6428571428571429,
|
| 922 |
+
"eval_type": "rule",
|
| 923 |
+
"num_demo": 1,
|
| 924 |
+
"num_query": 14
|
| 925 |
+
},
|
| 926 |
+
{
|
| 927 |
+
"name": "rebus",
|
| 928 |
+
"score": 0.4782608695652174,
|
| 929 |
+
"eval_type": "rule",
|
| 930 |
+
"num_demo": 1,
|
| 931 |
+
"num_query": 23
|
| 932 |
+
},
|
| 933 |
+
{
|
| 934 |
+
"name": "ishihara_test",
|
| 935 |
+
"score": 0.35714285714285715,
|
| 936 |
+
"eval_type": "rule",
|
| 937 |
+
"num_demo": 1,
|
| 938 |
+
"num_query": 14
|
| 939 |
+
},
|
| 940 |
+
{
|
| 941 |
+
"name": "paper_vqa",
|
| 942 |
+
"score": 0.5,
|
| 943 |
+
"eval_type": "rule",
|
| 944 |
+
"num_demo": 1,
|
| 945 |
+
"num_query": 14
|
| 946 |
+
},
|
| 947 |
+
{
|
| 948 |
+
"name": "signage_navigation",
|
| 949 |
+
"score": 0.7142857142857143,
|
| 950 |
+
"eval_type": "rule",
|
| 951 |
+
"num_demo": 1,
|
| 952 |
+
"num_query": 14
|
| 953 |
+
},
|
| 954 |
+
{
|
| 955 |
+
"name": "planning_visual_storage",
|
| 956 |
+
"score": 0.06666666666666667,
|
| 957 |
+
"eval_type": "rule",
|
| 958 |
+
"num_demo": 1,
|
| 959 |
+
"num_query": 15
|
| 960 |
+
},
|
| 961 |
+
{
|
| 962 |
+
"name": "GUI_Act_Web_Multi",
|
| 963 |
+
"score": 0.35714285714285715,
|
| 964 |
+
"eval_type": "rule",
|
| 965 |
+
"num_demo": 1,
|
| 966 |
+
"num_query": 14
|
| 967 |
+
},
|
| 968 |
+
{
|
| 969 |
+
"name": "webpage_code_understanding",
|
| 970 |
+
"score": 0.7777777777777778,
|
| 971 |
+
"eval_type": "rule",
|
| 972 |
+
"num_demo": 1,
|
| 973 |
+
"num_query": 9
|
| 974 |
+
},
|
| 975 |
+
{
|
| 976 |
+
"name": "chinese_idiom_recognition",
|
| 977 |
+
"score": 1.0,
|
| 978 |
+
"eval_type": "rule",
|
| 979 |
+
"num_demo": 1,
|
| 980 |
+
"num_query": 14
|
| 981 |
+
},
|
| 982 |
+
{
|
| 983 |
+
"name": "number_comparison",
|
| 984 |
+
"score": 0.9285714285714286,
|
| 985 |
+
"eval_type": "rule",
|
| 986 |
+
"num_demo": 1,
|
| 987 |
+
"num_query": 14
|
| 988 |
+
},
|
| 989 |
+
{
|
| 990 |
+
"name": "medical_counting_lymphocytes",
|
| 991 |
+
"score": 0.0,
|
| 992 |
+
"eval_type": "rule",
|
| 993 |
+
"num_demo": 1,
|
| 994 |
+
"num_query": 14
|
| 995 |
+
},
|
| 996 |
+
{
|
| 997 |
+
"name": "planning_screenshot_blocksworld",
|
| 998 |
+
"score": 0.13333333333333333,
|
| 999 |
+
"eval_type": "rule",
|
| 1000 |
+
"num_demo": 1,
|
| 1001 |
+
"num_query": 15
|
| 1002 |
+
},
|
| 1003 |
+
{
|
| 1004 |
+
"name": "product_ocr_qa",
|
| 1005 |
+
"score": 0.5,
|
| 1006 |
+
"eval_type": "rule",
|
| 1007 |
+
"num_demo": 1,
|
| 1008 |
+
"num_query": 14
|
| 1009 |
+
},
|
| 1010 |
+
{
|
| 1011 |
+
"name": "geometry_reasoning_circled_letter",
|
| 1012 |
+
"score": 0.8928571428571429,
|
| 1013 |
+
"eval_type": "rule",
|
| 1014 |
+
"num_demo": 1,
|
| 1015 |
+
"num_query": 14
|
| 1016 |
+
},
|
| 1017 |
+
{
|
| 1018 |
+
"name": "GUI_Act_Web_Single",
|
| 1019 |
+
"score": 0.0,
|
| 1020 |
+
"eval_type": "rule",
|
| 1021 |
+
"num_demo": 1,
|
| 1022 |
+
"num_query": 14
|
| 1023 |
+
},
|
| 1024 |
+
{
|
| 1025 |
+
"name": "extract_webpage_headline",
|
| 1026 |
+
"score": 0.6428571428571429,
|
| 1027 |
+
"eval_type": "rule",
|
| 1028 |
+
"num_demo": 1,
|
| 1029 |
+
"num_query": 14
|
| 1030 |
+
},
|
| 1031 |
+
{
|
| 1032 |
+
"name": "planning_screenshot_storage",
|
| 1033 |
+
"score": 0.2,
|
| 1034 |
+
"eval_type": "rule",
|
| 1035 |
+
"num_demo": 1,
|
| 1036 |
+
"num_query": 15
|
| 1037 |
+
},
|
| 1038 |
+
{
|
| 1039 |
+
"name": "game_platform_support_identification",
|
| 1040 |
+
"score": 0.9642857142857143,
|
| 1041 |
+
"eval_type": "rule",
|
| 1042 |
+
"num_demo": 1,
|
| 1043 |
+
"num_query": 14
|
| 1044 |
+
},
|
| 1045 |
+
{
|
| 1046 |
+
"name": "GUI_Act_Mobile_swipe",
|
| 1047 |
+
"score": 0.6410384962857598,
|
| 1048 |
+
"eval_type": "rule",
|
| 1049 |
+
"num_demo": 1,
|
| 1050 |
+
"num_query": 13
|
| 1051 |
+
},
|
| 1052 |
+
{
|
| 1053 |
+
"name": "mahjong",
|
| 1054 |
+
"score": 0.14285714285714285,
|
| 1055 |
+
"eval_type": "rule",
|
| 1056 |
+
"num_demo": 1,
|
| 1057 |
+
"num_query": 14
|
| 1058 |
+
},
|
| 1059 |
+
{
|
| 1060 |
+
"name": "soccer_offside",
|
| 1061 |
+
"score": 0.2222222222222222,
|
| 1062 |
+
"eval_type": "rule",
|
| 1063 |
+
"num_demo": 1,
|
| 1064 |
+
"num_query": 9
|
| 1065 |
+
},
|
| 1066 |
+
{
|
| 1067 |
+
"name": "scibench_calculus_wo_solution",
|
| 1068 |
+
"score": 0.4489795918367347,
|
| 1069 |
+
"eval_type": "rule",
|
| 1070 |
+
"num_demo": 1,
|
| 1071 |
+
"num_query": 49
|
| 1072 |
+
},
|
| 1073 |
+
{
|
| 1074 |
+
"name": "knowledge_graph_understanding",
|
| 1075 |
+
"score": 0.7333333333333333,
|
| 1076 |
+
"eval_type": "rule",
|
| 1077 |
+
"num_demo": 1,
|
| 1078 |
+
"num_query": 15
|
| 1079 |
+
},
|
| 1080 |
+
{
|
| 1081 |
+
"name": "geometry_reasoning_grid",
|
| 1082 |
+
"score": 0.9285714285714286,
|
| 1083 |
+
"eval_type": "rule",
|
| 1084 |
+
"num_demo": 1,
|
| 1085 |
+
"num_query": 14
|
| 1086 |
+
},
|
| 1087 |
+
{
|
| 1088 |
+
"name": "relative_reflectance_of_different_regions",
|
| 1089 |
+
"score": 0.35714285714285715,
|
| 1090 |
+
"eval_type": "rule",
|
| 1091 |
+
"num_demo": 1,
|
| 1092 |
+
"num_query": 14
|
| 1093 |
+
},
|
| 1094 |
+
{
|
| 1095 |
+
"name": "entertainment_web_game_style",
|
| 1096 |
+
"score": 0.7857142857142857,
|
| 1097 |
+
"eval_type": "rule",
|
| 1098 |
+
"num_demo": 1,
|
| 1099 |
+
"num_query": 14
|
| 1100 |
+
},
|
| 1101 |
+
{
|
| 1102 |
+
"name": "image_translation_en2cn",
|
| 1103 |
+
"score": 0.46216435287356755,
|
| 1104 |
+
"eval_type": "rule",
|
| 1105 |
+
"num_demo": 1,
|
| 1106 |
+
"num_query": 9
|
| 1107 |
+
},
|
| 1108 |
+
{
|
| 1109 |
+
"name": "realworld_qa_en2cn",
|
| 1110 |
+
"score": 0.8571428571428571,
|
| 1111 |
+
"eval_type": "rule",
|
| 1112 |
+
"num_demo": 1,
|
| 1113 |
+
"num_query": 14
|
| 1114 |
+
},
|
| 1115 |
+
{
|
| 1116 |
+
"name": "planning_screenshot_grippers",
|
| 1117 |
+
"score": 0.6,
|
| 1118 |
+
"eval_type": "rule",
|
| 1119 |
+
"num_demo": 1,
|
| 1120 |
+
"num_query": 15
|
| 1121 |
+
},
|
| 1122 |
+
{
|
| 1123 |
+
"name": "orchestra_score_recognition",
|
| 1124 |
+
"score": 0.21428571428571427,
|
| 1125 |
+
"eval_type": "rule",
|
| 1126 |
+
"num_demo": 1,
|
| 1127 |
+
"num_query": 14
|
| 1128 |
+
},
|
| 1129 |
+
{
|
| 1130 |
+
"name": "icon_arithmetic_puzzle",
|
| 1131 |
+
"score": 0.75,
|
| 1132 |
+
"eval_type": "rule",
|
| 1133 |
+
"num_demo": 1,
|
| 1134 |
+
"num_query": 14
|
| 1135 |
+
},
|
| 1136 |
+
{
|
| 1137 |
+
"name": "autonomous_driving_scene_analysis",
|
| 1138 |
+
"score": 0.9285714285714286,
|
| 1139 |
+
"eval_type": "rule",
|
| 1140 |
+
"num_demo": 1,
|
| 1141 |
+
"num_query": 14
|
| 1142 |
+
},
|
| 1143 |
+
{
|
| 1144 |
+
"name": "counting_single_image",
|
| 1145 |
+
"score": 0.6428571428571429,
|
| 1146 |
+
"eval_type": "rule",
|
| 1147 |
+
"num_demo": 1,
|
| 1148 |
+
"num_query": 14
|
| 1149 |
+
},
|
| 1150 |
+
{
|
| 1151 |
+
"name": "MMMU_pro_exam_screenshot",
|
| 1152 |
+
"score": 0.4444444444444444,
|
| 1153 |
+
"eval_type": "rule",
|
| 1154 |
+
"num_demo": 1,
|
| 1155 |
+
"num_query": 99
|
| 1156 |
+
},
|
| 1157 |
+
{
|
| 1158 |
+
"name": "GUI_Act_Mobile_tap",
|
| 1159 |
+
"score": 0.07142857142857142,
|
| 1160 |
+
"eval_type": "rule",
|
| 1161 |
+
"num_demo": 1,
|
| 1162 |
+
"num_query": 14
|
| 1163 |
+
},
|
| 1164 |
+
{
|
| 1165 |
+
"name": "road_map_find_highway_between_two_place",
|
| 1166 |
+
"score": 0.7058823529411765,
|
| 1167 |
+
"eval_type": "rule",
|
| 1168 |
+
"num_demo": 1,
|
| 1169 |
+
"num_query": 17
|
| 1170 |
+
},
|
| 1171 |
+
{
|
| 1172 |
+
"name": "clevrer_physics",
|
| 1173 |
+
"score": 0.35,
|
| 1174 |
+
"eval_type": "rule",
|
| 1175 |
+
"num_demo": 1,
|
| 1176 |
+
"num_query": 20
|
| 1177 |
+
},
|
| 1178 |
+
{
|
| 1179 |
+
"name": "MMMU_physics_chemistry_selected",
|
| 1180 |
+
"score": 0.7142857142857143,
|
| 1181 |
+
"eval_type": "rule",
|
| 1182 |
+
"num_demo": 1,
|
| 1183 |
+
"num_query": 14
|
| 1184 |
+
},
|
| 1185 |
+
{
|
| 1186 |
+
"name": "chess_sygyzy_endgames",
|
| 1187 |
+
"score": 0.11354838709677419,
|
| 1188 |
+
"eval_type": "rule",
|
| 1189 |
+
"num_demo": 1,
|
| 1190 |
+
"num_query": 14
|
| 1191 |
+
},
|
| 1192 |
+
{
|
| 1193 |
+
"name": "planning_screenshot_tyreworld",
|
| 1194 |
+
"score": 0.8666666666666667,
|
| 1195 |
+
"eval_type": "rule",
|
| 1196 |
+
"num_demo": 1,
|
| 1197 |
+
"num_query": 15
|
| 1198 |
+
},
|
| 1199 |
+
{
|
| 1200 |
+
"name": "music_sheet_note_count",
|
| 1201 |
+
"score": 0.11764705882352941,
|
| 1202 |
+
"eval_type": "rule",
|
| 1203 |
+
"num_demo": 1,
|
| 1204 |
+
"num_query": 17
|
| 1205 |
+
},
|
| 1206 |
+
{
|
| 1207 |
+
"name": "planning_screenshot_termes",
|
| 1208 |
+
"score": 0.0,
|
| 1209 |
+
"eval_type": "rule",
|
| 1210 |
+
"num_demo": 1,
|
| 1211 |
+
"num_query": 15
|
| 1212 |
+
},
|
| 1213 |
+
{
|
| 1214 |
+
"name": "hashtag_recommendation",
|
| 1215 |
+
"score": 0.869047619047619,
|
| 1216 |
+
"eval_type": "rule",
|
| 1217 |
+
"num_demo": 1,
|
| 1218 |
+
"num_query": 14
|
| 1219 |
+
},
|
| 1220 |
+
{
|
| 1221 |
+
"name": "llavaguard",
|
| 1222 |
+
"score": 0.5357142857142857,
|
| 1223 |
+
"eval_type": "rule",
|
| 1224 |
+
"num_demo": 1,
|
| 1225 |
+
"num_query": 14
|
| 1226 |
+
},
|
| 1227 |
+
{
|
| 1228 |
+
"name": "medical_multi_organ_segmentation_rater",
|
| 1229 |
+
"score": 0.42857142857142855,
|
| 1230 |
+
"eval_type": "rule",
|
| 1231 |
+
"num_demo": 1,
|
| 1232 |
+
"num_query": 14
|
| 1233 |
+
},
|
| 1234 |
+
{
|
| 1235 |
+
"name": "cultural_vqa",
|
| 1236 |
+
"score": 0.4666666666666667,
|
| 1237 |
+
"eval_type": "rule",
|
| 1238 |
+
"num_demo": 1,
|
| 1239 |
+
"num_query": 15
|
| 1240 |
+
},
|
| 1241 |
+
{
|
| 1242 |
+
"name": "logical_reasoning_fit_pattern",
|
| 1243 |
+
"score": 0.2857142857142857,
|
| 1244 |
+
"eval_type": "rule",
|
| 1245 |
+
"num_demo": 1,
|
| 1246 |
+
"num_query": 14
|
| 1247 |
+
},
|
| 1248 |
+
{
|
| 1249 |
+
"name": "multiple_states_identify_asia",
|
| 1250 |
+
"score": 0.9142857142857143,
|
| 1251 |
+
"eval_type": "rule",
|
| 1252 |
+
"num_demo": 1,
|
| 1253 |
+
"num_query": 14
|
| 1254 |
+
},
|
| 1255 |
+
{
|
| 1256 |
+
"name": "character_recognition_in_TV_shows",
|
| 1257 |
+
"score": 0.5714285714285714,
|
| 1258 |
+
"eval_type": "rule",
|
| 1259 |
+
"num_demo": 1,
|
| 1260 |
+
"num_query": 14
|
| 1261 |
+
},
|
| 1262 |
+
{
|
| 1263 |
+
"name": "multiple_states_identify_africa",
|
| 1264 |
+
"score": 0.8,
|
| 1265 |
+
"eval_type": "rule",
|
| 1266 |
+
"num_demo": 1,
|
| 1267 |
+
"num_query": 14
|
| 1268 |
+
},
|
| 1269 |
+
{
|
| 1270 |
+
"name": "highest_discount_game_price_identification",
|
| 1271 |
+
"score": 0.8571428571428571,
|
| 1272 |
+
"eval_type": "rule",
|
| 1273 |
+
"num_demo": 1,
|
| 1274 |
+
"num_query": 14
|
| 1275 |
+
},
|
| 1276 |
+
{
|
| 1277 |
+
"name": "multiple_states_identify_europe",
|
| 1278 |
+
"score": 0.7714285714285714,
|
| 1279 |
+
"eval_type": "rule",
|
| 1280 |
+
"num_demo": 1,
|
| 1281 |
+
"num_query": 14
|
| 1282 |
+
},
|
| 1283 |
+
{
|
| 1284 |
+
"name": "remaining_playback_time_calculation",
|
| 1285 |
+
"score": 0.07142857142857142,
|
| 1286 |
+
"eval_type": "rule",
|
| 1287 |
+
"num_demo": 1,
|
| 1288 |
+
"num_query": 14
|
| 1289 |
+
},
|
| 1290 |
+
{
|
| 1291 |
+
"name": "multiple_states_identify_americas",
|
| 1292 |
+
"score": 0.7285714285714285,
|
| 1293 |
+
"eval_type": "rule",
|
| 1294 |
+
"num_demo": 1,
|
| 1295 |
+
"num_query": 14
|
| 1296 |
+
},
|
| 1297 |
+
{
|
| 1298 |
+
"name": "medical_cell_recognition",
|
| 1299 |
+
"score": 0.8571428571428571,
|
| 1300 |
+
"eval_type": "rule",
|
| 1301 |
+
"num_demo": 1,
|
| 1302 |
+
"num_query": 14
|
| 1303 |
+
},
|
| 1304 |
+
{
|
| 1305 |
+
"name": "adapted_cvbench_distance",
|
| 1306 |
+
"score": 0.7857142857142857,
|
| 1307 |
+
"eval_type": "rule",
|
| 1308 |
+
"num_demo": 1,
|
| 1309 |
+
"num_query": 14
|
| 1310 |
+
},
|
| 1311 |
+
{
|
| 1312 |
+
"name": "adapted_cvbench_count",
|
| 1313 |
+
"score": 0.5,
|
| 1314 |
+
"eval_type": "rule",
|
| 1315 |
+
"num_demo": 1,
|
| 1316 |
+
"num_query": 14
|
| 1317 |
+
},
|
| 1318 |
+
{
|
| 1319 |
+
"name": "adapted_cvbench_depth",
|
| 1320 |
+
"score": 0.7142857142857143,
|
| 1321 |
+
"eval_type": "rule",
|
| 1322 |
+
"num_demo": 1,
|
| 1323 |
+
"num_query": 14
|
| 1324 |
+
},
|
| 1325 |
+
{
|
| 1326 |
+
"name": "adapted_cvbench_relation",
|
| 1327 |
+
"score": 0.5714285714285714,
|
| 1328 |
+
"eval_type": "rule",
|
| 1329 |
+
"num_demo": 1,
|
| 1330 |
+
"num_query": 14
|
| 1331 |
+
},
|
| 1332 |
+
{
|
| 1333 |
+
"name": "symbolic_graphics_programs_computer_aided_design",
|
| 1334 |
+
"score": 0.42857142857142855,
|
| 1335 |
+
"eval_type": "rule",
|
| 1336 |
+
"num_demo": 1,
|
| 1337 |
+
"num_query": 14
|
| 1338 |
+
},
|
| 1339 |
+
{
|
| 1340 |
+
"name": "chess_find_legal_moves",
|
| 1341 |
+
"score": 0.18247734432946935,
|
| 1342 |
+
"eval_type": "rule",
|
| 1343 |
+
"num_demo": 1,
|
| 1344 |
+
"num_query": 14
|
| 1345 |
+
},
|
| 1346 |
+
{
|
| 1347 |
+
"name": "symbolic_graphics_programs_scalable_vector_graphics",
|
| 1348 |
+
"score": 0.1111111111111111,
|
| 1349 |
+
"eval_type": "rule",
|
| 1350 |
+
"num_demo": 1,
|
| 1351 |
+
"num_query": 18
|
| 1352 |
+
},
|
| 1353 |
+
{
|
| 1354 |
+
"name": "table_understanding_complex_question_answering",
|
| 1355 |
+
"score": 0.6428571428571429,
|
| 1356 |
+
"eval_type": "rule",
|
| 1357 |
+
"num_demo": 1,
|
| 1358 |
+
"num_query": 14
|
| 1359 |
+
},
|
| 1360 |
+
{
|
| 1361 |
+
"name": "table_understanding_fact_verification",
|
| 1362 |
+
"score": 0.8333333333333334,
|
| 1363 |
+
"eval_type": "rule",
|
| 1364 |
+
"num_demo": 1,
|
| 1365 |
+
"num_query": 14
|
| 1366 |
+
},
|
| 1367 |
+
{
|
| 1368 |
+
"name": "panel_images_multi_question",
|
| 1369 |
+
"score": 0.738095238095238,
|
| 1370 |
+
"eval_type": "rule",
|
| 1371 |
+
"num_demo": 1,
|
| 1372 |
+
"num_query": 14
|
| 1373 |
+
},
|
| 1374 |
+
{
|
| 1375 |
+
"name": "panel_images_single_question",
|
| 1376 |
+
"score": 0.9285714285714286,
|
| 1377 |
+
"eval_type": "rule",
|
| 1378 |
+
"num_demo": 1,
|
| 1379 |
+
"num_query": 14
|
| 1380 |
+
},
|
| 1381 |
+
{
|
| 1382 |
+
"name": "MMSoc_Misinformation_GossipCop",
|
| 1383 |
+
"score": 0.7142857142857143,
|
| 1384 |
+
"eval_type": "rule",
|
| 1385 |
+
"num_demo": 1,
|
| 1386 |
+
"num_query": 14
|
| 1387 |
+
},
|
| 1388 |
+
{
|
| 1389 |
+
"name": "MMSoc_HatefulMemes",
|
| 1390 |
+
"score": 0.6428571428571429,
|
| 1391 |
+
"eval_type": "rule",
|
| 1392 |
+
"num_demo": 1,
|
| 1393 |
+
"num_query": 14
|
| 1394 |
+
},
|
| 1395 |
+
{
|
| 1396 |
+
"name": "distinguish_ai_generated_image",
|
| 1397 |
+
"score": 0.631578947368421,
|
| 1398 |
+
"eval_type": "rule",
|
| 1399 |
+
"num_demo": 1,
|
| 1400 |
+
"num_query": 19
|
| 1401 |
+
},
|
| 1402 |
+
{
|
| 1403 |
+
"name": "MMSoc_Memotion",
|
| 1404 |
+
"score": 0.635294117647059,
|
| 1405 |
+
"eval_type": "rule",
|
| 1406 |
+
"num_demo": 1,
|
| 1407 |
+
"num_query": 17
|
| 1408 |
+
},
|
| 1409 |
+
{
|
| 1410 |
+
"name": "MMSoc_Misinformation_PolitiFact",
|
| 1411 |
+
"score": 0.8571428571428571,
|
| 1412 |
+
"eval_type": "rule",
|
| 1413 |
+
"num_demo": 1,
|
| 1414 |
+
"num_query": 14
|
| 1415 |
+
},
|
| 1416 |
+
{
|
| 1417 |
+
"name": "text_entity_replace",
|
| 1418 |
+
"score": 0.7142857142857143,
|
| 1419 |
+
"eval_type": "rule",
|
| 1420 |
+
"num_demo": 1,
|
| 1421 |
+
"num_query": 14
|
| 1422 |
+
},
|
| 1423 |
+
{
|
| 1424 |
+
"name": "poetry_acrostic_alliteration",
|
| 1425 |
+
"score": 0.8666666666666667,
|
| 1426 |
+
"eval_type": "rule",
|
| 1427 |
+
"num_demo": 0,
|
| 1428 |
+
"num_query": 15
|
| 1429 |
+
},
|
| 1430 |
+
{
|
| 1431 |
+
"name": "background_change",
|
| 1432 |
+
"score": 0.9285714285714286,
|
| 1433 |
+
"eval_type": "rule",
|
| 1434 |
+
"num_demo": 1,
|
| 1435 |
+
"num_query": 14
|
| 1436 |
+
},
|
| 1437 |
+
{
|
| 1438 |
+
"name": "poetry_acrostic",
|
| 1439 |
+
"score": 1.0,
|
| 1440 |
+
"eval_type": "rule",
|
| 1441 |
+
"num_demo": 0,
|
| 1442 |
+
"num_query": 15
|
| 1443 |
+
},
|
| 1444 |
+
{
|
| 1445 |
+
"name": "face_attribute_edit",
|
| 1446 |
+
"score": 0.5714285714285714,
|
| 1447 |
+
"eval_type": "rule",
|
| 1448 |
+
"num_demo": 1,
|
| 1449 |
+
"num_query": 14
|
| 1450 |
+
},
|
| 1451 |
+
{
|
| 1452 |
+
"name": "poetry_limerick",
|
| 1453 |
+
"score": 0.6666666666666666,
|
| 1454 |
+
"eval_type": "rule",
|
| 1455 |
+
"num_demo": 0,
|
| 1456 |
+
"num_query": 15
|
| 1457 |
+
},
|
| 1458 |
+
{
|
| 1459 |
+
"name": "face_swap",
|
| 1460 |
+
"score": 0.5714285714285714,
|
| 1461 |
+
"eval_type": "rule",
|
| 1462 |
+
"num_demo": 1,
|
| 1463 |
+
"num_query": 14
|
| 1464 |
+
},
|
| 1465 |
+
{
|
| 1466 |
+
"name": "poetry_custom_rhyming_scheme",
|
| 1467 |
+
"score": 0.26666666666666666,
|
| 1468 |
+
"eval_type": "rule",
|
| 1469 |
+
"num_demo": 0,
|
| 1470 |
+
"num_query": 15
|
| 1471 |
+
},
|
| 1472 |
+
{
|
| 1473 |
+
"name": "text_style",
|
| 1474 |
+
"score": 0.7857142857142857,
|
| 1475 |
+
"eval_type": "rule",
|
| 1476 |
+
"num_demo": 1,
|
| 1477 |
+
"num_query": 14
|
| 1478 |
+
},
|
| 1479 |
+
{
|
| 1480 |
+
"name": "out_of_context",
|
| 1481 |
+
"score": 0.8571428571428571,
|
| 1482 |
+
"eval_type": "rule",
|
| 1483 |
+
"num_demo": 1,
|
| 1484 |
+
"num_query": 14
|
| 1485 |
+
},
|
| 1486 |
+
{
|
| 1487 |
+
"name": "poetry_petrarchian_sonnet_optional_meter",
|
| 1488 |
+
"score": 0.0,
|
| 1489 |
+
"eval_type": "rule",
|
| 1490 |
+
"num_demo": 0,
|
| 1491 |
+
"num_query": 15
|
| 1492 |
+
},
|
| 1493 |
+
{
|
| 1494 |
+
"name": "poetry_haiku",
|
| 1495 |
+
"score": 0.4,
|
| 1496 |
+
"eval_type": "rule",
|
| 1497 |
+
"num_demo": 0,
|
| 1498 |
+
"num_query": 15
|
| 1499 |
+
},
|
| 1500 |
+
{
|
| 1501 |
+
"name": "clip_stable_diffusion_generate",
|
| 1502 |
+
"score": 0.6428571428571429,
|
| 1503 |
+
"eval_type": "rule",
|
| 1504 |
+
"num_demo": 1,
|
| 1505 |
+
"num_query": 14
|
| 1506 |
+
},
|
| 1507 |
+
{
|
| 1508 |
+
"name": "veracity",
|
| 1509 |
+
"score": 0.8571428571428571,
|
| 1510 |
+
"eval_type": "rule",
|
| 1511 |
+
"num_demo": 1,
|
| 1512 |
+
"num_query": 14
|
| 1513 |
+
},
|
| 1514 |
+
{
|
| 1515 |
+
"name": "poetry_shakespearean_sonnet",
|
| 1516 |
+
"score": 0.13333333333333333,
|
| 1517 |
+
"eval_type": "rule",
|
| 1518 |
+
"num_demo": 0,
|
| 1519 |
+
"num_query": 15
|
| 1520 |
+
},
|
| 1521 |
+
{
|
| 1522 |
+
"name": "counterfactual_arithmetic",
|
| 1523 |
+
"score": 1.0,
|
| 1524 |
+
"eval_type": "rule",
|
| 1525 |
+
"num_demo": 1,
|
| 1526 |
+
"num_query": 14
|
| 1527 |
+
},
|
| 1528 |
+
{
|
| 1529 |
+
"name": "screenshot_lighteval_math",
|
| 1530 |
+
"score": 0.8,
|
| 1531 |
+
"eval_type": "rule",
|
| 1532 |
+
"num_demo": 1,
|
| 1533 |
+
"num_query": 15
|
| 1534 |
+
},
|
| 1535 |
+
{
|
| 1536 |
+
"name": "maze_2d_8x8",
|
| 1537 |
+
"score": 0.0,
|
| 1538 |
+
"eval_type": "rule",
|
| 1539 |
+
"num_demo": 1,
|
| 1540 |
+
"num_query": 14
|
| 1541 |
+
},
|
| 1542 |
+
{
|
| 1543 |
+
"name": "shape_composition_shapes",
|
| 1544 |
+
"score": 0.5147108843537415,
|
| 1545 |
+
"eval_type": "rule",
|
| 1546 |
+
"num_demo": 1,
|
| 1547 |
+
"num_query": 14
|
| 1548 |
+
},
|
| 1549 |
+
{
|
| 1550 |
+
"name": "screenshot_theoremqa",
|
| 1551 |
+
"score": 0.6428571428571429,
|
| 1552 |
+
"eval_type": "rule",
|
| 1553 |
+
"num_demo": 1,
|
| 1554 |
+
"num_query": 14
|
| 1555 |
+
},
|
| 1556 |
+
{
|
| 1557 |
+
"name": "shape_composition_colours",
|
| 1558 |
+
"score": 0.3873299319727891,
|
| 1559 |
+
"eval_type": "rule",
|
| 1560 |
+
"num_demo": 1,
|
| 1561 |
+
"num_query": 14
|
| 1562 |
+
},
|
| 1563 |
+
{
|
| 1564 |
+
"name": "number_puzzle_sudoku",
|
| 1565 |
+
"score": 0.0,
|
| 1566 |
+
"eval_type": "rule",
|
| 1567 |
+
"num_demo": 1,
|
| 1568 |
+
"num_query": 15
|
| 1569 |
+
},
|
| 1570 |
+
{
|
| 1571 |
+
"name": "autorater_artifact",
|
| 1572 |
+
"score": 0.7142857142857143,
|
| 1573 |
+
"eval_type": "rule",
|
| 1574 |
+
"num_demo": 1,
|
| 1575 |
+
"num_query": 14
|
| 1576 |
+
},
|
| 1577 |
+
{
|
| 1578 |
+
"name": "number_puzzle_kakuro_5x5",
|
| 1579 |
+
"score": 0.0,
|
| 1580 |
+
"eval_type": "rule",
|
| 1581 |
+
"num_demo": 1,
|
| 1582 |
+
"num_query": 15
|
| 1583 |
+
},
|
| 1584 |
+
{
|
| 1585 |
+
"name": "autorater_artifact_reason",
|
| 1586 |
+
"score": 0.9333333333333333,
|
| 1587 |
+
"eval_type": "rule",
|
| 1588 |
+
"num_demo": 0,
|
| 1589 |
+
"num_query": 15
|
| 1590 |
+
},
|
| 1591 |
+
{
|
| 1592 |
+
"name": "chess_puzzles_crushing",
|
| 1593 |
+
"score": 0.0,
|
| 1594 |
+
"eval_type": "rule",
|
| 1595 |
+
"num_demo": 1,
|
| 1596 |
+
"num_query": 14
|
| 1597 |
+
},
|
| 1598 |
+
{
|
| 1599 |
+
"name": "app_layout_understanding_twitter",
|
| 1600 |
+
"score": 0.8571428571428571,
|
| 1601 |
+
"eval_type": "rule",
|
| 1602 |
+
"num_demo": 1,
|
| 1603 |
+
"num_query": 14
|
| 1604 |
+
},
|
| 1605 |
+
{
|
| 1606 |
+
"name": "app_layout_understanding_youtube",
|
| 1607 |
+
"score": 0.8571428571428571,
|
| 1608 |
+
"eval_type": "rule",
|
| 1609 |
+
"num_demo": 1,
|
| 1610 |
+
"num_query": 14
|
| 1611 |
+
},
|
| 1612 |
+
{
|
| 1613 |
+
"name": "chess_puzzles_checkmate",
|
| 1614 |
+
"score": 0.0,
|
| 1615 |
+
"eval_type": "rule",
|
| 1616 |
+
"num_demo": 1,
|
| 1617 |
+
"num_query": 14
|
| 1618 |
+
},
|
| 1619 |
+
{
|
| 1620 |
+
"name": "app_layout_understanding_tiktok",
|
| 1621 |
+
"score": 0.7857142857142857,
|
| 1622 |
+
"eval_type": "rule",
|
| 1623 |
+
"num_demo": 1,
|
| 1624 |
+
"num_query": 14
|
| 1625 |
+
},
|
| 1626 |
+
{
|
| 1627 |
+
"name": "app_layout_understanding_excel",
|
| 1628 |
+
"score": 0.6428571428571429,
|
| 1629 |
+
"eval_type": "rule",
|
| 1630 |
+
"num_demo": 1,
|
| 1631 |
+
"num_query": 14
|
| 1632 |
+
},
|
| 1633 |
+
{
|
| 1634 |
+
"name": "chess_puzzles_equality",
|
| 1635 |
+
"score": 0.06666666666666667,
|
| 1636 |
+
"eval_type": "rule",
|
| 1637 |
+
"num_demo": 1,
|
| 1638 |
+
"num_query": 15
|
| 1639 |
+
},
|
| 1640 |
+
{
|
| 1641 |
+
"name": "app_layout_understanding_amazon",
|
| 1642 |
+
"score": 0.8571428571428571,
|
| 1643 |
+
"eval_type": "rule",
|
| 1644 |
+
"num_demo": 1,
|
| 1645 |
+
"num_query": 14
|
| 1646 |
+
},
|
| 1647 |
+
{
|
| 1648 |
+
"name": "app_layout_understanding_notes",
|
| 1649 |
+
"score": 0.7142857142857143,
|
| 1650 |
+
"eval_type": "rule",
|
| 1651 |
+
"num_demo": 1,
|
| 1652 |
+
"num_query": 14
|
| 1653 |
+
},
|
| 1654 |
+
{
|
| 1655 |
+
"name": "app_layout_understanding_instagram",
|
| 1656 |
+
"score": 0.8571428571428571,
|
| 1657 |
+
"eval_type": "rule",
|
| 1658 |
+
"num_demo": 1,
|
| 1659 |
+
"num_query": 14
|
| 1660 |
+
},
|
| 1661 |
+
{
|
| 1662 |
+
"name": "contain_position_length",
|
| 1663 |
+
"score": 0.8666666666666667,
|
| 1664 |
+
"eval_type": "rule",
|
| 1665 |
+
"num_demo": 0,
|
| 1666 |
+
"num_query": 15
|
| 1667 |
+
},
|
| 1668 |
+
{
|
| 1669 |
+
"name": "app_layout_understanding_zoom",
|
| 1670 |
+
"score": 0.7333333333333333,
|
| 1671 |
+
"eval_type": "rule",
|
| 1672 |
+
"num_demo": 1,
|
| 1673 |
+
"num_query": 15
|
| 1674 |
+
},
|
| 1675 |
+
{
|
| 1676 |
+
"name": "app_layout_understanding_word",
|
| 1677 |
+
"score": 0.7142857142857143,
|
| 1678 |
+
"eval_type": "rule",
|
| 1679 |
+
"num_demo": 1,
|
| 1680 |
+
"num_query": 14
|
| 1681 |
+
},
|
| 1682 |
+
{
|
| 1683 |
+
"name": "contain_repeat_length",
|
| 1684 |
+
"score": 0.6666666666666666,
|
| 1685 |
+
"eval_type": "rule",
|
| 1686 |
+
"num_demo": 0,
|
| 1687 |
+
"num_query": 15
|
| 1688 |
+
},
|
| 1689 |
+
{
|
| 1690 |
+
"name": "app_layout_understanding_iphone_settings",
|
| 1691 |
+
"score": 1.0,
|
| 1692 |
+
"eval_type": "rule",
|
| 1693 |
+
"num_demo": 1,
|
| 1694 |
+
"num_query": 14
|
| 1695 |
+
},
|
| 1696 |
+
{
|
| 1697 |
+
"name": "app_layout_understanding_leetcode",
|
| 1698 |
+
"score": 0.8571428571428571,
|
| 1699 |
+
"eval_type": "rule",
|
| 1700 |
+
"num_demo": 1,
|
| 1701 |
+
"num_query": 14
|
| 1702 |
+
},
|
| 1703 |
+
{
|
| 1704 |
+
"name": "app_layout_understanding_ppt",
|
| 1705 |
+
"score": 0.7142857142857143,
|
| 1706 |
+
"eval_type": "rule",
|
| 1707 |
+
"num_demo": 1,
|
| 1708 |
+
"num_query": 14
|
| 1709 |
+
},
|
| 1710 |
+
{
|
| 1711 |
+
"name": "app_layout_understanding_alipay",
|
| 1712 |
+
"score": 0.8235294117647058,
|
| 1713 |
+
"eval_type": "rule",
|
| 1714 |
+
"num_demo": 1,
|
| 1715 |
+
"num_query": 17
|
| 1716 |
+
},
|
| 1717 |
+
{
|
| 1718 |
+
"name": "ocr_table_to_markdown",
|
| 1719 |
+
"score": 0.9285714285714286,
|
| 1720 |
+
"eval_type": "rule",
|
| 1721 |
+
"num_demo": 1,
|
| 1722 |
+
"num_query": 14
|
| 1723 |
+
},
|
| 1724 |
+
{
|
| 1725 |
+
"name": "ocr_table_to_latex",
|
| 1726 |
+
"score": 0.6428571428571429,
|
| 1727 |
+
"eval_type": "rule",
|
| 1728 |
+
"num_demo": 1,
|
| 1729 |
+
"num_query": 14
|
| 1730 |
+
},
|
| 1731 |
+
{
|
| 1732 |
+
"name": "contain_length",
|
| 1733 |
+
"score": 0.9333333333333333,
|
| 1734 |
+
"eval_type": "rule",
|
| 1735 |
+
"num_demo": 0,
|
| 1736 |
+
"num_query": 15
|
| 1737 |
+
},
|
| 1738 |
+
{
|
| 1739 |
+
"name": "ocr_resume_employer_plain",
|
| 1740 |
+
"score": 0.7142857142857143,
|
| 1741 |
+
"eval_type": "rule",
|
| 1742 |
+
"num_demo": 1,
|
| 1743 |
+
"num_query": 14
|
| 1744 |
+
},
|
| 1745 |
+
{
|
| 1746 |
+
"name": "ocr_article_journal",
|
| 1747 |
+
"score": 0.9285714285714286,
|
| 1748 |
+
"eval_type": "rule",
|
| 1749 |
+
"num_demo": 1,
|
| 1750 |
+
"num_query": 14
|
| 1751 |
+
},
|
| 1752 |
+
{
|
| 1753 |
+
"name": "contain_contain_length",
|
| 1754 |
+
"score": 1.0,
|
| 1755 |
+
"eval_type": "rule",
|
| 1756 |
+
"num_demo": 0,
|
| 1757 |
+
"num_query": 15
|
| 1758 |
+
},
|
| 1759 |
+
{
|
| 1760 |
+
"name": "ocr_resume_experience_plain",
|
| 1761 |
+
"score": 0.7142857142857143,
|
| 1762 |
+
"eval_type": "rule",
|
| 1763 |
+
"num_demo": 1,
|
| 1764 |
+
"num_query": 14
|
| 1765 |
+
},
|
| 1766 |
+
{
|
| 1767 |
+
"name": "pictionary_skribbl_io",
|
| 1768 |
+
"score": 0.5,
|
| 1769 |
+
"eval_type": "rule",
|
| 1770 |
+
"num_demo": 1,
|
| 1771 |
+
"num_query": 20
|
| 1772 |
+
},
|
| 1773 |
+
{
|
| 1774 |
+
"name": "ocr_math_text_latex",
|
| 1775 |
+
"score": 0.6428571428571429,
|
| 1776 |
+
"eval_type": "rule",
|
| 1777 |
+
"num_demo": 1,
|
| 1778 |
+
"num_query": 14
|
| 1779 |
+
},
|
| 1780 |
+
{
|
| 1781 |
+
"name": "pictionary_doodle_guess",
|
| 1782 |
+
"score": 0.9333333333333333,
|
| 1783 |
+
"eval_type": "rule",
|
| 1784 |
+
"num_demo": 1,
|
| 1785 |
+
"num_query": 15
|
| 1786 |
+
},
|
| 1787 |
+
{
|
| 1788 |
+
"name": "ocr_article_authors",
|
| 1789 |
+
"score": 0.8214285714285714,
|
| 1790 |
+
"eval_type": "rule",
|
| 1791 |
+
"num_demo": 1,
|
| 1792 |
+
"num_query": 14
|
| 1793 |
+
},
|
| 1794 |
+
{
|
| 1795 |
+
"name": "pictionary_genai_output_chinese",
|
| 1796 |
+
"score": 0.5,
|
| 1797 |
+
"eval_type": "rule",
|
| 1798 |
+
"num_demo": 1,
|
| 1799 |
+
"num_query": 14
|
| 1800 |
+
},
|
| 1801 |
+
{
|
| 1802 |
+
"name": "pictionary_cartoon_drawing_guess",
|
| 1803 |
+
"score": 0.7857142857142857,
|
| 1804 |
+
"eval_type": "rule",
|
| 1805 |
+
"num_demo": 1,
|
| 1806 |
+
"num_query": 14
|
| 1807 |
+
},
|
| 1808 |
+
{
|
| 1809 |
+
"name": "ocr_table_to_csv",
|
| 1810 |
+
"score": 0.6428571428571429,
|
| 1811 |
+
"eval_type": "rule",
|
| 1812 |
+
"num_demo": 1,
|
| 1813 |
+
"num_query": 14
|
| 1814 |
+
},
|
| 1815 |
+
{
|
| 1816 |
+
"name": "pictionary_chinese_food_img2en",
|
| 1817 |
+
"score": 0.7857142857142857,
|
| 1818 |
+
"eval_type": "rule",
|
| 1819 |
+
"num_demo": 1,
|
| 1820 |
+
"num_query": 14
|
| 1821 |
+
},
|
| 1822 |
+
{
|
| 1823 |
+
"name": "ocr_math_equation",
|
| 1824 |
+
"score": 0.6428571428571429,
|
| 1825 |
+
"eval_type": "rule",
|
| 1826 |
+
"num_demo": 1,
|
| 1827 |
+
"num_query": 14
|
| 1828 |
+
},
|
| 1829 |
+
{
|
| 1830 |
+
"name": "ocr_resume_school_plain",
|
| 1831 |
+
"score": 0.7857142857142857,
|
| 1832 |
+
"eval_type": "rule",
|
| 1833 |
+
"num_demo": 1,
|
| 1834 |
+
"num_query": 14
|
| 1835 |
+
},
|
| 1836 |
+
{
|
| 1837 |
+
"name": "ocr_table_to_html",
|
| 1838 |
+
"score": 0.7142857142857143,
|
| 1839 |
+
"eval_type": "rule",
|
| 1840 |
+
"num_demo": 1,
|
| 1841 |
+
"num_query": 14
|
| 1842 |
+
},
|
| 1843 |
+
{
|
| 1844 |
+
"name": "ocr_resume_skill_plain",
|
| 1845 |
+
"score": 0.5714285714285714,
|
| 1846 |
+
"eval_type": "rule",
|
| 1847 |
+
"num_demo": 1,
|
| 1848 |
+
"num_query": 14
|
| 1849 |
+
},
|
| 1850 |
+
{
|
| 1851 |
+
"name": "crossword_mini_5x5",
|
| 1852 |
+
"score": 0.6857142857142858,
|
| 1853 |
+
"eval_type": "rule",
|
| 1854 |
+
"num_demo": 1,
|
| 1855 |
+
"num_query": 14
|
| 1856 |
+
},
|
| 1857 |
+
{
|
| 1858 |
+
"name": "reward_models_i2t_reward",
|
| 1859 |
+
"score": 0.5,
|
| 1860 |
+
"eval_type": "rule",
|
| 1861 |
+
"num_demo": 1,
|
| 1862 |
+
"num_query": 14
|
| 1863 |
+
},
|
| 1864 |
+
{
|
| 1865 |
+
"name": "memorization_chinese_celebrity",
|
| 1866 |
+
"score": 0.7857142857142857,
|
| 1867 |
+
"eval_type": "rule",
|
| 1868 |
+
"num_demo": 1,
|
| 1869 |
+
"num_query": 14
|
| 1870 |
+
},
|
| 1871 |
+
{
|
| 1872 |
+
"name": "memorization_papers",
|
| 1873 |
+
"score": 0.6,
|
| 1874 |
+
"eval_type": "rule",
|
| 1875 |
+
"num_demo": 1,
|
| 1876 |
+
"num_query": 15
|
| 1877 |
+
},
|
| 1878 |
+
{
|
| 1879 |
+
"name": "memorization_famous_treaty",
|
| 1880 |
+
"score": 0.75,
|
| 1881 |
+
"eval_type": "rule",
|
| 1882 |
+
"num_demo": 1,
|
| 1883 |
+
"num_query": 14
|
| 1884 |
+
},
|
| 1885 |
+
{
|
| 1886 |
+
"name": "memorization_indian_celebrity",
|
| 1887 |
+
"score": 0.8214285714285714,
|
| 1888 |
+
"eval_type": "rule",
|
| 1889 |
+
"num_demo": 1,
|
| 1890 |
+
"num_query": 14
|
| 1891 |
+
},
|
| 1892 |
+
{
|
| 1893 |
+
"name": "research_website_parsing_blogpost",
|
| 1894 |
+
"score": 0.07142857142857142,
|
| 1895 |
+
"eval_type": "rule",
|
| 1896 |
+
"num_demo": 1,
|
| 1897 |
+
"num_query": 14
|
| 1898 |
+
},
|
| 1899 |
+
{
|
| 1900 |
+
"name": "research_website_parsing_publication",
|
| 1901 |
+
"score": 0.07142857142857142,
|
| 1902 |
+
"eval_type": "rule",
|
| 1903 |
+
"num_demo": 1,
|
| 1904 |
+
"num_query": 14
|
| 1905 |
+
},
|
| 1906 |
+
{
|
| 1907 |
+
"name": "research_website_parsing_homepage",
|
| 1908 |
+
"score": 0.21428571428571427,
|
| 1909 |
+
"eval_type": "rule",
|
| 1910 |
+
"num_demo": 1,
|
| 1911 |
+
"num_query": 14
|
| 1912 |
+
},
|
| 1913 |
+
{
|
| 1914 |
+
"name": "ascii_art_30",
|
| 1915 |
+
"score": 0.07142857142857142,
|
| 1916 |
+
"eval_type": "llm",
|
| 1917 |
+
"num_demo": 1,
|
| 1918 |
+
"num_query": 14
|
| 1919 |
+
},
|
| 1920 |
+
{
|
| 1921 |
+
"name": "humor_explanation",
|
| 1922 |
+
"score": 0.8066666666666668,
|
| 1923 |
+
"eval_type": "llm",
|
| 1924 |
+
"num_demo": 1,
|
| 1925 |
+
"num_query": 15
|
| 1926 |
+
},
|
| 1927 |
+
{
|
| 1928 |
+
"name": "science_figure_explanation",
|
| 1929 |
+
"score": 0.8034482758620691,
|
| 1930 |
+
"eval_type": "llm",
|
| 1931 |
+
"num_demo": 1,
|
| 1932 |
+
"num_query": 29
|
| 1933 |
+
},
|
| 1934 |
+
{
|
| 1935 |
+
"name": "vibe_eval_phrase",
|
| 1936 |
+
"score": 0.7214285714285714,
|
| 1937 |
+
"eval_type": "llm",
|
| 1938 |
+
"num_demo": 1,
|
| 1939 |
+
"num_query": 14
|
| 1940 |
+
},
|
| 1941 |
+
{
|
| 1942 |
+
"name": "traffic_accident_analysis",
|
| 1943 |
+
"score": 0.6214285714285713,
|
| 1944 |
+
"eval_type": "llm",
|
| 1945 |
+
"num_demo": 1,
|
| 1946 |
+
"num_query": 14
|
| 1947 |
+
},
|
| 1948 |
+
{
|
| 1949 |
+
"name": "figurative_speech_explanation",
|
| 1950 |
+
"score": 0.8344827586206895,
|
| 1951 |
+
"eval_type": "llm",
|
| 1952 |
+
"num_demo": 1,
|
| 1953 |
+
"num_query": 29
|
| 1954 |
+
},
|
| 1955 |
+
{
|
| 1956 |
+
"name": "table2latex_complex",
|
| 1957 |
+
"score": 0.7666666666666667,
|
| 1958 |
+
"eval_type": "llm",
|
| 1959 |
+
"num_demo": 1,
|
| 1960 |
+
"num_query": 9
|
| 1961 |
+
},
|
| 1962 |
+
{
|
| 1963 |
+
"name": "unusual_images",
|
| 1964 |
+
"score": 0.8586206896551722,
|
| 1965 |
+
"eval_type": "llm",
|
| 1966 |
+
"num_demo": 1,
|
| 1967 |
+
"num_query": 29
|
| 1968 |
+
},
|
| 1969 |
+
{
|
| 1970 |
+
"name": "art_explanation",
|
| 1971 |
+
"score": 0.7793103448275862,
|
| 1972 |
+
"eval_type": "llm",
|
| 1973 |
+
"num_demo": 1,
|
| 1974 |
+
"num_query": 29
|
| 1975 |
+
},
|
| 1976 |
+
{
|
| 1977 |
+
"name": "ocr_open_ended_qa",
|
| 1978 |
+
"score": 0.8310344827586207,
|
| 1979 |
+
"eval_type": "llm",
|
| 1980 |
+
"num_demo": 1,
|
| 1981 |
+
"num_query": 29
|
| 1982 |
+
},
|
| 1983 |
+
{
|
| 1984 |
+
"name": "bar_chart_interpretation",
|
| 1985 |
+
"score": 0.7551724137931033,
|
| 1986 |
+
"eval_type": "llm",
|
| 1987 |
+
"num_demo": 1,
|
| 1988 |
+
"num_query": 29
|
| 1989 |
+
},
|
| 1990 |
+
{
|
| 1991 |
+
"name": "scibench_w_solution_open_ended",
|
| 1992 |
+
"score": 0.5880000000000001,
|
| 1993 |
+
"eval_type": "llm",
|
| 1994 |
+
"num_demo": 1,
|
| 1995 |
+
"num_query": 25
|
| 1996 |
+
},
|
| 1997 |
+
{
|
| 1998 |
+
"name": "GUI_Chat_Hard",
|
| 1999 |
+
"score": 0.3692307692307693,
|
| 2000 |
+
"eval_type": "llm",
|
| 2001 |
+
"num_demo": 1,
|
| 2002 |
+
"num_query": 26
|
| 2003 |
+
},
|
| 2004 |
+
{
|
| 2005 |
+
"name": "image_humor_understanding",
|
| 2006 |
+
"score": 0.8896551724137929,
|
| 2007 |
+
"eval_type": "llm",
|
| 2008 |
+
"num_demo": 1,
|
| 2009 |
+
"num_query": 29
|
| 2010 |
+
},
|
| 2011 |
+
{
|
| 2012 |
+
"name": "defeasible_reasoning",
|
| 2013 |
+
"score": 0.8620689655172415,
|
| 2014 |
+
"eval_type": "llm",
|
| 2015 |
+
"num_demo": 1,
|
| 2016 |
+
"num_query": 29
|
| 2017 |
+
},
|
| 2018 |
+
{
|
| 2019 |
+
"name": "funny_image_title",
|
| 2020 |
+
"score": 0.6714285714285714,
|
| 2021 |
+
"eval_type": "llm",
|
| 2022 |
+
"num_demo": 1,
|
| 2023 |
+
"num_query": 14
|
| 2024 |
+
},
|
| 2025 |
+
{
|
| 2026 |
+
"name": "tweets_captioning",
|
| 2027 |
+
"score": 0.5428571428571428,
|
| 2028 |
+
"eval_type": "llm",
|
| 2029 |
+
"num_demo": 1,
|
| 2030 |
+
"num_query": 14
|
| 2031 |
+
},
|
| 2032 |
+
{
|
| 2033 |
+
"name": "graph_interpretation",
|
| 2034 |
+
"score": 0.8896551724137931,
|
| 2035 |
+
"eval_type": "llm",
|
| 2036 |
+
"num_demo": 1,
|
| 2037 |
+
"num_query": 29
|
| 2038 |
+
},
|
| 2039 |
+
{
|
| 2040 |
+
"name": "meme_explain",
|
| 2041 |
+
"score": 0.8857142857142858,
|
| 2042 |
+
"eval_type": "llm",
|
| 2043 |
+
"num_demo": 1,
|
| 2044 |
+
"num_query": 14
|
| 2045 |
+
},
|
| 2046 |
+
{
|
| 2047 |
+
"name": "guess_image_generation_prompt",
|
| 2048 |
+
"score": 0.805263157894737,
|
| 2049 |
+
"eval_type": "llm",
|
| 2050 |
+
"num_demo": 1,
|
| 2051 |
+
"num_query": 19
|
| 2052 |
+
},
|
| 2053 |
+
{
|
| 2054 |
+
"name": "visualization_with_code",
|
| 2055 |
+
"score": 0.7142857142857143,
|
| 2056 |
+
"eval_type": "llm",
|
| 2057 |
+
"num_demo": 1,
|
| 2058 |
+
"num_query": 14
|
| 2059 |
+
},
|
| 2060 |
+
{
|
| 2061 |
+
"name": "iq_test_open_ended",
|
| 2062 |
+
"score": 0.6896551724137931,
|
| 2063 |
+
"eval_type": "llm",
|
| 2064 |
+
"num_demo": 1,
|
| 2065 |
+
"num_query": 29
|
| 2066 |
+
},
|
| 2067 |
+
{
|
| 2068 |
+
"name": "electrocardiogram",
|
| 2069 |
+
"score": 0.3928571428571428,
|
| 2070 |
+
"eval_type": "llm",
|
| 2071 |
+
"num_demo": 1,
|
| 2072 |
+
"num_query": 14
|
| 2073 |
+
},
|
| 2074 |
+
{
|
| 2075 |
+
"name": "image_captioning_with_additional_requirements",
|
| 2076 |
+
"score": 0.9214285714285716,
|
| 2077 |
+
"eval_type": "llm",
|
| 2078 |
+
"num_demo": 1,
|
| 2079 |
+
"num_query": 14
|
| 2080 |
+
},
|
| 2081 |
+
{
|
| 2082 |
+
"name": "docci_image_description_long",
|
| 2083 |
+
"score": 0.7214285714285713,
|
| 2084 |
+
"eval_type": "llm",
|
| 2085 |
+
"num_demo": 1,
|
| 2086 |
+
"num_query": 14
|
| 2087 |
+
},
|
| 2088 |
+
{
|
| 2089 |
+
"name": "GUI_Chat_Easy",
|
| 2090 |
+
"score": 0.6076923076923078,
|
| 2091 |
+
"eval_type": "llm",
|
| 2092 |
+
"num_demo": 1,
|
| 2093 |
+
"num_query": 26
|
| 2094 |
+
},
|
| 2095 |
+
{
|
| 2096 |
+
"name": "bridge_strategies_advanced",
|
| 2097 |
+
"score": 0.15714285714285717,
|
| 2098 |
+
"eval_type": "llm",
|
| 2099 |
+
"num_demo": 1,
|
| 2100 |
+
"num_query": 14
|
| 2101 |
+
},
|
| 2102 |
+
{
|
| 2103 |
+
"name": "bridge_strategies_worldclass",
|
| 2104 |
+
"score": 0.13571428571428573,
|
| 2105 |
+
"eval_type": "llm",
|
| 2106 |
+
"num_demo": 1,
|
| 2107 |
+
"num_query": 14
|
| 2108 |
+
},
|
| 2109 |
+
{
|
| 2110 |
+
"name": "bridge_strategies_expert",
|
| 2111 |
+
"score": 0.37142857142857144,
|
| 2112 |
+
"eval_type": "llm",
|
| 2113 |
+
"num_demo": 1,
|
| 2114 |
+
"num_query": 14
|
| 2115 |
+
},
|
| 2116 |
+
{
|
| 2117 |
+
"name": "multi_lingual_Ruozhiba_expalnation_Spanish",
|
| 2118 |
+
"score": 0.6642857142857144,
|
| 2119 |
+
"eval_type": "llm",
|
| 2120 |
+
"num_demo": 1,
|
| 2121 |
+
"num_query": 14
|
| 2122 |
+
},
|
| 2123 |
+
{
|
| 2124 |
+
"name": "multi_lingual_Ruozhiba_expalnation_Japanese",
|
| 2125 |
+
"score": 0.6357142857142858,
|
| 2126 |
+
"eval_type": "llm",
|
| 2127 |
+
"num_demo": 1,
|
| 2128 |
+
"num_query": 14
|
| 2129 |
+
},
|
| 2130 |
+
{
|
| 2131 |
+
"name": "multi_lingual_Ruozhiba_expalnation_French",
|
| 2132 |
+
"score": 0.6285714285714287,
|
| 2133 |
+
"eval_type": "llm",
|
| 2134 |
+
"num_demo": 1,
|
| 2135 |
+
"num_query": 14
|
| 2136 |
+
},
|
| 2137 |
+
{
|
| 2138 |
+
"name": "multi_lingual_Ruozhiba_expalnation_Arabic",
|
| 2139 |
+
"score": 0.6785714285714286,
|
| 2140 |
+
"eval_type": "llm",
|
| 2141 |
+
"num_demo": 1,
|
| 2142 |
+
"num_query": 14
|
| 2143 |
+
},
|
| 2144 |
+
{
|
| 2145 |
+
"name": "multi_lingual_Ruozhiba_expalnation_Russian",
|
| 2146 |
+
"score": 0.6642857142857143,
|
| 2147 |
+
"eval_type": "llm",
|
| 2148 |
+
"num_demo": 1,
|
| 2149 |
+
"num_query": 14
|
| 2150 |
+
},
|
| 2151 |
+
{
|
| 2152 |
+
"name": "multi_lingual_Ruozhiba_expalnation_English",
|
| 2153 |
+
"score": 0.7357142857142858,
|
| 2154 |
+
"eval_type": "llm",
|
| 2155 |
+
"num_demo": 1,
|
| 2156 |
+
"num_query": 14
|
| 2157 |
+
},
|
| 2158 |
+
{
|
| 2159 |
+
"name": "table_understanding_fetaqa",
|
| 2160 |
+
"score": 0.6857142857142858,
|
| 2161 |
+
"eval_type": "llm",
|
| 2162 |
+
"num_demo": 1,
|
| 2163 |
+
"num_query": 14
|
| 2164 |
+
},
|
| 2165 |
+
{
|
| 2166 |
+
"name": "red_teaming_celebrity",
|
| 2167 |
+
"score": 0.8700000000000001,
|
| 2168 |
+
"eval_type": "llm",
|
| 2169 |
+
"num_demo": 0,
|
| 2170 |
+
"num_query": 20
|
| 2171 |
+
},
|
| 2172 |
+
{
|
| 2173 |
+
"name": "red_teaming_captcha",
|
| 2174 |
+
"score": 0.12105263157894738,
|
| 2175 |
+
"eval_type": "llm",
|
| 2176 |
+
"num_demo": 1,
|
| 2177 |
+
"num_query": 19
|
| 2178 |
+
},
|
| 2179 |
+
{
|
| 2180 |
+
"name": "red_teaming_jailbreak",
|
| 2181 |
+
"score": 0.5399999999999999,
|
| 2182 |
+
"eval_type": "llm",
|
| 2183 |
+
"num_demo": 0,
|
| 2184 |
+
"num_query": 20
|
| 2185 |
+
},
|
| 2186 |
+
{
|
| 2187 |
+
"name": "red_teaming_visualmisleading",
|
| 2188 |
+
"score": 0.8578947368421055,
|
| 2189 |
+
"eval_type": "llm",
|
| 2190 |
+
"num_demo": 1,
|
| 2191 |
+
"num_query": 19
|
| 2192 |
+
},
|
| 2193 |
+
{
|
| 2194 |
+
"name": "red_teaming_racial",
|
| 2195 |
+
"score": 0.7750000000000001,
|
| 2196 |
+
"eval_type": "llm",
|
| 2197 |
+
"num_demo": 0,
|
| 2198 |
+
"num_query": 20
|
| 2199 |
+
},
|
| 2200 |
+
{
|
| 2201 |
+
"name": "red_teaming_politics",
|
| 2202 |
+
"score": 0.745,
|
| 2203 |
+
"eval_type": "llm",
|
| 2204 |
+
"num_demo": 0,
|
| 2205 |
+
"num_query": 20
|
| 2206 |
+
}
|
| 2207 |
+
]
|