Upload folder using huggingface_hub
Browse files- m04/benchmark_results.json +540 -0
- m04/config.json +49 -0
- m04/eval_results.json +212 -0
- m04/model.safetensors +3 -0
m04/benchmark_results.json
ADDED
|
@@ -0,0 +1,540 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_id": "m04",
|
| 3 |
+
"checkpoint": "results/m04/final.pt",
|
| 4 |
+
"trained_step": 15258,
|
| 5 |
+
"num_fewshot": 0,
|
| 6 |
+
"limit": null,
|
| 7 |
+
"results": {
|
| 8 |
+
"lambada_openai": {
|
| 9 |
+
"name": "lambada_openai",
|
| 10 |
+
"alias": "lambada_openai",
|
| 11 |
+
"sample_len": 5153,
|
| 12 |
+
"perplexity,none": 887.1387056167104,
|
| 13 |
+
"perplexity_stderr,none": 48.06586665128127,
|
| 14 |
+
"acc,none": 0.15796623326217737,
|
| 15 |
+
"acc_stderr,none": 0.005081114222421306
|
| 16 |
+
},
|
| 17 |
+
"hellaswag": {
|
| 18 |
+
"name": "hellaswag",
|
| 19 |
+
"alias": "hellaswag",
|
| 20 |
+
"sample_len": 10042,
|
| 21 |
+
"acc,none": 0.2724556861183031,
|
| 22 |
+
"acc_stderr,none": 0.0044431316326797,
|
| 23 |
+
"acc_norm,none": 0.27165903206532566,
|
| 24 |
+
"acc_norm_stderr,none": 0.004439059440526545
|
| 25 |
+
},
|
| 26 |
+
"piqa": {
|
| 27 |
+
"name": "piqa",
|
| 28 |
+
"alias": "piqa",
|
| 29 |
+
"sample_len": 1838,
|
| 30 |
+
"acc,none": 0.588683351468988,
|
| 31 |
+
"acc_stderr,none": 0.01148086057719274,
|
| 32 |
+
"acc_norm,none": 0.5908596300326442,
|
| 33 |
+
"acc_norm_stderr,none": 0.011471593460443175
|
| 34 |
+
},
|
| 35 |
+
"sciq": {
|
| 36 |
+
"name": "sciq",
|
| 37 |
+
"alias": "sciq",
|
| 38 |
+
"sample_len": 1000,
|
| 39 |
+
"acc,none": 0.685,
|
| 40 |
+
"acc_stderr,none": 0.014696631960792617,
|
| 41 |
+
"acc_norm,none": 0.605,
|
| 42 |
+
"acc_norm_stderr,none": 0.015466551464829328
|
| 43 |
+
},
|
| 44 |
+
"arc_easy": {
|
| 45 |
+
"name": "arc_easy",
|
| 46 |
+
"alias": "arc_easy",
|
| 47 |
+
"sample_len": 2376,
|
| 48 |
+
"acc,none": 0.4494949494949495,
|
| 49 |
+
"acc_stderr,none": 0.010207308833916014,
|
| 50 |
+
"acc_norm,none": 0.4036195286195286,
|
| 51 |
+
"acc_norm_stderr,none": 0.010067368960348095
|
| 52 |
+
},
|
| 53 |
+
"winogrande": {
|
| 54 |
+
"name": "winogrande",
|
| 55 |
+
"alias": "winogrande",
|
| 56 |
+
"sample_len": 1267,
|
| 57 |
+
"acc,none": 0.5201262825572218,
|
| 58 |
+
"acc_stderr,none": 0.014041096664344348
|
| 59 |
+
},
|
| 60 |
+
"blimp_adjunct_island": {
|
| 61 |
+
"name": "blimp_adjunct_island",
|
| 62 |
+
"alias": "blimp_adjunct_island",
|
| 63 |
+
"sample_len": 1000,
|
| 64 |
+
"acc,none": 0.774,
|
| 65 |
+
"acc_stderr,none": 0.013232501619085452
|
| 66 |
+
},
|
| 67 |
+
"blimp_anaphor_gender_agreement": {
|
| 68 |
+
"name": "blimp_anaphor_gender_agreement",
|
| 69 |
+
"alias": "blimp_anaphor_gender_agreement",
|
| 70 |
+
"sample_len": 1000,
|
| 71 |
+
"acc,none": 0.902,
|
| 72 |
+
"acc_stderr,none": 0.009406619184621247
|
| 73 |
+
},
|
| 74 |
+
"blimp_anaphor_number_agreement": {
|
| 75 |
+
"name": "blimp_anaphor_number_agreement",
|
| 76 |
+
"alias": "blimp_anaphor_number_agreement",
|
| 77 |
+
"sample_len": 1000,
|
| 78 |
+
"acc,none": 0.971,
|
| 79 |
+
"acc_stderr,none": 0.005309160685756948
|
| 80 |
+
},
|
| 81 |
+
"blimp_animate_subject_passive": {
|
| 82 |
+
"name": "blimp_animate_subject_passive",
|
| 83 |
+
"alias": "blimp_animate_subject_passive",
|
| 84 |
+
"sample_len": 1000,
|
| 85 |
+
"acc,none": 0.696,
|
| 86 |
+
"acc_stderr,none": 0.014553205687950535
|
| 87 |
+
},
|
| 88 |
+
"blimp_animate_subject_trans": {
|
| 89 |
+
"name": "blimp_animate_subject_trans",
|
| 90 |
+
"alias": "blimp_animate_subject_trans",
|
| 91 |
+
"sample_len": 1000,
|
| 92 |
+
"acc,none": 0.897,
|
| 93 |
+
"acc_stderr,none": 0.009616833339695739
|
| 94 |
+
},
|
| 95 |
+
"blimp_causative": {
|
| 96 |
+
"name": "blimp_causative",
|
| 97 |
+
"alias": "blimp_causative",
|
| 98 |
+
"sample_len": 1000,
|
| 99 |
+
"acc,none": 0.694,
|
| 100 |
+
"acc_stderr,none": 0.014580006055436898
|
| 101 |
+
},
|
| 102 |
+
"blimp_complex_NP_island": {
|
| 103 |
+
"name": "blimp_complex_NP_island",
|
| 104 |
+
"alias": "blimp_complex_NP_island",
|
| 105 |
+
"sample_len": 1000,
|
| 106 |
+
"acc,none": 0.606,
|
| 107 |
+
"acc_stderr,none": 0.015459721957493382
|
| 108 |
+
},
|
| 109 |
+
"blimp_coordinate_structure_constraint_complex_left_branch": {
|
| 110 |
+
"name": "blimp_coordinate_structure_constraint_complex_left_branch",
|
| 111 |
+
"alias": "blimp_coordinate_structure_constraint_complex_left_branch",
|
| 112 |
+
"sample_len": 1000,
|
| 113 |
+
"acc,none": 0.413,
|
| 114 |
+
"acc_stderr,none": 0.015577986829936455
|
| 115 |
+
},
|
| 116 |
+
"blimp_coordinate_structure_constraint_object_extraction": {
|
| 117 |
+
"name": "blimp_coordinate_structure_constraint_object_extraction",
|
| 118 |
+
"alias": "blimp_coordinate_structure_constraint_object_extraction",
|
| 119 |
+
"sample_len": 1000,
|
| 120 |
+
"acc,none": 0.82,
|
| 121 |
+
"acc_stderr,none": 0.012155153135512022
|
| 122 |
+
},
|
| 123 |
+
"blimp_determiner_noun_agreement_1": {
|
| 124 |
+
"name": "blimp_determiner_noun_agreement_1",
|
| 125 |
+
"alias": "blimp_determiner_noun_agreement_1",
|
| 126 |
+
"sample_len": 1000,
|
| 127 |
+
"acc,none": 0.961,
|
| 128 |
+
"acc_stderr,none": 0.006125072776426131
|
| 129 |
+
},
|
| 130 |
+
"blimp_determiner_noun_agreement_2": {
|
| 131 |
+
"name": "blimp_determiner_noun_agreement_2",
|
| 132 |
+
"alias": "blimp_determiner_noun_agreement_2",
|
| 133 |
+
"sample_len": 1000,
|
| 134 |
+
"acc,none": 0.955,
|
| 135 |
+
"acc_stderr,none": 0.006558812241406063
|
| 136 |
+
},
|
| 137 |
+
"blimp_determiner_noun_agreement_irregular_1": {
|
| 138 |
+
"name": "blimp_determiner_noun_agreement_irregular_1",
|
| 139 |
+
"alias": "blimp_determiner_noun_agreement_irregular_1",
|
| 140 |
+
"sample_len": 1000,
|
| 141 |
+
"acc,none": 0.859,
|
| 142 |
+
"acc_stderr,none": 0.01101091459599248
|
| 143 |
+
},
|
| 144 |
+
"blimp_determiner_noun_agreement_irregular_2": {
|
| 145 |
+
"name": "blimp_determiner_noun_agreement_irregular_2",
|
| 146 |
+
"alias": "blimp_determiner_noun_agreement_irregular_2",
|
| 147 |
+
"sample_len": 1000,
|
| 148 |
+
"acc,none": 0.939,
|
| 149 |
+
"acc_stderr,none": 0.007572076091557476
|
| 150 |
+
},
|
| 151 |
+
"blimp_determiner_noun_agreement_with_adj_2": {
|
| 152 |
+
"name": "blimp_determiner_noun_agreement_with_adj_2",
|
| 153 |
+
"alias": "blimp_determiner_noun_agreement_with_adj_2",
|
| 154 |
+
"sample_len": 1000,
|
| 155 |
+
"acc,none": 0.918,
|
| 156 |
+
"acc_stderr,none": 0.008680515615523732
|
| 157 |
+
},
|
| 158 |
+
"blimp_determiner_noun_agreement_with_adj_irregular_1": {
|
| 159 |
+
"name": "blimp_determiner_noun_agreement_with_adj_irregular_1",
|
| 160 |
+
"alias": "blimp_determiner_noun_agreement_with_adj_irregular_1",
|
| 161 |
+
"sample_len": 1000,
|
| 162 |
+
"acc,none": 0.763,
|
| 163 |
+
"acc_stderr,none": 0.013454070462577935
|
| 164 |
+
},
|
| 165 |
+
"blimp_determiner_noun_agreement_with_adj_irregular_2": {
|
| 166 |
+
"name": "blimp_determiner_noun_agreement_with_adj_irregular_2",
|
| 167 |
+
"alias": "blimp_determiner_noun_agreement_with_adj_irregular_2",
|
| 168 |
+
"sample_len": 1000,
|
| 169 |
+
"acc,none": 0.878,
|
| 170 |
+
"acc_stderr,none": 0.010354864712936776
|
| 171 |
+
},
|
| 172 |
+
"blimp_determiner_noun_agreement_with_adjective_1": {
|
| 173 |
+
"name": "blimp_determiner_noun_agreement_with_adjective_1",
|
| 174 |
+
"alias": "blimp_determiner_noun_agreement_with_adjective_1",
|
| 175 |
+
"sample_len": 1000,
|
| 176 |
+
"acc,none": 0.905,
|
| 177 |
+
"acc_stderr,none": 0.009276910103103353
|
| 178 |
+
},
|
| 179 |
+
"blimp_distractor_agreement_relational_noun": {
|
| 180 |
+
"name": "blimp_distractor_agreement_relational_noun",
|
| 181 |
+
"alias": "blimp_distractor_agreement_relational_noun",
|
| 182 |
+
"sample_len": 1000,
|
| 183 |
+
"acc,none": 0.847,
|
| 184 |
+
"acc_stderr,none": 0.011389500459665504
|
| 185 |
+
},
|
| 186 |
+
"blimp_distractor_agreement_relative_clause": {
|
| 187 |
+
"name": "blimp_distractor_agreement_relative_clause",
|
| 188 |
+
"alias": "blimp_distractor_agreement_relative_clause",
|
| 189 |
+
"sample_len": 1000,
|
| 190 |
+
"acc,none": 0.726,
|
| 191 |
+
"acc_stderr,none": 0.01411109928825969
|
| 192 |
+
},
|
| 193 |
+
"blimp_drop_argument": {
|
| 194 |
+
"name": "blimp_drop_argument",
|
| 195 |
+
"alias": "blimp_drop_argument",
|
| 196 |
+
"sample_len": 1000,
|
| 197 |
+
"acc,none": 0.722,
|
| 198 |
+
"acc_stderr,none": 0.014174516461485272
|
| 199 |
+
},
|
| 200 |
+
"blimp_ellipsis_n_bar_1": {
|
| 201 |
+
"name": "blimp_ellipsis_n_bar_1",
|
| 202 |
+
"alias": "blimp_ellipsis_n_bar_1",
|
| 203 |
+
"sample_len": 1000,
|
| 204 |
+
"acc,none": 0.775,
|
| 205 |
+
"acc_stderr,none": 0.013211720158614833
|
| 206 |
+
},
|
| 207 |
+
"blimp_ellipsis_n_bar_2": {
|
| 208 |
+
"name": "blimp_ellipsis_n_bar_2",
|
| 209 |
+
"alias": "blimp_ellipsis_n_bar_2",
|
| 210 |
+
"sample_len": 1000,
|
| 211 |
+
"acc,none": 0.825,
|
| 212 |
+
"acc_stderr,none": 0.012021627157731998
|
| 213 |
+
},
|
| 214 |
+
"blimp_existential_there_object_raising": {
|
| 215 |
+
"name": "blimp_existential_there_object_raising",
|
| 216 |
+
"alias": "blimp_existential_there_object_raising",
|
| 217 |
+
"sample_len": 1000,
|
| 218 |
+
"acc,none": 0.742,
|
| 219 |
+
"acc_stderr,none": 0.013842963108656601
|
| 220 |
+
},
|
| 221 |
+
"blimp_existential_there_quantifiers_1": {
|
| 222 |
+
"name": "blimp_existential_there_quantifiers_1",
|
| 223 |
+
"alias": "blimp_existential_there_quantifiers_1",
|
| 224 |
+
"sample_len": 1000,
|
| 225 |
+
"acc,none": 0.965,
|
| 226 |
+
"acc_stderr,none": 0.005814534272734903
|
| 227 |
+
},
|
| 228 |
+
"blimp_existential_there_quantifiers_2": {
|
| 229 |
+
"name": "blimp_existential_there_quantifiers_2",
|
| 230 |
+
"alias": "blimp_existential_there_quantifiers_2",
|
| 231 |
+
"sample_len": 1000,
|
| 232 |
+
"acc,none": 0.187,
|
| 233 |
+
"acc_stderr,none": 0.012336254828074166
|
| 234 |
+
},
|
| 235 |
+
"blimp_existential_there_subject_raising": {
|
| 236 |
+
"name": "blimp_existential_there_subject_raising",
|
| 237 |
+
"alias": "blimp_existential_there_subject_raising",
|
| 238 |
+
"sample_len": 1000,
|
| 239 |
+
"acc,none": 0.869,
|
| 240 |
+
"acc_stderr,none": 0.0106748748448379
|
| 241 |
+
},
|
| 242 |
+
"blimp_expletive_it_object_raising": {
|
| 243 |
+
"name": "blimp_expletive_it_object_raising",
|
| 244 |
+
"alias": "blimp_expletive_it_object_raising",
|
| 245 |
+
"sample_len": 1000,
|
| 246 |
+
"acc,none": 0.727,
|
| 247 |
+
"acc_stderr,none": 0.014095022868717513
|
| 248 |
+
},
|
| 249 |
+
"blimp_inchoative": {
|
| 250 |
+
"name": "blimp_inchoative",
|
| 251 |
+
"alias": "blimp_inchoative",
|
| 252 |
+
"sample_len": 1000,
|
| 253 |
+
"acc,none": 0.654,
|
| 254 |
+
"acc_stderr,none": 0.01505026612756434
|
| 255 |
+
},
|
| 256 |
+
"blimp_intransitive": {
|
| 257 |
+
"name": "blimp_intransitive",
|
| 258 |
+
"alias": "blimp_intransitive",
|
| 259 |
+
"sample_len": 1000,
|
| 260 |
+
"acc,none": 0.736,
|
| 261 |
+
"acc_stderr,none": 0.01394627184944057
|
| 262 |
+
},
|
| 263 |
+
"blimp_irregular_past_participle_adjectives": {
|
| 264 |
+
"name": "blimp_irregular_past_participle_adjectives",
|
| 265 |
+
"alias": "blimp_irregular_past_participle_adjectives",
|
| 266 |
+
"sample_len": 1000,
|
| 267 |
+
"acc,none": 0.856,
|
| 268 |
+
"acc_stderr,none": 0.01110798754893916
|
| 269 |
+
},
|
| 270 |
+
"blimp_irregular_past_participle_verbs": {
|
| 271 |
+
"name": "blimp_irregular_past_participle_verbs",
|
| 272 |
+
"alias": "blimp_irregular_past_participle_verbs",
|
| 273 |
+
"sample_len": 1000,
|
| 274 |
+
"acc,none": 0.895,
|
| 275 |
+
"acc_stderr,none": 0.00969892102602496
|
| 276 |
+
},
|
| 277 |
+
"blimp_irregular_plural_subject_verb_agreement_1": {
|
| 278 |
+
"name": "blimp_irregular_plural_subject_verb_agreement_1",
|
| 279 |
+
"alias": "blimp_irregular_plural_subject_verb_agreement_1",
|
| 280 |
+
"sample_len": 1000,
|
| 281 |
+
"acc,none": 0.867,
|
| 282 |
+
"acc_stderr,none": 0.010743669132397295
|
| 283 |
+
},
|
| 284 |
+
"blimp_irregular_plural_subject_verb_agreement_2": {
|
| 285 |
+
"name": "blimp_irregular_plural_subject_verb_agreement_2",
|
| 286 |
+
"alias": "blimp_irregular_plural_subject_verb_agreement_2",
|
| 287 |
+
"sample_len": 1000,
|
| 288 |
+
"acc,none": 0.875,
|
| 289 |
+
"acc_stderr,none": 0.010463483381956722
|
| 290 |
+
},
|
| 291 |
+
"blimp_left_branch_island_echo_question": {
|
| 292 |
+
"name": "blimp_left_branch_island_echo_question",
|
| 293 |
+
"alias": "blimp_left_branch_island_echo_question",
|
| 294 |
+
"sample_len": 1000,
|
| 295 |
+
"acc,none": 0.588,
|
| 296 |
+
"acc_stderr,none": 0.015572363292015072
|
| 297 |
+
},
|
| 298 |
+
"blimp_left_branch_island_simple_question": {
|
| 299 |
+
"name": "blimp_left_branch_island_simple_question",
|
| 300 |
+
"alias": "blimp_left_branch_island_simple_question",
|
| 301 |
+
"sample_len": 1000,
|
| 302 |
+
"acc,none": 0.535,
|
| 303 |
+
"acc_stderr,none": 0.015780495050030086
|
| 304 |
+
},
|
| 305 |
+
"blimp_matrix_question_npi_licensor_present": {
|
| 306 |
+
"name": "blimp_matrix_question_npi_licensor_present",
|
| 307 |
+
"alias": "blimp_matrix_question_npi_licensor_present",
|
| 308 |
+
"sample_len": 1000,
|
| 309 |
+
"acc,none": 0.491,
|
| 310 |
+
"acc_stderr,none": 0.0158167369950053
|
| 311 |
+
},
|
| 312 |
+
"blimp_npi_present_1": {
|
| 313 |
+
"name": "blimp_npi_present_1",
|
| 314 |
+
"alias": "blimp_npi_present_1",
|
| 315 |
+
"sample_len": 1000,
|
| 316 |
+
"acc,none": 0.59,
|
| 317 |
+
"acc_stderr,none": 0.015560917136921659
|
| 318 |
+
},
|
| 319 |
+
"blimp_npi_present_2": {
|
| 320 |
+
"name": "blimp_npi_present_2",
|
| 321 |
+
"alias": "blimp_npi_present_2",
|
| 322 |
+
"sample_len": 1000,
|
| 323 |
+
"acc,none": 0.66,
|
| 324 |
+
"acc_stderr,none": 0.014987482264363972
|
| 325 |
+
},
|
| 326 |
+
"blimp_only_npi_licensor_present": {
|
| 327 |
+
"name": "blimp_only_npi_licensor_present",
|
| 328 |
+
"alias": "blimp_only_npi_licensor_present",
|
| 329 |
+
"sample_len": 1000,
|
| 330 |
+
"acc,none": 0.988,
|
| 331 |
+
"acc_stderr,none": 0.00344497719409982
|
| 332 |
+
},
|
| 333 |
+
"blimp_only_npi_scope": {
|
| 334 |
+
"name": "blimp_only_npi_scope",
|
| 335 |
+
"alias": "blimp_only_npi_scope",
|
| 336 |
+
"sample_len": 1000,
|
| 337 |
+
"acc,none": 0.78,
|
| 338 |
+
"acc_stderr,none": 0.013106173040661862
|
| 339 |
+
},
|
| 340 |
+
"blimp_passive_1": {
|
| 341 |
+
"name": "blimp_passive_1",
|
| 342 |
+
"alias": "blimp_passive_1",
|
| 343 |
+
"sample_len": 1000,
|
| 344 |
+
"acc,none": 0.873,
|
| 345 |
+
"acc_stderr,none": 0.010534798620855644
|
| 346 |
+
},
|
| 347 |
+
"blimp_passive_2": {
|
| 348 |
+
"name": "blimp_passive_2",
|
| 349 |
+
"alias": "blimp_passive_2",
|
| 350 |
+
"sample_len": 1000,
|
| 351 |
+
"acc,none": 0.872,
|
| 352 |
+
"acc_stderr,none": 0.010570133761108595
|
| 353 |
+
},
|
| 354 |
+
"blimp_principle_A_c_command": {
|
| 355 |
+
"name": "blimp_principle_A_c_command",
|
| 356 |
+
"alias": "blimp_principle_A_c_command",
|
| 357 |
+
"sample_len": 1000,
|
| 358 |
+
"acc,none": 0.557,
|
| 359 |
+
"acc_stderr,none": 0.015716169953204184
|
| 360 |
+
},
|
| 361 |
+
"blimp_principle_A_case_1": {
|
| 362 |
+
"name": "blimp_principle_A_case_1",
|
| 363 |
+
"alias": "blimp_principle_A_case_1",
|
| 364 |
+
"sample_len": 1000,
|
| 365 |
+
"acc,none": 1.0,
|
| 366 |
+
"acc_stderr,none": 0.0
|
| 367 |
+
},
|
| 368 |
+
"blimp_principle_A_case_2": {
|
| 369 |
+
"name": "blimp_principle_A_case_2",
|
| 370 |
+
"alias": "blimp_principle_A_case_2",
|
| 371 |
+
"sample_len": 1000,
|
| 372 |
+
"acc,none": 0.867,
|
| 373 |
+
"acc_stderr,none": 0.010743669132397295
|
| 374 |
+
},
|
| 375 |
+
"blimp_principle_A_domain_1": {
|
| 376 |
+
"name": "blimp_principle_A_domain_1",
|
| 377 |
+
"alias": "blimp_principle_A_domain_1",
|
| 378 |
+
"sample_len": 1000,
|
| 379 |
+
"acc,none": 0.936,
|
| 380 |
+
"acc_stderr,none": 0.007743640226919376
|
| 381 |
+
},
|
| 382 |
+
"blimp_principle_A_domain_2": {
|
| 383 |
+
"name": "blimp_principle_A_domain_2",
|
| 384 |
+
"alias": "blimp_principle_A_domain_2",
|
| 385 |
+
"sample_len": 1000,
|
| 386 |
+
"acc,none": 0.72,
|
| 387 |
+
"acc_stderr,none": 0.014205696104091548
|
| 388 |
+
},
|
| 389 |
+
"blimp_principle_A_domain_3": {
|
| 390 |
+
"name": "blimp_principle_A_domain_3",
|
| 391 |
+
"alias": "blimp_principle_A_domain_3",
|
| 392 |
+
"sample_len": 1000,
|
| 393 |
+
"acc,none": 0.616,
|
| 394 |
+
"acc_stderr,none": 0.015387682761896958
|
| 395 |
+
},
|
| 396 |
+
"blimp_principle_A_reconstruction": {
|
| 397 |
+
"name": "blimp_principle_A_reconstruction",
|
| 398 |
+
"alias": "blimp_principle_A_reconstruction",
|
| 399 |
+
"sample_len": 1000,
|
| 400 |
+
"acc,none": 0.489,
|
| 401 |
+
"acc_stderr,none": 0.01581547119529257
|
| 402 |
+
},
|
| 403 |
+
"blimp_regular_plural_subject_verb_agreement_1": {
|
| 404 |
+
"name": "blimp_regular_plural_subject_verb_agreement_1",
|
| 405 |
+
"alias": "blimp_regular_plural_subject_verb_agreement_1",
|
| 406 |
+
"sample_len": 1000,
|
| 407 |
+
"acc,none": 0.9,
|
| 408 |
+
"acc_stderr,none": 0.00949157995752499
|
| 409 |
+
},
|
| 410 |
+
"blimp_regular_plural_subject_verb_agreement_2": {
|
| 411 |
+
"name": "blimp_regular_plural_subject_verb_agreement_2",
|
| 412 |
+
"alias": "blimp_regular_plural_subject_verb_agreement_2",
|
| 413 |
+
"sample_len": 1000,
|
| 414 |
+
"acc,none": 0.888,
|
| 415 |
+
"acc_stderr,none": 0.009977753031397177
|
| 416 |
+
},
|
| 417 |
+
"blimp_sentential_negation_npi_licensor_present": {
|
| 418 |
+
"name": "blimp_sentential_negation_npi_licensor_present",
|
| 419 |
+
"alias": "blimp_sentential_negation_npi_licensor_present",
|
| 420 |
+
"sample_len": 1000,
|
| 421 |
+
"acc,none": 0.999,
|
| 422 |
+
"acc_stderr,none": 0.001
|
| 423 |
+
},
|
| 424 |
+
"blimp_sentential_negation_npi_scope": {
|
| 425 |
+
"name": "blimp_sentential_negation_npi_scope",
|
| 426 |
+
"alias": "blimp_sentential_negation_npi_scope",
|
| 427 |
+
"sample_len": 1000,
|
| 428 |
+
"acc,none": 0.345,
|
| 429 |
+
"acc_stderr,none": 0.015039986742055365
|
| 430 |
+
},
|
| 431 |
+
"blimp_sentential_subject_island": {
|
| 432 |
+
"name": "blimp_sentential_subject_island",
|
| 433 |
+
"alias": "blimp_sentential_subject_island",
|
| 434 |
+
"sample_len": 1000,
|
| 435 |
+
"acc,none": 0.404,
|
| 436 |
+
"acc_stderr,none": 0.015524980677122437
|
| 437 |
+
},
|
| 438 |
+
"blimp_superlative_quantifiers_1": {
|
| 439 |
+
"name": "blimp_superlative_quantifiers_1",
|
| 440 |
+
"alias": "blimp_superlative_quantifiers_1",
|
| 441 |
+
"sample_len": 1000,
|
| 442 |
+
"acc,none": 0.805,
|
| 443 |
+
"acc_stderr,none": 0.012535235623319256
|
| 444 |
+
},
|
| 445 |
+
"blimp_superlative_quantifiers_2": {
|
| 446 |
+
"name": "blimp_superlative_quantifiers_2",
|
| 447 |
+
"alias": "blimp_superlative_quantifiers_2",
|
| 448 |
+
"sample_len": 1000,
|
| 449 |
+
"acc,none": 0.841,
|
| 450 |
+
"acc_stderr,none": 0.011569479368271195
|
| 451 |
+
},
|
| 452 |
+
"blimp_tough_vs_raising_1": {
|
| 453 |
+
"name": "blimp_tough_vs_raising_1",
|
| 454 |
+
"alias": "blimp_tough_vs_raising_1",
|
| 455 |
+
"sample_len": 1000,
|
| 456 |
+
"acc,none": 0.515,
|
| 457 |
+
"acc_stderr,none": 0.01581217964181488
|
| 458 |
+
},
|
| 459 |
+
"blimp_tough_vs_raising_2": {
|
| 460 |
+
"name": "blimp_tough_vs_raising_2",
|
| 461 |
+
"alias": "blimp_tough_vs_raising_2",
|
| 462 |
+
"sample_len": 1000,
|
| 463 |
+
"acc,none": 0.895,
|
| 464 |
+
"acc_stderr,none": 0.00969892102602496
|
| 465 |
+
},
|
| 466 |
+
"blimp_transitive": {
|
| 467 |
+
"name": "blimp_transitive",
|
| 468 |
+
"alias": "blimp_transitive",
|
| 469 |
+
"sample_len": 1000,
|
| 470 |
+
"acc,none": 0.8,
|
| 471 |
+
"acc_stderr,none": 0.012655439943366653
|
| 472 |
+
},
|
| 473 |
+
"blimp_wh_island": {
|
| 474 |
+
"name": "blimp_wh_island",
|
| 475 |
+
"alias": "blimp_wh_island",
|
| 476 |
+
"sample_len": 1000,
|
| 477 |
+
"acc,none": 0.786,
|
| 478 |
+
"acc_stderr,none": 0.012975838021968847
|
| 479 |
+
},
|
| 480 |
+
"blimp_wh_questions_object_gap": {
|
| 481 |
+
"name": "blimp_wh_questions_object_gap",
|
| 482 |
+
"alias": "blimp_wh_questions_object_gap",
|
| 483 |
+
"sample_len": 1000,
|
| 484 |
+
"acc,none": 0.669,
|
| 485 |
+
"acc_stderr,none": 0.0148882725882038
|
| 486 |
+
},
|
| 487 |
+
"blimp_wh_questions_subject_gap": {
|
| 488 |
+
"name": "blimp_wh_questions_subject_gap",
|
| 489 |
+
"alias": "blimp_wh_questions_subject_gap",
|
| 490 |
+
"sample_len": 1000,
|
| 491 |
+
"acc,none": 0.869,
|
| 492 |
+
"acc_stderr,none": 0.0106748748448379
|
| 493 |
+
},
|
| 494 |
+
"blimp_wh_questions_subject_gap_long_distance": {
|
| 495 |
+
"name": "blimp_wh_questions_subject_gap_long_distance",
|
| 496 |
+
"alias": "blimp_wh_questions_subject_gap_long_distance",
|
| 497 |
+
"sample_len": 1000,
|
| 498 |
+
"acc,none": 0.868,
|
| 499 |
+
"acc_stderr,none": 0.010709373963527965
|
| 500 |
+
},
|
| 501 |
+
"blimp_wh_vs_that_no_gap": {
|
| 502 |
+
"name": "blimp_wh_vs_that_no_gap",
|
| 503 |
+
"alias": "blimp_wh_vs_that_no_gap",
|
| 504 |
+
"sample_len": 1000,
|
| 505 |
+
"acc,none": 0.944,
|
| 506 |
+
"acc_stderr,none": 0.007274401481697097
|
| 507 |
+
},
|
| 508 |
+
"blimp_wh_vs_that_no_gap_long_distance": {
|
| 509 |
+
"name": "blimp_wh_vs_that_no_gap_long_distance",
|
| 510 |
+
"alias": "blimp_wh_vs_that_no_gap_long_distance",
|
| 511 |
+
"sample_len": 1000,
|
| 512 |
+
"acc,none": 0.975,
|
| 513 |
+
"acc_stderr,none": 0.004939574819698493
|
| 514 |
+
},
|
| 515 |
+
"blimp_wh_vs_that_with_gap": {
|
| 516 |
+
"name": "blimp_wh_vs_that_with_gap",
|
| 517 |
+
"alias": "blimp_wh_vs_that_with_gap",
|
| 518 |
+
"sample_len": 1000,
|
| 519 |
+
"acc,none": 0.572,
|
| 520 |
+
"acc_stderr,none": 0.015654426245029267
|
| 521 |
+
},
|
| 522 |
+
"blimp_wh_vs_that_with_gap_long_distance": {
|
| 523 |
+
"name": "blimp_wh_vs_that_with_gap_long_distance",
|
| 524 |
+
"alias": "blimp_wh_vs_that_with_gap_long_distance",
|
| 525 |
+
"sample_len": 1000,
|
| 526 |
+
"acc,none": 0.176,
|
| 527 |
+
"acc_stderr,none": 0.012048616898597498
|
| 528 |
+
},
|
| 529 |
+
"blimp": {
|
| 530 |
+
"alias": "blimp",
|
| 531 |
+
"name": "blimp",
|
| 532 |
+
"sample_len": 67000,
|
| 533 |
+
"acc,none": 0.7616716417910447,
|
| 534 |
+
"acc_stderr,none": 0.0014786292668924085,
|
| 535 |
+
"sample_count": {
|
| 536 |
+
"acc,none": 67000
|
| 537 |
+
}
|
| 538 |
+
}
|
| 539 |
+
}
|
| 540 |
+
}
|
m04/config.json
ADDED
|
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_id": "m04",
|
| 3 |
+
"run_name": "m04-gqa-learned-dense-adamw",
|
| 4 |
+
"data_dir": "data/fineweb-edu",
|
| 5 |
+
"vocab_size": 50257,
|
| 6 |
+
"block_size": 1024,
|
| 7 |
+
"d_model": 512,
|
| 8 |
+
"num_layers": 12,
|
| 9 |
+
"num_heads": 8,
|
| 10 |
+
"ffn_mult": 4,
|
| 11 |
+
"dropout": 0.0,
|
| 12 |
+
"bias": false,
|
| 13 |
+
"norm_type": "rmsnorm",
|
| 14 |
+
"tie_embeddings": true,
|
| 15 |
+
"attention_type": "gqa",
|
| 16 |
+
"gqa_num_kv_heads": 2,
|
| 17 |
+
"mla_latent_dim": 64,
|
| 18 |
+
"pos_type": "learned",
|
| 19 |
+
"rope_base": 10000,
|
| 20 |
+
"ffn_type": "dense",
|
| 21 |
+
"moe_num_experts": 8,
|
| 22 |
+
"moe_top_k": 2,
|
| 23 |
+
"optimizer": "adamw",
|
| 24 |
+
"lr": 0.0006,
|
| 25 |
+
"min_lr": 6e-05,
|
| 26 |
+
"weight_decay": 0.1,
|
| 27 |
+
"beta1": 0.9,
|
| 28 |
+
"beta2": 0.95,
|
| 29 |
+
"grad_clip": 1.0,
|
| 30 |
+
"muon_lr": 0.02,
|
| 31 |
+
"muon_momentum": 0.95,
|
| 32 |
+
"training_tokens": 1000000000,
|
| 33 |
+
"warmup_tokens": 6000000,
|
| 34 |
+
"lr_schedule": "cosine",
|
| 35 |
+
"batch_size": 4,
|
| 36 |
+
"grad_accum_steps": 16,
|
| 37 |
+
"device": "cuda",
|
| 38 |
+
"dtype": "bfloat16",
|
| 39 |
+
"compile": true,
|
| 40 |
+
"seed": 1337,
|
| 41 |
+
"eval_interval": 250,
|
| 42 |
+
"eval_iters": 100,
|
| 43 |
+
"log_interval": 10,
|
| 44 |
+
"checkpoint_interval": 1000,
|
| 45 |
+
"results_dir": "results",
|
| 46 |
+
"mlflow_experiment": "slm-ablation",
|
| 47 |
+
"mlflow_tracking_uri": "sqlite:///mlflow.db",
|
| 48 |
+
"mlflow_artifact_location": "./mlartifacts"
|
| 49 |
+
}
|
m04/eval_results.json
ADDED
|
@@ -0,0 +1,212 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_id": "m04",
|
| 3 |
+
"checkpoint": "results/m04/final.pt",
|
| 4 |
+
"trained_step": 15258,
|
| 5 |
+
"total_params": 59298816,
|
| 6 |
+
"active_params": 59298816,
|
| 7 |
+
"flops_per_forward": 118597632.0,
|
| 8 |
+
"val_loss": 3.5466248601675034,
|
| 9 |
+
"val_perplexity": 34.696015739304045,
|
| 10 |
+
"bits_per_token": 5.116698097657171,
|
| 11 |
+
"ppl_tokens_evaluated": 655360,
|
| 12 |
+
"prefill_tok_per_sec": 187406.36219991438,
|
| 13 |
+
"inference_tok_per_sec": 291.6789108904974,
|
| 14 |
+
"inference_tok_per_sec_kvcache": 289.28245120485207,
|
| 15 |
+
"kvcache_speedup": 0.9917839117050701,
|
| 16 |
+
"gen_tokens": 128,
|
| 17 |
+
"peak_infer_vram_gb": 0.365,
|
| 18 |
+
"peak_train_vram_gb": 3.989,
|
| 19 |
+
"closeness_answer_ppl": 31.783,
|
| 20 |
+
"closeness_hit_rate": 0.45,
|
| 21 |
+
"closeness_token_f1": 0.068,
|
| 22 |
+
"closeness_n": 20,
|
| 23 |
+
"closeness_items": [
|
| 24 |
+
{
|
| 25 |
+
"prompt": "The capital of France is",
|
| 26 |
+
"answer": " Paris",
|
| 27 |
+
"generated": " the capital of the French Empire",
|
| 28 |
+
"answer_nll": 2.9767,
|
| 29 |
+
"answer_ppl": 19.624,
|
| 30 |
+
"hit": false,
|
| 31 |
+
"token_f1": 0.0
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"prompt": "The capital of Japan is",
|
| 35 |
+
"answer": " Tokyo",
|
| 36 |
+
"generated": " the capital of the United States",
|
| 37 |
+
"answer_nll": 3.5593,
|
| 38 |
+
"answer_ppl": 35.137,
|
| 39 |
+
"hit": false,
|
| 40 |
+
"token_f1": 0.0
|
| 41 |
+
},
|
| 42 |
+
{
|
| 43 |
+
"prompt": "The largest planet in the solar system is",
|
| 44 |
+
"answer": " Jupiter",
|
| 45 |
+
"generated": " the Sun.\nThe Sun",
|
| 46 |
+
"answer_nll": 4.1744,
|
| 47 |
+
"answer_ppl": 65.003,
|
| 48 |
+
"hit": false,
|
| 49 |
+
"token_f1": 0.0
|
| 50 |
+
},
|
| 51 |
+
{
|
| 52 |
+
"prompt": "Water is made of hydrogen and",
|
| 53 |
+
"answer": " oxygen",
|
| 54 |
+
"generated": " oxygen.\nThe process of",
|
| 55 |
+
"answer_nll": 0.9468,
|
| 56 |
+
"answer_ppl": 2.578,
|
| 57 |
+
"hit": true,
|
| 58 |
+
"token_f1": 0.0
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"prompt": "The opposite of hot is",
|
| 62 |
+
"answer": " cold",
|
| 63 |
+
"generated": " the cold.\nThe cold",
|
| 64 |
+
"answer_nll": 4.0504,
|
| 65 |
+
"answer_ppl": 57.418,
|
| 66 |
+
"hit": true,
|
| 67 |
+
"token_f1": 0.4
|
| 68 |
+
},
|
| 69 |
+
{
|
| 70 |
+
"prompt": "The opposite of up is",
|
| 71 |
+
"answer": " down",
|
| 72 |
+
"generated": " the same as the opposite of",
|
| 73 |
+
"answer_nll": 7.5578,
|
| 74 |
+
"answer_ppl": 1915.625,
|
| 75 |
+
"hit": false,
|
| 76 |
+
"token_f1": 0.0
|
| 77 |
+
},
|
| 78 |
+
{
|
| 79 |
+
"prompt": "Two plus two equals",
|
| 80 |
+
"answer": " four",
|
| 81 |
+
"generated": " to the sum of the two",
|
| 82 |
+
"answer_nll": 4.8203,
|
| 83 |
+
"answer_ppl": 124.004,
|
| 84 |
+
"hit": false,
|
| 85 |
+
"token_f1": 0.0
|
| 86 |
+
},
|
| 87 |
+
{
|
| 88 |
+
"prompt": "The sun rises in the",
|
| 89 |
+
"answer": " east",
|
| 90 |
+
"generated": " sky, and the sun rises",
|
| 91 |
+
"answer_nll": 4.6267,
|
| 92 |
+
"answer_ppl": 102.172,
|
| 93 |
+
"hit": false,
|
| 94 |
+
"token_f1": 0.0
|
| 95 |
+
},
|
| 96 |
+
{
|
| 97 |
+
"prompt": "The first president of the United States was George",
|
| 98 |
+
"answer": " Washington",
|
| 99 |
+
"generated": " Washington, who was the first",
|
| 100 |
+
"answer_nll": 0.4903,
|
| 101 |
+
"answer_ppl": 1.633,
|
| 102 |
+
"hit": true,
|
| 103 |
+
"token_f1": 0.0
|
| 104 |
+
},
|
| 105 |
+
{
|
| 106 |
+
"prompt": "The speed of light is approximately 300,000 kilometers per",
|
| 107 |
+
"answer": " second",
|
| 108 |
+
"generated": " second.\nThe speed of",
|
| 109 |
+
"answer_nll": 0.1589,
|
| 110 |
+
"answer_ppl": 1.172,
|
| 111 |
+
"hit": true,
|
| 112 |
+
"token_f1": 0.0
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"prompt": "A triangle has three",
|
| 116 |
+
"answer": " sides",
|
| 117 |
+
"generated": " sides, and the sides of",
|
| 118 |
+
"answer_nll": 0.8456,
|
| 119 |
+
"answer_ppl": 2.329,
|
| 120 |
+
"hit": true,
|
| 121 |
+
"token_f1": 0.333
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"prompt": "The chemical symbol for gold is",
|
| 125 |
+
"answer": " Au",
|
| 126 |
+
"generated": " the gold.\nThe gold",
|
| 127 |
+
"answer_nll": 8.1955,
|
| 128 |
+
"answer_ppl": 3624.438,
|
| 129 |
+
"hit": false,
|
| 130 |
+
"token_f1": 0.0
|
| 131 |
+
},
|
| 132 |
+
{
|
| 133 |
+
"prompt": "The freezing point of water in Celsius is",
|
| 134 |
+
"answer": " zero",
|
| 135 |
+
"generated": " the temperature of the Earth.",
|
| 136 |
+
"answer_nll": 4.5415,
|
| 137 |
+
"answer_ppl": 93.832,
|
| 138 |
+
"hit": false,
|
| 139 |
+
"token_f1": 0.0
|
| 140 |
+
},
|
| 141 |
+
{
|
| 142 |
+
"prompt": "The author of Romeo and Juliet is William",
|
| 143 |
+
"answer": " Shakespeare",
|
| 144 |
+
"generated": " Shakespeare.\nThe author of",
|
| 145 |
+
"answer_nll": 0.6901,
|
| 146 |
+
"answer_ppl": 1.994,
|
| 147 |
+
"hit": true,
|
| 148 |
+
"token_f1": 0.0
|
| 149 |
+
},
|
| 150 |
+
{
|
| 151 |
+
"prompt": "The Earth revolves around the",
|
| 152 |
+
"answer": " Sun",
|
| 153 |
+
"generated": " sun, and the Sun is",
|
| 154 |
+
"answer_nll": 1.1193,
|
| 155 |
+
"answer_ppl": 3.063,
|
| 156 |
+
"hit": true,
|
| 157 |
+
"token_f1": 0.333
|
| 158 |
+
},
|
| 159 |
+
{
|
| 160 |
+
"prompt": "The plural of mouse is",
|
| 161 |
+
"answer": " mice",
|
| 162 |
+
"generated": " a word that is used to",
|
| 163 |
+
"answer_nll": 9.0947,
|
| 164 |
+
"answer_ppl": 8907.852,
|
| 165 |
+
"hit": false,
|
| 166 |
+
"token_f1": 0.0
|
| 167 |
+
},
|
| 168 |
+
{
|
| 169 |
+
"prompt": "Photosynthesis occurs in the",
|
| 170 |
+
"answer": " leaves",
|
| 171 |
+
"generated": " plant cells.\nThe plant",
|
| 172 |
+
"answer_nll": 3.9642,
|
| 173 |
+
"answer_ppl": 52.68,
|
| 174 |
+
"hit": false,
|
| 175 |
+
"token_f1": 0.0
|
| 176 |
+
},
|
| 177 |
+
{
|
| 178 |
+
"prompt": "The human heart pumps",
|
| 179 |
+
"answer": " blood",
|
| 180 |
+
"generated": " more blood than it needs to",
|
| 181 |
+
"answer_nll": 1.912,
|
| 182 |
+
"answer_ppl": 6.767,
|
| 183 |
+
"hit": true,
|
| 184 |
+
"token_f1": 0.286
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"prompt": "The currency used in the United States is the",
|
| 188 |
+
"answer": " dollar",
|
| 189 |
+
"generated": " United States dollar.\nThe",
|
| 190 |
+
"answer_nll": 2.7842,
|
| 191 |
+
"answer_ppl": 16.187,
|
| 192 |
+
"hit": true,
|
| 193 |
+
"token_f1": 0.0
|
| 194 |
+
},
|
| 195 |
+
{
|
| 196 |
+
"prompt": "The tallest animal in the world is the",
|
| 197 |
+
"answer": " giraffe",
|
| 198 |
+
"generated": " giant elephant.\nThe elephant",
|
| 199 |
+
"answer_nll": 2.6698,
|
| 200 |
+
"answer_ppl": 14.438,
|
| 201 |
+
"hit": false,
|
| 202 |
+
"token_f1": 0.0
|
| 203 |
+
}
|
| 204 |
+
],
|
| 205 |
+
"generation_samples": {
|
| 206 |
+
"The most important thing about sleep is": "The most important thing about sleep is that it\u2019s a good idea to start off the habit by gradually increasing your activity level.\nAnd this is not a bad idea. It\u2019s actually good for you and your body, and it will help you burn off the energy you need, and more importantly, you\u2019",
|
| 207 |
+
"Scientists have discovered that": "Scientists have discovered that, in many ways, the human face is actually a kind of plastic. In the past, people have been using plastic to create a plastic box, but that\u2019s not a problem.\nWhat is one of the mysteries of the world?\nThere are many things we can ask this question",
|
| 208 |
+
"In the history of mathematics,": "In the history of mathematics, the term has been used to describe the behavior of one variable after another. The result is sometimes known as a time difference. In the world of mathematics, the term time of the time (in the context of time itself) can be used to describe what the individual is doing, or what the individual",
|
| 209 |
+
"Python code for prime number": "Python code for prime number systems, for example. However, the main reason for this is that their implementation is so poor and not enough to support the use of a large number of memory devices. As a result, it is recommended that the system be used efficiently.\nThe main reason for this is that it allows a lot",
|
| 210 |
+
"The best way to learn a new language is": "The best way to learn a new language is to learn about it. Teaching about the language of the new language is important because it allows you to learn a new language.\nThe best way to learn a new language is to use it as an alternative, in which you can change the meaning of the new language and to learn new words.\n"
|
| 211 |
+
}
|
| 212 |
+
}
|
m04/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f0295767376d96b0ddd689f345e82e7c100ffacab1b4ed14b7e6bdc07ab28715
|
| 3 |
+
size 237204864
|