Delete folder add_sub_sorl_v1_abs10_K1_10K with huggingface_hub
Browse files
add_sub_sorl_v1_abs10_K1_10K/config.json
DELETED
|
@@ -1,37 +0,0 @@
|
|
| 1 |
-
{
|
| 2 |
-
"architectures": [
|
| 3 |
-
"SorlModelWrapper"
|
| 4 |
-
],
|
| 5 |
-
"attention_bias": false,
|
| 6 |
-
"attention_dropout": 0.0,
|
| 7 |
-
"bos_token_id": null,
|
| 8 |
-
"dtype": "float32",
|
| 9 |
-
"eos_token_id": null,
|
| 10 |
-
"head_dim": 128,
|
| 11 |
-
"hidden_act": "silu",
|
| 12 |
-
"hidden_size": 510,
|
| 13 |
-
"initializer_range": 0.02,
|
| 14 |
-
"intermediate_size": 2040,
|
| 15 |
-
"layer_types": [
|
| 16 |
-
"full_attention",
|
| 17 |
-
"full_attention"
|
| 18 |
-
],
|
| 19 |
-
"max_position_embeddings": 128,
|
| 20 |
-
"max_window_layers": 28,
|
| 21 |
-
"model_type": "qwen3",
|
| 22 |
-
"num_attention_heads": 3,
|
| 23 |
-
"num_hidden_layers": 2,
|
| 24 |
-
"num_key_value_heads": 3,
|
| 25 |
-
"pad_token_id": null,
|
| 26 |
-
"rms_norm_eps": 1e-06,
|
| 27 |
-
"rope_parameters": {
|
| 28 |
-
"rope_theta": 10000.0,
|
| 29 |
-
"rope_type": "default"
|
| 30 |
-
},
|
| 31 |
-
"sliding_window": null,
|
| 32 |
-
"tie_word_embeddings": false,
|
| 33 |
-
"transformers_version": "5.5.0",
|
| 34 |
-
"use_cache": true,
|
| 35 |
-
"use_sliding_window": false,
|
| 36 |
-
"vocab_size": 151654
|
| 37 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
add_sub_sorl_v1_abs10_K1_10K/generation_config.json
DELETED
|
@@ -1,7 +0,0 @@
|
|
| 1 |
-
{
|
| 2 |
-
"_from_model_config": true,
|
| 3 |
-
"output_attentions": false,
|
| 4 |
-
"output_hidden_states": false,
|
| 5 |
-
"transformers_version": "5.5.0",
|
| 6 |
-
"use_cache": true
|
| 7 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
add_sub_sorl_v1_abs10_K1_10K/metrics.json
DELETED
|
@@ -1,1297 +0,0 @@
|
|
| 1 |
-
{
|
| 2 |
-
"history": {
|
| 3 |
-
"step": [
|
| 4 |
-
50,
|
| 5 |
-
100,
|
| 6 |
-
150,
|
| 7 |
-
207,
|
| 8 |
-
257,
|
| 9 |
-
307,
|
| 10 |
-
364,
|
| 11 |
-
414,
|
| 12 |
-
464,
|
| 13 |
-
521,
|
| 14 |
-
571,
|
| 15 |
-
621,
|
| 16 |
-
678,
|
| 17 |
-
728,
|
| 18 |
-
778,
|
| 19 |
-
835,
|
| 20 |
-
885,
|
| 21 |
-
935,
|
| 22 |
-
992,
|
| 23 |
-
1042,
|
| 24 |
-
1092,
|
| 25 |
-
1149,
|
| 26 |
-
1199,
|
| 27 |
-
1249,
|
| 28 |
-
1306,
|
| 29 |
-
1356,
|
| 30 |
-
1406,
|
| 31 |
-
1463,
|
| 32 |
-
1513,
|
| 33 |
-
1563
|
| 34 |
-
],
|
| 35 |
-
"loss": [
|
| 36 |
-
8.773706436157227,
|
| 37 |
-
4.004265308380127,
|
| 38 |
-
3.2570762634277344,
|
| 39 |
-
3.041517734527588,
|
| 40 |
-
2.8982486724853516,
|
| 41 |
-
2.4261059761047363,
|
| 42 |
-
-1.0320148468017578,
|
| 43 |
-
-6.537303924560547,
|
| 44 |
-
-7.572015285491943,
|
| 45 |
-
-8.820948600769043,
|
| 46 |
-
-9.133442878723145,
|
| 47 |
-
-9.698243141174316,
|
| 48 |
-
-10.157323837280273,
|
| 49 |
-
-9.581876754760742,
|
| 50 |
-
-10.299582481384277,
|
| 51 |
-
-8.835819244384766,
|
| 52 |
-
-7.875391006469727,
|
| 53 |
-
-6.175548553466797,
|
| 54 |
-
-4.922896385192871,
|
| 55 |
-
-5.274993896484375,
|
| 56 |
-
-3.509742021560669,
|
| 57 |
-
-2.9279913902282715,
|
| 58 |
-
-2.248248815536499,
|
| 59 |
-
-1.4432270526885986,
|
| 60 |
-
-0.9849211573600769,
|
| 61 |
-
-0.7613394260406494,
|
| 62 |
-
-0.9093202948570251,
|
| 63 |
-
-0.20168478786945343,
|
| 64 |
-
-0.5009633302688599,
|
| 65 |
-
-0.4359404742717743
|
| 66 |
-
],
|
| 67 |
-
"base_loss": [
|
| 68 |
-
6.400631904602051,
|
| 69 |
-
2.4112229347229004,
|
| 70 |
-
1.8464170694351196,
|
| 71 |
-
1.8753893375396729,
|
| 72 |
-
1.8903639316558838,
|
| 73 |
-
1.8746846914291382,
|
| 74 |
-
1.8820133209228516,
|
| 75 |
-
1.855467438697815,
|
| 76 |
-
1.7718533277511597,
|
| 77 |
-
1.7219583988189697,
|
| 78 |
-
1.7517011165618896,
|
| 79 |
-
1.6724598407745361,
|
| 80 |
-
1.647750973701477,
|
| 81 |
-
1.4754587411880493,
|
| 82 |
-
1.4356294870376587,
|
| 83 |
-
1.2360666990280151,
|
| 84 |
-
1.1430914402008057,
|
| 85 |
-
0.8601657748222351,
|
| 86 |
-
0.6953770518302917,
|
| 87 |
-
0.6893697381019592,
|
| 88 |
-
0.4810207784175873,
|
| 89 |
-
0.39060816168785095,
|
| 90 |
-
0.3150084316730499,
|
| 91 |
-
0.2136605978012085,
|
| 92 |
-
0.163777157664299,
|
| 93 |
-
0.12215958535671234,
|
| 94 |
-
0.13819025456905365,
|
| 95 |
-
0.08133143931627274,
|
| 96 |
-
0.08307908475399017,
|
| 97 |
-
0.07490883022546768
|
| 98 |
-
],
|
| 99 |
-
"info_loss": [
|
| 100 |
-
-0.21113967895507812,
|
| 101 |
-
-0.04423642158508301,
|
| 102 |
-
-0.04978370666503906,
|
| 103 |
-
-0.0719677209854126,
|
| 104 |
-
-0.087005615234375,
|
| 105 |
-
-0.13271820545196533,
|
| 106 |
-
-0.4793226718902588,
|
| 107 |
-
-1.025976300239563,
|
| 108 |
-
-1.1140811443328857,
|
| 109 |
-
-1.2247107028961182,
|
| 110 |
-
-1.256365180015564,
|
| 111 |
-
-1.3006340265274048,
|
| 112 |
-
-1.3388965129852295,
|
| 113 |
-
-1.2538009881973267,
|
| 114 |
-
-1.3090555667877197,
|
| 115 |
-
-1.1411150693893433,
|
| 116 |
-
-1.0133163928985596,
|
| 117 |
-
-0.8001939654350281,
|
| 118 |
-
-0.6398270130157471,
|
| 119 |
-
-0.6569589376449585,
|
| 120 |
-
-0.44588613510131836,
|
| 121 |
-
-0.368633508682251,
|
| 122 |
-
-0.2934108078479767,
|
| 123 |
-
-0.1994001418352127,
|
| 124 |
-
-0.14748595654964447,
|
| 125 |
-
-0.11586211621761322,
|
| 126 |
-
-0.12713156640529633,
|
| 127 |
-
-0.04881247505545616,
|
| 128 |
-
-0.07891618460416794,
|
| 129 |
-
-0.06897100806236267
|
| 130 |
-
],
|
| 131 |
-
"abs_loss": [
|
| 132 |
-
2.0780529975891113,
|
| 133 |
-
1.8440923690795898,
|
| 134 |
-
1.8333244323730469,
|
| 135 |
-
1.8426448106765747,
|
| 136 |
-
1.8403841257095337,
|
| 137 |
-
1.8378067016601562,
|
| 138 |
-
1.8460584878921509,
|
| 139 |
-
1.7702795267105103,
|
| 140 |
-
1.5657355785369873,
|
| 141 |
-
1.3797943592071533,
|
| 142 |
-
1.3185033798217773,
|
| 143 |
-
1.2104538679122925,
|
| 144 |
-
1.125502586364746,
|
| 145 |
-
0.9860450029373169,
|
| 146 |
-
0.8728498220443726,
|
| 147 |
-
0.7434267997741699,
|
| 148 |
-
0.7268884778022766,
|
| 149 |
-
0.5966061949729919,
|
| 150 |
-
0.5761818885803223,
|
| 151 |
-
0.4872754216194153,
|
| 152 |
-
0.42560893297195435,
|
| 153 |
-
0.3502724766731262,
|
| 154 |
-
0.29791387915611267,
|
| 155 |
-
0.267427533864975,
|
| 156 |
-
0.24475598335266113,
|
| 157 |
-
0.22809018194675446,
|
| 158 |
-
0.13244910538196564,
|
| 159 |
-
0.15626277029514313,
|
| 160 |
-
0.18973781168460846,
|
| 161 |
-
0.15587982535362244
|
| 162 |
-
],
|
| 163 |
-
"zipf_loss": [
|
| 164 |
-
4.276666164398193,
|
| 165 |
-
1.8509973287582397,
|
| 166 |
-
1.7251636981964111,
|
| 167 |
-
1.7015411853790283,
|
| 168 |
-
1.6939024925231934,
|
| 169 |
-
1.6948225498199463,
|
| 170 |
-
1.6945927143096924,
|
| 171 |
-
1.6899633407592773,
|
| 172 |
-
1.6403698921203613,
|
| 173 |
-
1.5662205219268799,
|
| 174 |
-
1.546657681465149,
|
| 175 |
-
1.514591932296753,
|
| 176 |
-
1.4713406562805176,
|
| 177 |
-
1.3820693492889404,
|
| 178 |
-
1.2680578231811523,
|
| 179 |
-
1.26492178440094,
|
| 180 |
-
1.0419921875,
|
| 181 |
-
0.9065651893615723,
|
| 182 |
-
0.7223787903785706,
|
| 183 |
-
0.5564987659454346,
|
| 184 |
-
0.42553776502609253,
|
| 185 |
-
0.33270835876464844,
|
| 186 |
-
0.3410593867301941,
|
| 187 |
-
0.3103709816932678,
|
| 188 |
-
0.30168575048446655,
|
| 189 |
-
0.25231313705444336,
|
| 190 |
-
0.21056024730205536,
|
| 191 |
-
0.18948225677013397,
|
| 192 |
-
0.18614569306373596,
|
| 193 |
-
0.16327276825904846
|
| 194 |
-
],
|
| 195 |
-
"denoise_loss": [],
|
| 196 |
-
"ortho_loss": [
|
| 197 |
-
0.3211989104747772,
|
| 198 |
-
0.19110506772994995,
|
| 199 |
-
0.11888417601585388,
|
| 200 |
-
0.0979497954249382,
|
| 201 |
-
0.09260708838701248,
|
| 202 |
-
0.10610710084438324,
|
| 203 |
-
0.14702288806438446,
|
| 204 |
-
0.21001708507537842,
|
| 205 |
-
0.2301187366247177,
|
| 206 |
-
0.24748258292675018,
|
| 207 |
-
0.2629227936267853,
|
| 208 |
-
0.26881012320518494,
|
| 209 |
-
0.269039511680603,
|
| 210 |
-
0.26941126585006714,
|
| 211 |
-
0.26823461055755615,
|
| 212 |
-
0.26427075266838074,
|
| 213 |
-
0.2743089199066162,
|
| 214 |
-
0.2734338045120239,
|
| 215 |
-
0.27655836939811707,
|
| 216 |
-
0.2692953944206238,
|
| 217 |
-
0.27160370349884033,
|
| 218 |
-
0.27964547276496887,
|
| 219 |
-
0.2874664068222046,
|
| 220 |
-
0.2921176552772522,
|
| 221 |
-
0.2956086993217468,
|
| 222 |
-
0.29919639229774475,
|
| 223 |
-
0.3037196099758148,
|
| 224 |
-
0.3071216940879822,
|
| 225 |
-
0.311845988035202,
|
| 226 |
-
0.3125845789909363
|
| 227 |
-
],
|
| 228 |
-
"lr": [
|
| 229 |
-
7.840000000000001e-05,
|
| 230 |
-
8e-05,
|
| 231 |
-
8e-05,
|
| 232 |
-
8e-05,
|
| 233 |
-
8e-05,
|
| 234 |
-
8e-05,
|
| 235 |
-
8e-05,
|
| 236 |
-
8e-05,
|
| 237 |
-
8e-05,
|
| 238 |
-
8e-05,
|
| 239 |
-
8e-05,
|
| 240 |
-
8e-05,
|
| 241 |
-
8e-05,
|
| 242 |
-
8e-05,
|
| 243 |
-
8e-05,
|
| 244 |
-
8e-05,
|
| 245 |
-
8e-05,
|
| 246 |
-
8e-05,
|
| 247 |
-
7.656578947368422e-05,
|
| 248 |
-
7.064473684210527e-05,
|
| 249 |
-
6.472368421052631e-05,
|
| 250 |
-
5.7973684210526325e-05,
|
| 251 |
-
5.2052631578947374e-05,
|
| 252 |
-
4.613157894736842e-05,
|
| 253 |
-
3.938157894736842e-05,
|
| 254 |
-
3.346052631578948e-05,
|
| 255 |
-
2.7539473684210523e-05,
|
| 256 |
-
2.078947368421053e-05,
|
| 257 |
-
1.4868421052631572e-05,
|
| 258 |
-
8.947368421052635e-06
|
| 259 |
-
],
|
| 260 |
-
"emb_lr": [],
|
| 261 |
-
"eval_step": [
|
| 262 |
-
150,
|
| 263 |
-
307,
|
| 264 |
-
464,
|
| 265 |
-
621,
|
| 266 |
-
778,
|
| 267 |
-
935,
|
| 268 |
-
1092,
|
| 269 |
-
1199,
|
| 270 |
-
1356,
|
| 271 |
-
1513
|
| 272 |
-
],
|
| 273 |
-
"eval_accuracy": [
|
| 274 |
-
0.0,
|
| 275 |
-
0.0,
|
| 276 |
-
0.0,
|
| 277 |
-
0.0,
|
| 278 |
-
0.0,
|
| 279 |
-
0.0,
|
| 280 |
-
0.0,
|
| 281 |
-
0.0,
|
| 282 |
-
0.0,
|
| 283 |
-
0.0
|
| 284 |
-
]
|
| 285 |
-
},
|
| 286 |
-
"final_accuracy": 0.0,
|
| 287 |
-
"sft_eval": {
|
| 288 |
-
"config": {
|
| 289 |
-
"ops": "add_sub",
|
| 290 |
-
"K": null,
|
| 291 |
-
"mode": "sft",
|
| 292 |
-
"n_digits": 6,
|
| 293 |
-
"n_per_split": 250
|
| 294 |
-
},
|
| 295 |
-
"splits": {
|
| 296 |
-
"add_S0": {
|
| 297 |
-
"full_accuracy": 0.944,
|
| 298 |
-
"n_examples": 250,
|
| 299 |
-
"per_subtask": {
|
| 300 |
-
"SA": {
|
| 301 |
-
"accuracy": 0.9908076165462902,
|
| 302 |
-
"count": 1523
|
| 303 |
-
},
|
| 304 |
-
"SS": {
|
| 305 |
-
"accuracy": 1.0,
|
| 306 |
-
"count": 227
|
| 307 |
-
}
|
| 308 |
-
}
|
| 309 |
-
},
|
| 310 |
-
"add_S1": {
|
| 311 |
-
"full_accuracy": 0.952,
|
| 312 |
-
"n_examples": 250,
|
| 313 |
-
"per_subtask": {
|
| 314 |
-
"SA": {
|
| 315 |
-
"accuracy": 0.9870848708487084,
|
| 316 |
-
"count": 542
|
| 317 |
-
},
|
| 318 |
-
"SC": {
|
| 319 |
-
"accuracy": 0.9952038369304557,
|
| 320 |
-
"count": 417
|
| 321 |
-
},
|
| 322 |
-
"SS": {
|
| 323 |
-
"accuracy": 1.0,
|
| 324 |
-
"count": 70
|
| 325 |
-
},
|
| 326 |
-
"UC": {
|
| 327 |
-
"accuracy": 0.9958391123439667,
|
| 328 |
-
"count": 721
|
| 329 |
-
}
|
| 330 |
-
}
|
| 331 |
-
},
|
| 332 |
-
"add_S2": {
|
| 333 |
-
"full_accuracy": 0.912,
|
| 334 |
-
"n_examples": 250,
|
| 335 |
-
"per_subtask": {
|
| 336 |
-
"SA": {
|
| 337 |
-
"accuracy": 0.970108695652174,
|
| 338 |
-
"count": 368
|
| 339 |
-
},
|
| 340 |
-
"SC": {
|
| 341 |
-
"accuracy": 0.9781931464174455,
|
| 342 |
-
"count": 321
|
| 343 |
-
},
|
| 344 |
-
"SS": {
|
| 345 |
-
"accuracy": 0.9736842105263158,
|
| 346 |
-
"count": 228
|
| 347 |
-
},
|
| 348 |
-
"UC": {
|
| 349 |
-
"accuracy": 0.9962335216572504,
|
| 350 |
-
"count": 531
|
| 351 |
-
},
|
| 352 |
-
"US": {
|
| 353 |
-
"accuracy": 1.0,
|
| 354 |
-
"count": 302
|
| 355 |
-
}
|
| 356 |
-
}
|
| 357 |
-
},
|
| 358 |
-
"add_S3": {
|
| 359 |
-
"full_accuracy": 0.756,
|
| 360 |
-
"n_examples": 250,
|
| 361 |
-
"per_subtask": {
|
| 362 |
-
"SA": {
|
| 363 |
-
"accuracy": 0.990228013029316,
|
| 364 |
-
"count": 307
|
| 365 |
-
},
|
| 366 |
-
"SC": {
|
| 367 |
-
"accuracy": 0.993127147766323,
|
| 368 |
-
"count": 291
|
| 369 |
-
},
|
| 370 |
-
"SS": {
|
| 371 |
-
"accuracy": 0.9823008849557522,
|
| 372 |
-
"count": 113
|
| 373 |
-
},
|
| 374 |
-
"UC": {
|
| 375 |
-
"accuracy": 0.8799171842650103,
|
| 376 |
-
"count": 483
|
| 377 |
-
},
|
| 378 |
-
"US": {
|
| 379 |
-
"accuracy": 1.0,
|
| 380 |
-
"count": 556
|
| 381 |
-
}
|
| 382 |
-
}
|
| 383 |
-
},
|
| 384 |
-
"add_S4": {
|
| 385 |
-
"full_accuracy": 0.544,
|
| 386 |
-
"n_examples": 250,
|
| 387 |
-
"per_subtask": {
|
| 388 |
-
"SA": {
|
| 389 |
-
"accuracy": 1.0,
|
| 390 |
-
"count": 238
|
| 391 |
-
},
|
| 392 |
-
"SC": {
|
| 393 |
-
"accuracy": 0.988929889298893,
|
| 394 |
-
"count": 271
|
| 395 |
-
},
|
| 396 |
-
"SS": {
|
| 397 |
-
"accuracy": 1.0,
|
| 398 |
-
"count": 59
|
| 399 |
-
},
|
| 400 |
-
"UC": {
|
| 401 |
-
"accuracy": 0.782716049382716,
|
| 402 |
-
"count": 405
|
| 403 |
-
},
|
| 404 |
-
"US": {
|
| 405 |
-
"accuracy": 0.9047619047619048,
|
| 406 |
-
"count": 777
|
| 407 |
-
}
|
| 408 |
-
}
|
| 409 |
-
},
|
| 410 |
-
"add_S5": {
|
| 411 |
-
"full_accuracy": 0.26,
|
| 412 |
-
"n_examples": 250,
|
| 413 |
-
"per_subtask": {
|
| 414 |
-
"SA": {
|
| 415 |
-
"accuracy": 0.988,
|
| 416 |
-
"count": 250
|
| 417 |
-
},
|
| 418 |
-
"SC": {
|
| 419 |
-
"accuracy": 1.0,
|
| 420 |
-
"count": 250
|
| 421 |
-
},
|
| 422 |
-
"UC": {
|
| 423 |
-
"accuracy": 0.452,
|
| 424 |
-
"count": 250
|
| 425 |
-
},
|
| 426 |
-
"US": {
|
| 427 |
-
"accuracy": 0.736,
|
| 428 |
-
"count": 1000
|
| 429 |
-
}
|
| 430 |
-
}
|
| 431 |
-
},
|
| 432 |
-
"add_S6": {
|
| 433 |
-
"full_accuracy": 0.656,
|
| 434 |
-
"n_examples": 250,
|
| 435 |
-
"per_subtask": {
|
| 436 |
-
"SC": {
|
| 437 |
-
"accuracy": 1.0,
|
| 438 |
-
"count": 250
|
| 439 |
-
},
|
| 440 |
-
"UC": {
|
| 441 |
-
"accuracy": 0.812,
|
| 442 |
-
"count": 250
|
| 443 |
-
},
|
| 444 |
-
"US": {
|
| 445 |
-
"accuracy": 0.8448,
|
| 446 |
-
"count": 1250
|
| 447 |
-
}
|
| 448 |
-
}
|
| 449 |
-
},
|
| 450 |
-
"add_random": {
|
| 451 |
-
"full_accuracy": 0.91,
|
| 452 |
-
"n_examples": 200,
|
| 453 |
-
"per_subtask": {
|
| 454 |
-
"SA": {
|
| 455 |
-
"accuracy": 0.9886621315192744,
|
| 456 |
-
"count": 441
|
| 457 |
-
},
|
| 458 |
-
"SC": {
|
| 459 |
-
"accuracy": 0.9936908517350158,
|
| 460 |
-
"count": 317
|
| 461 |
-
},
|
| 462 |
-
"SS": {
|
| 463 |
-
"accuracy": 0.9814814814814815,
|
| 464 |
-
"count": 54
|
| 465 |
-
},
|
| 466 |
-
"UC": {
|
| 467 |
-
"accuracy": 0.981203007518797,
|
| 468 |
-
"count": 532
|
| 469 |
-
},
|
| 470 |
-
"US": {
|
| 471 |
-
"accuracy": 0.9642857142857143,
|
| 472 |
-
"count": 56
|
| 473 |
-
}
|
| 474 |
-
}
|
| 475 |
-
},
|
| 476 |
-
"add_C3": {
|
| 477 |
-
"full_accuracy": 0.772,
|
| 478 |
-
"n_examples": 250,
|
| 479 |
-
"per_subtask": {
|
| 480 |
-
"SA": {
|
| 481 |
-
"accuracy": 0.9786666666666667,
|
| 482 |
-
"count": 750
|
| 483 |
-
},
|
| 484 |
-
"SC": {
|
| 485 |
-
"accuracy": 0.996,
|
| 486 |
-
"count": 250
|
| 487 |
-
},
|
| 488 |
-
"UC": {
|
| 489 |
-
"accuracy": 0.9096638655462185,
|
| 490 |
-
"count": 476
|
| 491 |
-
},
|
| 492 |
-
"US": {
|
| 493 |
-
"accuracy": 1.0,
|
| 494 |
-
"count": 274
|
| 495 |
-
}
|
| 496 |
-
}
|
| 497 |
-
},
|
| 498 |
-
"add_C4": {
|
| 499 |
-
"full_accuracy": 0.836,
|
| 500 |
-
"n_examples": 250,
|
| 501 |
-
"per_subtask": {
|
| 502 |
-
"SA": {
|
| 503 |
-
"accuracy": 0.986,
|
| 504 |
-
"count": 500
|
| 505 |
-
},
|
| 506 |
-
"SC": {
|
| 507 |
-
"accuracy": 0.996,
|
| 508 |
-
"count": 250
|
| 509 |
-
},
|
| 510 |
-
"UC": {
|
| 511 |
-
"accuracy": 0.9515885022692889,
|
| 512 |
-
"count": 661
|
| 513 |
-
},
|
| 514 |
-
"US": {
|
| 515 |
-
"accuracy": 0.9616519174041298,
|
| 516 |
-
"count": 339
|
| 517 |
-
}
|
| 518 |
-
}
|
| 519 |
-
},
|
| 520 |
-
"add_C5": {
|
| 521 |
-
"full_accuracy": 0.78,
|
| 522 |
-
"n_examples": 250,
|
| 523 |
-
"per_subtask": {
|
| 524 |
-
"SA": {
|
| 525 |
-
"accuracy": 0.996,
|
| 526 |
-
"count": 250
|
| 527 |
-
},
|
| 528 |
-
"SC": {
|
| 529 |
-
"accuracy": 1.0,
|
| 530 |
-
"count": 250
|
| 531 |
-
},
|
| 532 |
-
"UC": {
|
| 533 |
-
"accuracy": 0.9344262295081968,
|
| 534 |
-
"count": 732
|
| 535 |
-
},
|
| 536 |
-
"US": {
|
| 537 |
-
"accuracy": 0.9305019305019305,
|
| 538 |
-
"count": 518
|
| 539 |
-
}
|
| 540 |
-
}
|
| 541 |
-
},
|
| 542 |
-
"add_C6": {
|
| 543 |
-
"full_accuracy": 0.86,
|
| 544 |
-
"n_examples": 250,
|
| 545 |
-
"per_subtask": {
|
| 546 |
-
"SC": {
|
| 547 |
-
"accuracy": 1.0,
|
| 548 |
-
"count": 250
|
| 549 |
-
},
|
| 550 |
-
"UC": {
|
| 551 |
-
"accuracy": 0.9673423423423423,
|
| 552 |
-
"count": 888
|
| 553 |
-
},
|
| 554 |
-
"US": {
|
| 555 |
-
"accuracy": 0.9673202614379085,
|
| 556 |
-
"count": 612
|
| 557 |
-
}
|
| 558 |
-
}
|
| 559 |
-
},
|
| 560 |
-
"sub_M0": {
|
| 561 |
-
"full_accuracy": 0.988,
|
| 562 |
-
"n_examples": 250,
|
| 563 |
-
"per_subtask": {
|
| 564 |
-
"MD": {
|
| 565 |
-
"accuracy": 0.998671096345515,
|
| 566 |
-
"count": 1505
|
| 567 |
-
},
|
| 568 |
-
"ME": {
|
| 569 |
-
"accuracy": 0.9959183673469387,
|
| 570 |
-
"count": 245
|
| 571 |
-
}
|
| 572 |
-
}
|
| 573 |
-
},
|
| 574 |
-
"sub_M1": {
|
| 575 |
-
"full_accuracy": 0.932,
|
| 576 |
-
"n_examples": 250,
|
| 577 |
-
"per_subtask": {
|
| 578 |
-
"MD": {
|
| 579 |
-
"accuracy": 0.9929971988795518,
|
| 580 |
-
"count": 714
|
| 581 |
-
},
|
| 582 |
-
"MB": {
|
| 583 |
-
"accuracy": 0.9786096256684492,
|
| 584 |
-
"count": 374
|
| 585 |
-
},
|
| 586 |
-
"ME": {
|
| 587 |
-
"accuracy": 0.9866666666666667,
|
| 588 |
-
"count": 75
|
| 589 |
-
},
|
| 590 |
-
"UB": {
|
| 591 |
-
"accuracy": 0.9948892674616695,
|
| 592 |
-
"count": 587
|
| 593 |
-
}
|
| 594 |
-
}
|
| 595 |
-
},
|
| 596 |
-
"sub_M2": {
|
| 597 |
-
"full_accuracy": 0.896,
|
| 598 |
-
"n_examples": 250,
|
| 599 |
-
"per_subtask": {
|
| 600 |
-
"MD": {
|
| 601 |
-
"accuracy": 0.9909747292418772,
|
| 602 |
-
"count": 554
|
| 603 |
-
},
|
| 604 |
-
"MB": {
|
| 605 |
-
"accuracy": 0.9743589743589743,
|
| 606 |
-
"count": 273
|
| 607 |
-
},
|
| 608 |
-
"ME": {
|
| 609 |
-
"accuracy": 0.9908675799086758,
|
| 610 |
-
"count": 219
|
| 611 |
-
},
|
| 612 |
-
"UB": {
|
| 613 |
-
"accuracy": 0.9674418604651163,
|
| 614 |
-
"count": 430
|
| 615 |
-
},
|
| 616 |
-
"UD": {
|
| 617 |
-
"accuracy": 0.9963503649635036,
|
| 618 |
-
"count": 274
|
| 619 |
-
}
|
| 620 |
-
}
|
| 621 |
-
},
|
| 622 |
-
"sub_M3": {
|
| 623 |
-
"full_accuracy": 0.272,
|
| 624 |
-
"n_examples": 250,
|
| 625 |
-
"per_subtask": {
|
| 626 |
-
"MD": {
|
| 627 |
-
"accuracy": 0.9978165938864629,
|
| 628 |
-
"count": 458
|
| 629 |
-
},
|
| 630 |
-
"MB": {
|
| 631 |
-
"accuracy": 0.9923371647509579,
|
| 632 |
-
"count": 261
|
| 633 |
-
},
|
| 634 |
-
"ME": {
|
| 635 |
-
"accuracy": 1.0,
|
| 636 |
-
"count": 124
|
| 637 |
-
},
|
| 638 |
-
"UB": {
|
| 639 |
-
"accuracy": 0.5438144329896907,
|
| 640 |
-
"count": 388
|
| 641 |
-
},
|
| 642 |
-
"UD": {
|
| 643 |
-
"accuracy": 0.9730250481695568,
|
| 644 |
-
"count": 519
|
| 645 |
-
}
|
| 646 |
-
}
|
| 647 |
-
},
|
| 648 |
-
"sub_M4": {
|
| 649 |
-
"full_accuracy": 0.06,
|
| 650 |
-
"n_examples": 250,
|
| 651 |
-
"per_subtask": {
|
| 652 |
-
"MD": {
|
| 653 |
-
"accuracy": 1.0,
|
| 654 |
-
"count": 500
|
| 655 |
-
},
|
| 656 |
-
"MB": {
|
| 657 |
-
"accuracy": 0.98,
|
| 658 |
-
"count": 250
|
| 659 |
-
},
|
| 660 |
-
"UB": {
|
| 661 |
-
"accuracy": 0.192,
|
| 662 |
-
"count": 250
|
| 663 |
-
},
|
| 664 |
-
"UD": {
|
| 665 |
-
"accuracy": 0.6426666666666667,
|
| 666 |
-
"count": 750
|
| 667 |
-
}
|
| 668 |
-
}
|
| 669 |
-
},
|
| 670 |
-
"sub_M5": {
|
| 671 |
-
"full_accuracy": 0.016,
|
| 672 |
-
"n_examples": 250,
|
| 673 |
-
"per_subtask": {
|
| 674 |
-
"MD": {
|
| 675 |
-
"accuracy": 1.0,
|
| 676 |
-
"count": 250
|
| 677 |
-
},
|
| 678 |
-
"MB": {
|
| 679 |
-
"accuracy": 0.984,
|
| 680 |
-
"count": 250
|
| 681 |
-
},
|
| 682 |
-
"UB": {
|
| 683 |
-
"accuracy": 0.096,
|
| 684 |
-
"count": 250
|
| 685 |
-
},
|
| 686 |
-
"UD": {
|
| 687 |
-
"accuracy": 0.476,
|
| 688 |
-
"count": 1000
|
| 689 |
-
}
|
| 690 |
-
}
|
| 691 |
-
},
|
| 692 |
-
"sub_random": {
|
| 693 |
-
"full_accuracy": 0.935,
|
| 694 |
-
"n_examples": 200,
|
| 695 |
-
"per_subtask": {
|
| 696 |
-
"MD": {
|
| 697 |
-
"accuracy": 0.993103448275862,
|
| 698 |
-
"count": 580
|
| 699 |
-
},
|
| 700 |
-
"MB": {
|
| 701 |
-
"accuracy": 0.9887640449438202,
|
| 702 |
-
"count": 267
|
| 703 |
-
},
|
| 704 |
-
"ME": {
|
| 705 |
-
"accuracy": 1.0,
|
| 706 |
-
"count": 63
|
| 707 |
-
},
|
| 708 |
-
"UB": {
|
| 709 |
-
"accuracy": 0.984375,
|
| 710 |
-
"count": 448
|
| 711 |
-
},
|
| 712 |
-
"UD": {
|
| 713 |
-
"accuracy": 1.0,
|
| 714 |
-
"count": 42
|
| 715 |
-
}
|
| 716 |
-
}
|
| 717 |
-
},
|
| 718 |
-
"sub_B3": {
|
| 719 |
-
"full_accuracy": 0.724,
|
| 720 |
-
"n_examples": 250,
|
| 721 |
-
"per_subtask": {
|
| 722 |
-
"MD": {
|
| 723 |
-
"accuracy": 0.9986666666666667,
|
| 724 |
-
"count": 750
|
| 725 |
-
},
|
| 726 |
-
"MB": {
|
| 727 |
-
"accuracy": 0.996,
|
| 728 |
-
"count": 250
|
| 729 |
-
},
|
| 730 |
-
"UB": {
|
| 731 |
-
"accuracy": 0.8700787401574803,
|
| 732 |
-
"count": 508
|
| 733 |
-
},
|
| 734 |
-
"UD": {
|
| 735 |
-
"accuracy": 0.9628099173553719,
|
| 736 |
-
"count": 242
|
| 737 |
-
}
|
| 738 |
-
}
|
| 739 |
-
},
|
| 740 |
-
"sub_B4": {
|
| 741 |
-
"full_accuracy": 0.56,
|
| 742 |
-
"n_examples": 250,
|
| 743 |
-
"per_subtask": {
|
| 744 |
-
"MD": {
|
| 745 |
-
"accuracy": 0.988,
|
| 746 |
-
"count": 500
|
| 747 |
-
},
|
| 748 |
-
"MB": {
|
| 749 |
-
"accuracy": 0.976,
|
| 750 |
-
"count": 250
|
| 751 |
-
},
|
| 752 |
-
"UB": {
|
| 753 |
-
"accuracy": 0.8257328990228013,
|
| 754 |
-
"count": 614
|
| 755 |
-
},
|
| 756 |
-
"UD": {
|
| 757 |
-
"accuracy": 0.8860103626943006,
|
| 758 |
-
"count": 386
|
| 759 |
-
}
|
| 760 |
-
}
|
| 761 |
-
},
|
| 762 |
-
"sub_B5": {
|
| 763 |
-
"full_accuracy": 0.52,
|
| 764 |
-
"n_examples": 250,
|
| 765 |
-
"per_subtask": {
|
| 766 |
-
"MD": {
|
| 767 |
-
"accuracy": 1.0,
|
| 768 |
-
"count": 250
|
| 769 |
-
},
|
| 770 |
-
"MB": {
|
| 771 |
-
"accuracy": 0.992,
|
| 772 |
-
"count": 250
|
| 773 |
-
},
|
| 774 |
-
"UB": {
|
| 775 |
-
"accuracy": 0.8597168597168597,
|
| 776 |
-
"count": 777
|
| 777 |
-
},
|
| 778 |
-
"UD": {
|
| 779 |
-
"accuracy": 0.8752642706131079,
|
| 780 |
-
"count": 473
|
| 781 |
-
}
|
| 782 |
-
}
|
| 783 |
-
}
|
| 784 |
-
},
|
| 785 |
-
"summary": {
|
| 786 |
-
"overall_accuracy": 0.6812962962962963,
|
| 787 |
-
"total_examples": 5400,
|
| 788 |
-
"n_splits": 22
|
| 789 |
-
}
|
| 790 |
-
},
|
| 791 |
-
"sorl_eval": {
|
| 792 |
-
"config": {
|
| 793 |
-
"ops": "add_sub",
|
| 794 |
-
"K": 1,
|
| 795 |
-
"mode": "sorl",
|
| 796 |
-
"n_digits": 6,
|
| 797 |
-
"n_per_split": 250
|
| 798 |
-
},
|
| 799 |
-
"splits": {
|
| 800 |
-
"add_S0": {
|
| 801 |
-
"full_accuracy": 0.0,
|
| 802 |
-
"n_examples": 250,
|
| 803 |
-
"per_subtask": {
|
| 804 |
-
"SA": {
|
| 805 |
-
"accuracy": 0.2593565331582403,
|
| 806 |
-
"count": 1523
|
| 807 |
-
},
|
| 808 |
-
"SS": {
|
| 809 |
-
"accuracy": 0.23348017621145375,
|
| 810 |
-
"count": 227
|
| 811 |
-
}
|
| 812 |
-
}
|
| 813 |
-
},
|
| 814 |
-
"add_S1": {
|
| 815 |
-
"full_accuracy": 0.0,
|
| 816 |
-
"n_examples": 250,
|
| 817 |
-
"per_subtask": {
|
| 818 |
-
"SA": {
|
| 819 |
-
"accuracy": 0.31365313653136534,
|
| 820 |
-
"count": 542
|
| 821 |
-
},
|
| 822 |
-
"SC": {
|
| 823 |
-
"accuracy": 0.09112709832134293,
|
| 824 |
-
"count": 417
|
| 825 |
-
},
|
| 826 |
-
"SS": {
|
| 827 |
-
"accuracy": 0.12857142857142856,
|
| 828 |
-
"count": 70
|
| 829 |
-
},
|
| 830 |
-
"UC": {
|
| 831 |
-
"accuracy": 0.2565880721220527,
|
| 832 |
-
"count": 721
|
| 833 |
-
}
|
| 834 |
-
}
|
| 835 |
-
},
|
| 836 |
-
"add_S2": {
|
| 837 |
-
"full_accuracy": 0.0,
|
| 838 |
-
"n_examples": 250,
|
| 839 |
-
"per_subtask": {
|
| 840 |
-
"SA": {
|
| 841 |
-
"accuracy": 0.29891304347826086,
|
| 842 |
-
"count": 368
|
| 843 |
-
},
|
| 844 |
-
"SC": {
|
| 845 |
-
"accuracy": 0.13707165109034267,
|
| 846 |
-
"count": 321
|
| 847 |
-
},
|
| 848 |
-
"SS": {
|
| 849 |
-
"accuracy": 0.20614035087719298,
|
| 850 |
-
"count": 228
|
| 851 |
-
},
|
| 852 |
-
"UC": {
|
| 853 |
-
"accuracy": 0.3465160075329567,
|
| 854 |
-
"count": 531
|
| 855 |
-
},
|
| 856 |
-
"US": {
|
| 857 |
-
"accuracy": 0.0695364238410596,
|
| 858 |
-
"count": 302
|
| 859 |
-
}
|
| 860 |
-
}
|
| 861 |
-
},
|
| 862 |
-
"add_S3": {
|
| 863 |
-
"full_accuracy": 0.0,
|
| 864 |
-
"n_examples": 250,
|
| 865 |
-
"per_subtask": {
|
| 866 |
-
"SA": {
|
| 867 |
-
"accuracy": 0.2931596091205212,
|
| 868 |
-
"count": 307
|
| 869 |
-
},
|
| 870 |
-
"SC": {
|
| 871 |
-
"accuracy": 0.12027491408934708,
|
| 872 |
-
"count": 291
|
| 873 |
-
},
|
| 874 |
-
"SS": {
|
| 875 |
-
"accuracy": 0.3185840707964602,
|
| 876 |
-
"count": 113
|
| 877 |
-
},
|
| 878 |
-
"UC": {
|
| 879 |
-
"accuracy": 0.40993788819875776,
|
| 880 |
-
"count": 483
|
| 881 |
-
},
|
| 882 |
-
"US": {
|
| 883 |
-
"accuracy": 0.04316546762589928,
|
| 884 |
-
"count": 556
|
| 885 |
-
}
|
| 886 |
-
}
|
| 887 |
-
},
|
| 888 |
-
"add_S4": {
|
| 889 |
-
"full_accuracy": 0.0,
|
| 890 |
-
"n_examples": 250,
|
| 891 |
-
"per_subtask": {
|
| 892 |
-
"SA": {
|
| 893 |
-
"accuracy": 0.2857142857142857,
|
| 894 |
-
"count": 238
|
| 895 |
-
},
|
| 896 |
-
"SC": {
|
| 897 |
-
"accuracy": 0.08487084870848709,
|
| 898 |
-
"count": 271
|
| 899 |
-
},
|
| 900 |
-
"SS": {
|
| 901 |
-
"accuracy": 0.4576271186440678,
|
| 902 |
-
"count": 59
|
| 903 |
-
},
|
| 904 |
-
"UC": {
|
| 905 |
-
"accuracy": 0.5308641975308642,
|
| 906 |
-
"count": 405
|
| 907 |
-
},
|
| 908 |
-
"US": {
|
| 909 |
-
"accuracy": 0.04504504504504504,
|
| 910 |
-
"count": 777
|
| 911 |
-
}
|
| 912 |
-
}
|
| 913 |
-
},
|
| 914 |
-
"add_S5": {
|
| 915 |
-
"full_accuracy": 0.0,
|
| 916 |
-
"n_examples": 250,
|
| 917 |
-
"per_subtask": {
|
| 918 |
-
"SA": {
|
| 919 |
-
"accuracy": 0.588,
|
| 920 |
-
"count": 250
|
| 921 |
-
},
|
| 922 |
-
"SC": {
|
| 923 |
-
"accuracy": 0.084,
|
| 924 |
-
"count": 250
|
| 925 |
-
},
|
| 926 |
-
"UC": {
|
| 927 |
-
"accuracy": 0.488,
|
| 928 |
-
"count": 250
|
| 929 |
-
},
|
| 930 |
-
"US": {
|
| 931 |
-
"accuracy": 0.059,
|
| 932 |
-
"count": 1000
|
| 933 |
-
}
|
| 934 |
-
}
|
| 935 |
-
},
|
| 936 |
-
"add_S6": {
|
| 937 |
-
"full_accuracy": 0.0,
|
| 938 |
-
"n_examples": 250,
|
| 939 |
-
"per_subtask": {
|
| 940 |
-
"SC": {
|
| 941 |
-
"accuracy": 0.08,
|
| 942 |
-
"count": 250
|
| 943 |
-
},
|
| 944 |
-
"UC": {
|
| 945 |
-
"accuracy": 0.992,
|
| 946 |
-
"count": 250
|
| 947 |
-
},
|
| 948 |
-
"US": {
|
| 949 |
-
"accuracy": 0.1616,
|
| 950 |
-
"count": 1250
|
| 951 |
-
}
|
| 952 |
-
}
|
| 953 |
-
},
|
| 954 |
-
"add_random": {
|
| 955 |
-
"full_accuracy": 0.0,
|
| 956 |
-
"n_examples": 200,
|
| 957 |
-
"per_subtask": {
|
| 958 |
-
"SA": {
|
| 959 |
-
"accuracy": 0.2925170068027211,
|
| 960 |
-
"count": 441
|
| 961 |
-
},
|
| 962 |
-
"SC": {
|
| 963 |
-
"accuracy": 0.11356466876971609,
|
| 964 |
-
"count": 317
|
| 965 |
-
},
|
| 966 |
-
"SS": {
|
| 967 |
-
"accuracy": 0.2962962962962963,
|
| 968 |
-
"count": 54
|
| 969 |
-
},
|
| 970 |
-
"UC": {
|
| 971 |
-
"accuracy": 0.2725563909774436,
|
| 972 |
-
"count": 532
|
| 973 |
-
},
|
| 974 |
-
"US": {
|
| 975 |
-
"accuracy": 0.07142857142857142,
|
| 976 |
-
"count": 56
|
| 977 |
-
}
|
| 978 |
-
}
|
| 979 |
-
},
|
| 980 |
-
"add_C3": {
|
| 981 |
-
"full_accuracy": 0.0,
|
| 982 |
-
"n_examples": 250,
|
| 983 |
-
"per_subtask": {
|
| 984 |
-
"SA": {
|
| 985 |
-
"accuracy": 0.33866666666666667,
|
| 986 |
-
"count": 750
|
| 987 |
-
},
|
| 988 |
-
"SC": {
|
| 989 |
-
"accuracy": 0.044,
|
| 990 |
-
"count": 250
|
| 991 |
-
},
|
| 992 |
-
"UC": {
|
| 993 |
-
"accuracy": 0.17016806722689076,
|
| 994 |
-
"count": 476
|
| 995 |
-
},
|
| 996 |
-
"US": {
|
| 997 |
-
"accuracy": 0.0036496350364963502,
|
| 998 |
-
"count": 274
|
| 999 |
-
}
|
| 1000 |
-
}
|
| 1001 |
-
},
|
| 1002 |
-
"add_C4": {
|
| 1003 |
-
"full_accuracy": 0.0,
|
| 1004 |
-
"n_examples": 250,
|
| 1005 |
-
"per_subtask": {
|
| 1006 |
-
"SA": {
|
| 1007 |
-
"accuracy": 0.412,
|
| 1008 |
-
"count": 500
|
| 1009 |
-
},
|
| 1010 |
-
"SC": {
|
| 1011 |
-
"accuracy": 0.056,
|
| 1012 |
-
"count": 250
|
| 1013 |
-
},
|
| 1014 |
-
"UC": {
|
| 1015 |
-
"accuracy": 0.21331316187594554,
|
| 1016 |
-
"count": 661
|
| 1017 |
-
},
|
| 1018 |
-
"US": {
|
| 1019 |
-
"accuracy": 0.0,
|
| 1020 |
-
"count": 339
|
| 1021 |
-
}
|
| 1022 |
-
}
|
| 1023 |
-
},
|
| 1024 |
-
"add_C5": {
|
| 1025 |
-
"full_accuracy": 0.0,
|
| 1026 |
-
"n_examples": 250,
|
| 1027 |
-
"per_subtask": {
|
| 1028 |
-
"SA": {
|
| 1029 |
-
"accuracy": 0.62,
|
| 1030 |
-
"count": 250
|
| 1031 |
-
},
|
| 1032 |
-
"SC": {
|
| 1033 |
-
"accuracy": 0.092,
|
| 1034 |
-
"count": 250
|
| 1035 |
-
},
|
| 1036 |
-
"UC": {
|
| 1037 |
-
"accuracy": 0.22814207650273224,
|
| 1038 |
-
"count": 732
|
| 1039 |
-
},
|
| 1040 |
-
"US": {
|
| 1041 |
-
"accuracy": 0.0694980694980695,
|
| 1042 |
-
"count": 518
|
| 1043 |
-
}
|
| 1044 |
-
}
|
| 1045 |
-
},
|
| 1046 |
-
"add_C6": {
|
| 1047 |
-
"full_accuracy": 0.0,
|
| 1048 |
-
"n_examples": 250,
|
| 1049 |
-
"per_subtask": {
|
| 1050 |
-
"SC": {
|
| 1051 |
-
"accuracy": 0.1,
|
| 1052 |
-
"count": 250
|
| 1053 |
-
},
|
| 1054 |
-
"UC": {
|
| 1055 |
-
"accuracy": 0.3536036036036036,
|
| 1056 |
-
"count": 888
|
| 1057 |
-
},
|
| 1058 |
-
"US": {
|
| 1059 |
-
"accuracy": 0.1323529411764706,
|
| 1060 |
-
"count": 612
|
| 1061 |
-
}
|
| 1062 |
-
}
|
| 1063 |
-
},
|
| 1064 |
-
"sub_M0": {
|
| 1065 |
-
"full_accuracy": 0.0,
|
| 1066 |
-
"n_examples": 250,
|
| 1067 |
-
"per_subtask": {
|
| 1068 |
-
"MD": {
|
| 1069 |
-
"accuracy": 0.3282392026578073,
|
| 1070 |
-
"count": 1505
|
| 1071 |
-
},
|
| 1072 |
-
"ME": {
|
| 1073 |
-
"accuracy": 0.5918367346938775,
|
| 1074 |
-
"count": 245
|
| 1075 |
-
}
|
| 1076 |
-
}
|
| 1077 |
-
},
|
| 1078 |
-
"sub_M1": {
|
| 1079 |
-
"full_accuracy": 0.0,
|
| 1080 |
-
"n_examples": 250,
|
| 1081 |
-
"per_subtask": {
|
| 1082 |
-
"MD": {
|
| 1083 |
-
"accuracy": 0.4887955182072829,
|
| 1084 |
-
"count": 714
|
| 1085 |
-
},
|
| 1086 |
-
"MB": {
|
| 1087 |
-
"accuracy": 0.26737967914438504,
|
| 1088 |
-
"count": 374
|
| 1089 |
-
},
|
| 1090 |
-
"ME": {
|
| 1091 |
-
"accuracy": 0.52,
|
| 1092 |
-
"count": 75
|
| 1093 |
-
},
|
| 1094 |
-
"UB": {
|
| 1095 |
-
"accuracy": 0.1839863713798978,
|
| 1096 |
-
"count": 587
|
| 1097 |
-
}
|
| 1098 |
-
}
|
| 1099 |
-
},
|
| 1100 |
-
"sub_M2": {
|
| 1101 |
-
"full_accuracy": 0.0,
|
| 1102 |
-
"n_examples": 250,
|
| 1103 |
-
"per_subtask": {
|
| 1104 |
-
"MD": {
|
| 1105 |
-
"accuracy": 0.5379061371841155,
|
| 1106 |
-
"count": 554
|
| 1107 |
-
},
|
| 1108 |
-
"MB": {
|
| 1109 |
-
"accuracy": 0.18681318681318682,
|
| 1110 |
-
"count": 273
|
| 1111 |
-
},
|
| 1112 |
-
"ME": {
|
| 1113 |
-
"accuracy": 0.410958904109589,
|
| 1114 |
-
"count": 219
|
| 1115 |
-
},
|
| 1116 |
-
"UB": {
|
| 1117 |
-
"accuracy": 0.19534883720930232,
|
| 1118 |
-
"count": 430
|
| 1119 |
-
},
|
| 1120 |
-
"UD": {
|
| 1121 |
-
"accuracy": 0.4124087591240876,
|
| 1122 |
-
"count": 274
|
| 1123 |
-
}
|
| 1124 |
-
}
|
| 1125 |
-
},
|
| 1126 |
-
"sub_M3": {
|
| 1127 |
-
"full_accuracy": 0.0,
|
| 1128 |
-
"n_examples": 250,
|
| 1129 |
-
"per_subtask": {
|
| 1130 |
-
"MD": {
|
| 1131 |
-
"accuracy": 0.611353711790393,
|
| 1132 |
-
"count": 458
|
| 1133 |
-
},
|
| 1134 |
-
"MB": {
|
| 1135 |
-
"accuracy": 0.19157088122605365,
|
| 1136 |
-
"count": 261
|
| 1137 |
-
},
|
| 1138 |
-
"ME": {
|
| 1139 |
-
"accuracy": 0.41935483870967744,
|
| 1140 |
-
"count": 124
|
| 1141 |
-
},
|
| 1142 |
-
"UB": {
|
| 1143 |
-
"accuracy": 0.18814432989690721,
|
| 1144 |
-
"count": 388
|
| 1145 |
-
},
|
| 1146 |
-
"UD": {
|
| 1147 |
-
"accuracy": 0.23121387283236994,
|
| 1148 |
-
"count": 519
|
| 1149 |
-
}
|
| 1150 |
-
}
|
| 1151 |
-
},
|
| 1152 |
-
"sub_M4": {
|
| 1153 |
-
"full_accuracy": 0.0,
|
| 1154 |
-
"n_examples": 250,
|
| 1155 |
-
"per_subtask": {
|
| 1156 |
-
"MD": {
|
| 1157 |
-
"accuracy": 0.602,
|
| 1158 |
-
"count": 500
|
| 1159 |
-
},
|
| 1160 |
-
"MB": {
|
| 1161 |
-
"accuracy": 0.16,
|
| 1162 |
-
"count": 250
|
| 1163 |
-
},
|
| 1164 |
-
"UB": {
|
| 1165 |
-
"accuracy": 0.06,
|
| 1166 |
-
"count": 250
|
| 1167 |
-
},
|
| 1168 |
-
"UD": {
|
| 1169 |
-
"accuracy": 0.13466666666666666,
|
| 1170 |
-
"count": 750
|
| 1171 |
-
}
|
| 1172 |
-
}
|
| 1173 |
-
},
|
| 1174 |
-
"sub_M5": {
|
| 1175 |
-
"full_accuracy": 0.0,
|
| 1176 |
-
"n_examples": 250,
|
| 1177 |
-
"per_subtask": {
|
| 1178 |
-
"MD": {
|
| 1179 |
-
"accuracy": 1.0,
|
| 1180 |
-
"count": 250
|
| 1181 |
-
},
|
| 1182 |
-
"MB": {
|
| 1183 |
-
"accuracy": 0.072,
|
| 1184 |
-
"count": 250
|
| 1185 |
-
},
|
| 1186 |
-
"UB": {
|
| 1187 |
-
"accuracy": 0.084,
|
| 1188 |
-
"count": 250
|
| 1189 |
-
},
|
| 1190 |
-
"UD": {
|
| 1191 |
-
"accuracy": 0.413,
|
| 1192 |
-
"count": 1000
|
| 1193 |
-
}
|
| 1194 |
-
}
|
| 1195 |
-
},
|
| 1196 |
-
"sub_random": {
|
| 1197 |
-
"full_accuracy": 0.0,
|
| 1198 |
-
"n_examples": 200,
|
| 1199 |
-
"per_subtask": {
|
| 1200 |
-
"MD": {
|
| 1201 |
-
"accuracy": 0.47586206896551725,
|
| 1202 |
-
"count": 580
|
| 1203 |
-
},
|
| 1204 |
-
"MB": {
|
| 1205 |
-
"accuracy": 0.20599250936329588,
|
| 1206 |
-
"count": 267
|
| 1207 |
-
},
|
| 1208 |
-
"ME": {
|
| 1209 |
-
"accuracy": 0.5714285714285714,
|
| 1210 |
-
"count": 63
|
| 1211 |
-
},
|
| 1212 |
-
"UB": {
|
| 1213 |
-
"accuracy": 0.18080357142857142,
|
| 1214 |
-
"count": 448
|
| 1215 |
-
},
|
| 1216 |
-
"UD": {
|
| 1217 |
-
"accuracy": 0.30952380952380953,
|
| 1218 |
-
"count": 42
|
| 1219 |
-
}
|
| 1220 |
-
}
|
| 1221 |
-
},
|
| 1222 |
-
"sub_B3": {
|
| 1223 |
-
"full_accuracy": 0.0,
|
| 1224 |
-
"n_examples": 250,
|
| 1225 |
-
"per_subtask": {
|
| 1226 |
-
"MD": {
|
| 1227 |
-
"accuracy": 0.44,
|
| 1228 |
-
"count": 750
|
| 1229 |
-
},
|
| 1230 |
-
"MB": {
|
| 1231 |
-
"accuracy": 0.212,
|
| 1232 |
-
"count": 250
|
| 1233 |
-
},
|
| 1234 |
-
"UB": {
|
| 1235 |
-
"accuracy": 0.20078740157480315,
|
| 1236 |
-
"count": 508
|
| 1237 |
-
},
|
| 1238 |
-
"UD": {
|
| 1239 |
-
"accuracy": 0.16942148760330578,
|
| 1240 |
-
"count": 242
|
| 1241 |
-
}
|
| 1242 |
-
}
|
| 1243 |
-
},
|
| 1244 |
-
"sub_B4": {
|
| 1245 |
-
"full_accuracy": 0.0,
|
| 1246 |
-
"n_examples": 250,
|
| 1247 |
-
"per_subtask": {
|
| 1248 |
-
"MD": {
|
| 1249 |
-
"accuracy": 0.622,
|
| 1250 |
-
"count": 500
|
| 1251 |
-
},
|
| 1252 |
-
"MB": {
|
| 1253 |
-
"accuracy": 0.184,
|
| 1254 |
-
"count": 250
|
| 1255 |
-
},
|
| 1256 |
-
"UB": {
|
| 1257 |
-
"accuracy": 0.14006514657980457,
|
| 1258 |
-
"count": 614
|
| 1259 |
-
},
|
| 1260 |
-
"UD": {
|
| 1261 |
-
"accuracy": 0.19948186528497408,
|
| 1262 |
-
"count": 386
|
| 1263 |
-
}
|
| 1264 |
-
}
|
| 1265 |
-
},
|
| 1266 |
-
"sub_B5": {
|
| 1267 |
-
"full_accuracy": 0.0,
|
| 1268 |
-
"n_examples": 250,
|
| 1269 |
-
"per_subtask": {
|
| 1270 |
-
"MD": {
|
| 1271 |
-
"accuracy": 1.0,
|
| 1272 |
-
"count": 250
|
| 1273 |
-
},
|
| 1274 |
-
"MB": {
|
| 1275 |
-
"accuracy": 0.068,
|
| 1276 |
-
"count": 250
|
| 1277 |
-
},
|
| 1278 |
-
"UB": {
|
| 1279 |
-
"accuracy": 0.17631917631917632,
|
| 1280 |
-
"count": 777
|
| 1281 |
-
},
|
| 1282 |
-
"UD": {
|
| 1283 |
-
"accuracy": 0.3678646934460888,
|
| 1284 |
-
"count": 473
|
| 1285 |
-
}
|
| 1286 |
-
}
|
| 1287 |
-
}
|
| 1288 |
-
},
|
| 1289 |
-
"summary": {
|
| 1290 |
-
"overall_accuracy": 0.0,
|
| 1291 |
-
"total_examples": 5400,
|
| 1292 |
-
"n_splits": 22
|
| 1293 |
-
}
|
| 1294 |
-
},
|
| 1295 |
-
"sorl_overall_accuracy": 0.0,
|
| 1296 |
-
"sft_overall_accuracy": 0.6812962962962963
|
| 1297 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
add_sub_sorl_v1_abs10_K1_10K/model.safetensors
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:fdc469755c92ef1653fb3f6ae13eac568959c2ae6791d4bfece6a727903c7bc8
|
| 3 |
-
size 650303660
|
|
|
|
|
|
|
|
|
|
|
|
add_sub_sorl_v1_abs10_K1_10K/train_config.json
DELETED
|
@@ -1,35 +0,0 @@
|
|
| 1 |
-
{
|
| 2 |
-
"mode": "sorl",
|
| 3 |
-
"ops": "add_sub",
|
| 4 |
-
"n_digits": 6,
|
| 5 |
-
"n_layer": 2,
|
| 6 |
-
"n_head": 3,
|
| 7 |
-
"n_embd": 510,
|
| 8 |
-
"abs_vocab": 10,
|
| 9 |
-
"K": 1,
|
| 10 |
-
"alpha_info_gain": 10.0,
|
| 11 |
-
"alpha_abs": 0.1,
|
| 12 |
-
"alpha_soft_zipf": 1.0,
|
| 13 |
-
"batch_size": 64,
|
| 14 |
-
"num_epochs": 10,
|
| 15 |
-
"dataset_size": 10000,
|
| 16 |
-
"lr": 8e-05,
|
| 17 |
-
"output_dir": "ckpt/sweep/as_sorl_abs10_K1_10K",
|
| 18 |
-
"device": "cuda",
|
| 19 |
-
"push_to_hub": true,
|
| 20 |
-
"no_wandb": false,
|
| 21 |
-
"n_params": 162499262,
|
| 22 |
-
"run_name": "add_sub_sorl_v1_abs10_K1_10K",
|
| 23 |
-
"git_commit": "800625019270114adcda289bbd550c4f1109a514",
|
| 24 |
-
"timestamp": "2026-04-11T21:33:55.579231+00:00",
|
| 25 |
-
"tokenizer": "Qwen/Qwen3-0.6B",
|
| 26 |
-
"dataset_repo": "thoughtworks/arithmetic-sorl-data",
|
| 27 |
-
"dataset_config": "add_sub_6digit",
|
| 28 |
-
"model_repo": "thoughtworks/arithmetic-sorl",
|
| 29 |
-
"trainer_version": "v1",
|
| 30 |
-
"wandb_run_id": "ulvxvuq3",
|
| 31 |
-
"wandb_url": "https://wandb.ai/nlp_and_interpretability/sorl-arithmetic/runs/ulvxvuq3",
|
| 32 |
-
"final_accuracy": 0.0,
|
| 33 |
-
"sft_accuracy": 0.6812962962962963,
|
| 34 |
-
"eval_method": "ArithmeticEvaluator"
|
| 35 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|