Archive verified research artifact: summary.json
Browse files
artifacts/experiments/runs/gate-search-mira-mmqa_ti-transfer_p2_r1-s17/summary.json
ADDED
|
@@ -0,0 +1,328 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"development_trial": true,
|
| 3 |
+
"effective_config_sha256": "0d251dfb7109568e3201187d3da43337c8ffd3df499076c9b07967cd8cbb92fd",
|
| 4 |
+
"history": [
|
| 5 |
+
{
|
| 6 |
+
"checkpoint": "/root/paper/experiments/runs/gate-search-mira-mmqa_ti-transfer_p2_r1-s17/epoch1.pt",
|
| 7 |
+
"checkpoint_sha256": "53cfa531df88678d2a717b6bc795aecbb6ce0262b32d320f9755e9540b0b82a6",
|
| 8 |
+
"epoch": 1,
|
| 9 |
+
"examples_without_trainable_graph_path": 0,
|
| 10 |
+
"mean_active_slots": 0.03793278119809067,
|
| 11 |
+
"mean_training_loss": 1.5066449747160902,
|
| 12 |
+
"questions_with_active_slots": 3,
|
| 13 |
+
"scores": [
|
| 14 |
+
{
|
| 15 |
+
"complete": true,
|
| 16 |
+
"dataset": "mmqa_ti",
|
| 17 |
+
"groups": {
|
| 18 |
+
"modality:image": {
|
| 19 |
+
"em": 23.56687898089172,
|
| 20 |
+
"f1": 26.235668789808916,
|
| 21 |
+
"n": 314
|
| 22 |
+
},
|
| 23 |
+
"modality:image+text": {
|
| 24 |
+
"em": 11.578947368421053,
|
| 25 |
+
"f1": 14.694736842105263,
|
| 26 |
+
"n": 95
|
| 27 |
+
},
|
| 28 |
+
"modality:text": {
|
| 29 |
+
"em": 25.609756097560975,
|
| 30 |
+
"f1": 31.6875,
|
| 31 |
+
"n": 656
|
| 32 |
+
},
|
| 33 |
+
"type:Compare(Compose(TableQ,ImageQ),Compose(TableQ,TextQ))": {
|
| 34 |
+
"em": 11.764705882352942,
|
| 35 |
+
"f1": 14.705882352941176,
|
| 36 |
+
"n": 17
|
| 37 |
+
},
|
| 38 |
+
"type:Compose(ImageQ,TextQ)": {
|
| 39 |
+
"em": 24.0,
|
| 40 |
+
"f1": 24.0,
|
| 41 |
+
"n": 25
|
| 42 |
+
},
|
| 43 |
+
"type:Compose(TextQ,ImageListQ)": {
|
| 44 |
+
"em": 4.0,
|
| 45 |
+
"f1": 8.92,
|
| 46 |
+
"n": 50
|
| 47 |
+
},
|
| 48 |
+
"type:ImageListQ": {
|
| 49 |
+
"em": 1.5625,
|
| 50 |
+
"f1": 6.1484375,
|
| 51 |
+
"n": 128
|
| 52 |
+
},
|
| 53 |
+
"type:ImageQ": {
|
| 54 |
+
"em": 38.70967741935484,
|
| 55 |
+
"f1": 40.05913978494624,
|
| 56 |
+
"n": 186
|
| 57 |
+
},
|
| 58 |
+
"type:Intersect(ImageListQ,TextQ)": {
|
| 59 |
+
"em": 33.333333333333336,
|
| 60 |
+
"f1": 33.333333333333336,
|
| 61 |
+
"n": 3
|
| 62 |
+
},
|
| 63 |
+
"type:TextQ": {
|
| 64 |
+
"em": 25.609756097560975,
|
| 65 |
+
"f1": 31.6875,
|
| 66 |
+
"n": 656
|
| 67 |
+
}
|
| 68 |
+
},
|
| 69 |
+
"metric_scale": "0-100",
|
| 70 |
+
"n_expected": 1065,
|
| 71 |
+
"n_missing": 0,
|
| 72 |
+
"n_predictions": 1065,
|
| 73 |
+
"parse_success_count": 1062,
|
| 74 |
+
"scores": {
|
| 75 |
+
"em": 23.75586854460094,
|
| 76 |
+
"f1": 28.564319248826287
|
| 77 |
+
},
|
| 78 |
+
"status_counts": {
|
| 79 |
+
"ok": 1065
|
| 80 |
+
}
|
| 81 |
+
}
|
| 82 |
+
],
|
| 83 |
+
"seconds": 3919.6008298397064
|
| 84 |
+
},
|
| 85 |
+
{
|
| 86 |
+
"checkpoint": "/root/paper/experiments/runs/gate-search-mira-mmqa_ti-transfer_p2_r1-s17/epoch2.pt",
|
| 87 |
+
"checkpoint_sha256": "0a5abe7045af7100a714f31868b175e22d98266cfd01bb5570ad927632af935b",
|
| 88 |
+
"epoch": 2,
|
| 89 |
+
"examples_without_trainable_graph_path": 0,
|
| 90 |
+
"mean_active_slots": 11.87184185353726,
|
| 91 |
+
"mean_training_loss": 1.2646497696589114,
|
| 92 |
+
"questions_with_active_slots": 1065,
|
| 93 |
+
"scores": [
|
| 94 |
+
{
|
| 95 |
+
"complete": true,
|
| 96 |
+
"dataset": "mmqa_ti",
|
| 97 |
+
"groups": {
|
| 98 |
+
"modality:image": {
|
| 99 |
+
"em": 24.840764331210192,
|
| 100 |
+
"f1": 28.831210191082803,
|
| 101 |
+
"n": 314
|
| 102 |
+
},
|
| 103 |
+
"modality:image+text": {
|
| 104 |
+
"em": 13.68421052631579,
|
| 105 |
+
"f1": 17.884210526315787,
|
| 106 |
+
"n": 95
|
| 107 |
+
},
|
| 108 |
+
"modality:text": {
|
| 109 |
+
"em": 27.591463414634145,
|
| 110 |
+
"f1": 33.36890243902439,
|
| 111 |
+
"n": 656
|
| 112 |
+
},
|
| 113 |
+
"type:Compare(Compose(TableQ,ImageQ),Compose(TableQ,TextQ))": {
|
| 114 |
+
"em": 17.647058823529413,
|
| 115 |
+
"f1": 20.58823529411765,
|
| 116 |
+
"n": 17
|
| 117 |
+
},
|
| 118 |
+
"type:Compose(ImageQ,TextQ)": {
|
| 119 |
+
"em": 28.0,
|
| 120 |
+
"f1": 28.0,
|
| 121 |
+
"n": 25
|
| 122 |
+
},
|
| 123 |
+
"type:Compose(TextQ,ImageListQ)": {
|
| 124 |
+
"em": 6.0,
|
| 125 |
+
"f1": 11.64,
|
| 126 |
+
"n": 50
|
| 127 |
+
},
|
| 128 |
+
"type:ImageListQ": {
|
| 129 |
+
"em": 0.0,
|
| 130 |
+
"f1": 8.3515625,
|
| 131 |
+
"n": 128
|
| 132 |
+
},
|
| 133 |
+
"type:ImageQ": {
|
| 134 |
+
"em": 41.935483870967744,
|
| 135 |
+
"f1": 42.924731182795696,
|
| 136 |
+
"n": 186
|
| 137 |
+
},
|
| 138 |
+
"type:Intersect(ImageListQ,TextQ)": {
|
| 139 |
+
"em": 0.0,
|
| 140 |
+
"f1": 22.333333333333332,
|
| 141 |
+
"n": 3
|
| 142 |
+
},
|
| 143 |
+
"type:TextQ": {
|
| 144 |
+
"em": 27.591463414634145,
|
| 145 |
+
"f1": 33.36890243902439,
|
| 146 |
+
"n": 656
|
| 147 |
+
}
|
| 148 |
+
},
|
| 149 |
+
"metric_scale": "0-100",
|
| 150 |
+
"n_expected": 1065,
|
| 151 |
+
"n_missing": 0,
|
| 152 |
+
"n_predictions": 1065,
|
| 153 |
+
"parse_success_count": 1063,
|
| 154 |
+
"scores": {
|
| 155 |
+
"em": 25.539906103286384,
|
| 156 |
+
"f1": 30.64976525821596
|
| 157 |
+
},
|
| 158 |
+
"status_counts": {
|
| 159 |
+
"ok": 1065
|
| 160 |
+
}
|
| 161 |
+
}
|
| 162 |
+
],
|
| 163 |
+
"seconds": 3795.9463155269623
|
| 164 |
+
},
|
| 165 |
+
{
|
| 166 |
+
"checkpoint": "/root/paper/experiments/runs/gate-search-mira-mmqa_ti-transfer_p2_r1-s17/epoch3.pt",
|
| 167 |
+
"checkpoint_sha256": "67f5e20ec17be4b1095d4821bdc4ce745e76057db2ee3595438dae6598c4290d",
|
| 168 |
+
"epoch": 3,
|
| 169 |
+
"examples_without_trainable_graph_path": 0,
|
| 170 |
+
"mean_active_slots": 12.094504832409564,
|
| 171 |
+
"mean_training_loss": 1.101622267909884,
|
| 172 |
+
"questions_with_active_slots": 1065,
|
| 173 |
+
"scores": [
|
| 174 |
+
{
|
| 175 |
+
"complete": true,
|
| 176 |
+
"dataset": "mmqa_ti",
|
| 177 |
+
"groups": {
|
| 178 |
+
"modality:image": {
|
| 179 |
+
"em": 21.97452229299363,
|
| 180 |
+
"f1": 25.71656050955414,
|
| 181 |
+
"n": 314
|
| 182 |
+
},
|
| 183 |
+
"modality:image+text": {
|
| 184 |
+
"em": 9.473684210526315,
|
| 185 |
+
"f1": 13.326315789473684,
|
| 186 |
+
"n": 95
|
| 187 |
+
},
|
| 188 |
+
"modality:text": {
|
| 189 |
+
"em": 28.201219512195124,
|
| 190 |
+
"f1": 34.49390243902439,
|
| 191 |
+
"n": 656
|
| 192 |
+
},
|
| 193 |
+
"type:Compare(Compose(TableQ,ImageQ),Compose(TableQ,TextQ))": {
|
| 194 |
+
"em": 5.882352941176471,
|
| 195 |
+
"f1": 8.823529411764707,
|
| 196 |
+
"n": 17
|
| 197 |
+
},
|
| 198 |
+
"type:Compose(ImageQ,TextQ)": {
|
| 199 |
+
"em": 16.0,
|
| 200 |
+
"f1": 16.0,
|
| 201 |
+
"n": 25
|
| 202 |
+
},
|
| 203 |
+
"type:Compose(TextQ,ImageListQ)": {
|
| 204 |
+
"em": 8.0,
|
| 205 |
+
"f1": 12.98,
|
| 206 |
+
"n": 50
|
| 207 |
+
},
|
| 208 |
+
"type:ImageListQ": {
|
| 209 |
+
"em": 0.78125,
|
| 210 |
+
"f1": 8.5234375,
|
| 211 |
+
"n": 128
|
| 212 |
+
},
|
| 213 |
+
"type:ImageQ": {
|
| 214 |
+
"em": 36.55913978494624,
|
| 215 |
+
"f1": 37.54838709677419,
|
| 216 |
+
"n": 186
|
| 217 |
+
},
|
| 218 |
+
"type:Intersect(ImageListQ,TextQ)": {
|
| 219 |
+
"em": 0.0,
|
| 220 |
+
"f1": 22.333333333333332,
|
| 221 |
+
"n": 3
|
| 222 |
+
},
|
| 223 |
+
"type:TextQ": {
|
| 224 |
+
"em": 28.201219512195124,
|
| 225 |
+
"f1": 34.49390243902439,
|
| 226 |
+
"n": 656
|
| 227 |
+
}
|
| 228 |
+
},
|
| 229 |
+
"metric_scale": "0-100",
|
| 230 |
+
"n_expected": 1065,
|
| 231 |
+
"n_missing": 0,
|
| 232 |
+
"n_predictions": 1065,
|
| 233 |
+
"parse_success_count": 1061,
|
| 234 |
+
"scores": {
|
| 235 |
+
"em": 24.694835680751172,
|
| 236 |
+
"f1": 30.017840375586854
|
| 237 |
+
},
|
| 238 |
+
"status_counts": {
|
| 239 |
+
"ok": 1065
|
| 240 |
+
}
|
| 241 |
+
}
|
| 242 |
+
],
|
| 243 |
+
"seconds": 3882.2563173770905
|
| 244 |
+
}
|
| 245 |
+
],
|
| 246 |
+
"selected": {
|
| 247 |
+
"checkpoint": "/root/paper/experiments/runs/gate-search-mira-mmqa_ti-transfer_p2_r1-s17/epoch2.pt",
|
| 248 |
+
"checkpoint_sha256": "0a5abe7045af7100a714f31868b175e22d98266cfd01bb5570ad927632af935b",
|
| 249 |
+
"epoch": 2,
|
| 250 |
+
"examples_without_trainable_graph_path": 0,
|
| 251 |
+
"mean_active_slots": 11.87184185353726,
|
| 252 |
+
"mean_training_loss": 1.2646497696589114,
|
| 253 |
+
"questions_with_active_slots": 1065,
|
| 254 |
+
"scores": [
|
| 255 |
+
{
|
| 256 |
+
"complete": true,
|
| 257 |
+
"dataset": "mmqa_ti",
|
| 258 |
+
"groups": {
|
| 259 |
+
"modality:image": {
|
| 260 |
+
"em": 24.840764331210192,
|
| 261 |
+
"f1": 28.831210191082803,
|
| 262 |
+
"n": 314
|
| 263 |
+
},
|
| 264 |
+
"modality:image+text": {
|
| 265 |
+
"em": 13.68421052631579,
|
| 266 |
+
"f1": 17.884210526315787,
|
| 267 |
+
"n": 95
|
| 268 |
+
},
|
| 269 |
+
"modality:text": {
|
| 270 |
+
"em": 27.591463414634145,
|
| 271 |
+
"f1": 33.36890243902439,
|
| 272 |
+
"n": 656
|
| 273 |
+
},
|
| 274 |
+
"type:Compare(Compose(TableQ,ImageQ),Compose(TableQ,TextQ))": {
|
| 275 |
+
"em": 17.647058823529413,
|
| 276 |
+
"f1": 20.58823529411765,
|
| 277 |
+
"n": 17
|
| 278 |
+
},
|
| 279 |
+
"type:Compose(ImageQ,TextQ)": {
|
| 280 |
+
"em": 28.0,
|
| 281 |
+
"f1": 28.0,
|
| 282 |
+
"n": 25
|
| 283 |
+
},
|
| 284 |
+
"type:Compose(TextQ,ImageListQ)": {
|
| 285 |
+
"em": 6.0,
|
| 286 |
+
"f1": 11.64,
|
| 287 |
+
"n": 50
|
| 288 |
+
},
|
| 289 |
+
"type:ImageListQ": {
|
| 290 |
+
"em": 0.0,
|
| 291 |
+
"f1": 8.3515625,
|
| 292 |
+
"n": 128
|
| 293 |
+
},
|
| 294 |
+
"type:ImageQ": {
|
| 295 |
+
"em": 41.935483870967744,
|
| 296 |
+
"f1": 42.924731182795696,
|
| 297 |
+
"n": 186
|
| 298 |
+
},
|
| 299 |
+
"type:Intersect(ImageListQ,TextQ)": {
|
| 300 |
+
"em": 0.0,
|
| 301 |
+
"f1": 22.333333333333332,
|
| 302 |
+
"n": 3
|
| 303 |
+
},
|
| 304 |
+
"type:TextQ": {
|
| 305 |
+
"em": 27.591463414634145,
|
| 306 |
+
"f1": 33.36890243902439,
|
| 307 |
+
"n": 656
|
| 308 |
+
}
|
| 309 |
+
},
|
| 310 |
+
"metric_scale": "0-100",
|
| 311 |
+
"n_expected": 1065,
|
| 312 |
+
"n_missing": 0,
|
| 313 |
+
"n_predictions": 1065,
|
| 314 |
+
"parse_success_count": 1063,
|
| 315 |
+
"scores": {
|
| 316 |
+
"em": 25.539906103286384,
|
| 317 |
+
"f1": 30.64976525821596
|
| 318 |
+
},
|
| 319 |
+
"status_counts": {
|
| 320 |
+
"ok": 1065
|
| 321 |
+
}
|
| 322 |
+
}
|
| 323 |
+
],
|
| 324 |
+
"seconds": 3795.9463155269623
|
| 325 |
+
},
|
| 326 |
+
"selection_rule": "Separate training per dataset; ScienceQA accuracy; MMQA list F1, tie list EM; tie earlier epoch; val_selection only",
|
| 327 |
+
"test_used": false
|
| 328 |
+
}
|