Upload folder using huggingface_hub
Browse files- m06/benchmark_results.json +540 -0
- m06/config.json +49 -0
- m06/eval_results.json +212 -0
- m06/model.safetensors +3 -0
m06/benchmark_results.json
ADDED
|
@@ -0,0 +1,540 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_id": "m06",
|
| 3 |
+
"checkpoint": "results/m06/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": 545.4353787064736,
|
| 13 |
+
"perplexity_stderr,none": 29.51773277871608,
|
| 14 |
+
"acc,none": 0.18086551523384437,
|
| 15 |
+
"acc_stderr,none": 0.005362503178578457
|
| 16 |
+
},
|
| 17 |
+
"hellaswag": {
|
| 18 |
+
"name": "hellaswag",
|
| 19 |
+
"alias": "hellaswag",
|
| 20 |
+
"sample_len": 10042,
|
| 21 |
+
"acc,none": 0.27703644692292373,
|
| 22 |
+
"acc_stderr,none": 0.004466200055292599,
|
| 23 |
+
"acc_norm,none": 0.27912766381198967,
|
| 24 |
+
"acc_norm_stderr,none": 0.004476536569056845
|
| 25 |
+
},
|
| 26 |
+
"piqa": {
|
| 27 |
+
"name": "piqa",
|
| 28 |
+
"alias": "piqa",
|
| 29 |
+
"sample_len": 1838,
|
| 30 |
+
"acc,none": 0.5963003264417845,
|
| 31 |
+
"acc_stderr,none": 0.011447407541749126,
|
| 32 |
+
"acc_norm,none": 0.5832426550598476,
|
| 33 |
+
"acc_norm_stderr,none": 0.011503015163618171
|
| 34 |
+
},
|
| 35 |
+
"sciq": {
|
| 36 |
+
"name": "sciq",
|
| 37 |
+
"alias": "sciq",
|
| 38 |
+
"sample_len": 1000,
|
| 39 |
+
"acc,none": 0.71,
|
| 40 |
+
"acc_stderr,none": 0.014356395999905621,
|
| 41 |
+
"acc_norm,none": 0.614,
|
| 42 |
+
"acc_norm_stderr,none": 0.015402637476784335
|
| 43 |
+
},
|
| 44 |
+
"arc_easy": {
|
| 45 |
+
"name": "arc_easy",
|
| 46 |
+
"alias": "arc_easy",
|
| 47 |
+
"sample_len": 2376,
|
| 48 |
+
"acc,none": 0.4532828282828283,
|
| 49 |
+
"acc_stderr,none": 0.010214901516731604,
|
| 50 |
+
"acc_norm,none": 0.4057239057239057,
|
| 51 |
+
"acc_norm_stderr,none": 0.010075755540128987
|
| 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": 0.772,
|
| 65 |
+
"acc_stderr,none": 0.013273740700804434
|
| 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": 0.897,
|
| 72 |
+
"acc_stderr,none": 0.009616833339695739
|
| 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": 0.986,
|
| 79 |
+
"acc_stderr,none": 0.003717232548256541
|
| 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": 0.738,
|
| 86 |
+
"acc_stderr,none": 0.013912208651021217
|
| 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": 0.901,
|
| 93 |
+
"acc_stderr,none": 0.009449248027662746
|
| 94 |
+
},
|
| 95 |
+
"blimp_causative": {
|
| 96 |
+
"name": "blimp_causative",
|
| 97 |
+
"alias": "blimp_causative",
|
| 98 |
+
"sample_len": 1000,
|
| 99 |
+
"acc,none": 0.698,
|
| 100 |
+
"acc_stderr,none": 0.014526080235459496
|
| 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": 0.518,
|
| 107 |
+
"acc_stderr,none": 0.01580904569940659
|
| 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": 0.54,
|
| 114 |
+
"acc_stderr,none": 0.0157685969143944
|
| 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": 0.822,
|
| 121 |
+
"acc_stderr,none": 0.012102167676183571
|
| 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": 0.969,
|
| 128 |
+
"acc_stderr,none": 0.005483527064679181
|
| 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": 0.963,
|
| 135 |
+
"acc_stderr,none": 0.0059721576223896395
|
| 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": 0.877,
|
| 142 |
+
"acc_stderr,none": 0.010391293421849803
|
| 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": 0.935,
|
| 149 |
+
"acc_stderr,none": 0.007799733061832043
|
| 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": 0.915,
|
| 156 |
+
"acc_stderr,none": 0.008823426366942316
|
| 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": 0.832,
|
| 163 |
+
"acc_stderr,none": 0.011828605831454352
|
| 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": 0.906,
|
| 170 |
+
"acc_stderr,none": 0.009233052000787672
|
| 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": 0.93,
|
| 177 |
+
"acc_stderr,none": 0.008072494358323525
|
| 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": 0.89,
|
| 184 |
+
"acc_stderr,none": 0.00989939381972442
|
| 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": 0.801,
|
| 191 |
+
"acc_stderr,none": 0.01263164908309911
|
| 192 |
+
},
|
| 193 |
+
"blimp_drop_argument": {
|
| 194 |
+
"name": "blimp_drop_argument",
|
| 195 |
+
"alias": "blimp_drop_argument",
|
| 196 |
+
"sample_len": 1000,
|
| 197 |
+
"acc,none": 0.743,
|
| 198 |
+
"acc_stderr,none": 0.013825416526895026
|
| 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": 0.825,
|
| 205 |
+
"acc_stderr,none": 0.012021627157731998
|
| 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": 0.852,
|
| 212 |
+
"acc_stderr,none": 0.011234866364235145
|
| 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": 0.809,
|
| 219 |
+
"acc_stderr,none": 0.012436787112179482
|
| 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": 0.955,
|
| 226 |
+
"acc_stderr,none": 0.006558812241406063
|
| 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": 0.172,
|
| 233 |
+
"acc_stderr,none": 0.011939788882495308
|
| 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": 0.852,
|
| 240 |
+
"acc_stderr,none": 0.011234866364235145
|
| 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": 0.762,
|
| 247 |
+
"acc_stderr,none": 0.013473586661967093
|
| 248 |
+
},
|
| 249 |
+
"blimp_inchoative": {
|
| 250 |
+
"name": "blimp_inchoative",
|
| 251 |
+
"alias": "blimp_inchoative",
|
| 252 |
+
"sample_len": 1000,
|
| 253 |
+
"acc,none": 0.656,
|
| 254 |
+
"acc_stderr,none": 0.015029633724408919
|
| 255 |
+
},
|
| 256 |
+
"blimp_intransitive": {
|
| 257 |
+
"name": "blimp_intransitive",
|
| 258 |
+
"alias": "blimp_intransitive",
|
| 259 |
+
"sample_len": 1000,
|
| 260 |
+
"acc,none": 0.744,
|
| 261 |
+
"acc_stderr,none": 0.013807775152234145
|
| 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": 0.927,
|
| 268 |
+
"acc_stderr,none": 0.008230354715243976
|
| 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": 0.925,
|
| 275 |
+
"acc_stderr,none": 0.008333333333333333
|
| 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": 0.895,
|
| 282 |
+
"acc_stderr,none": 0.00969892102602496
|
| 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": 0.88,
|
| 289 |
+
"acc_stderr,none": 0.010281328012747462
|
| 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": 0.536,
|
| 296 |
+
"acc_stderr,none": 0.015778243024904673
|
| 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": 0.624,
|
| 303 |
+
"acc_stderr,none": 0.0153251055088981
|
| 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": 0.665,
|
| 310 |
+
"acc_stderr,none": 0.014933117490932629
|
| 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": 0.583,
|
| 317 |
+
"acc_stderr,none": 0.015599819048769583
|
| 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": 0.616,
|
| 324 |
+
"acc_stderr,none": 0.015387682761896958
|
| 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": 0.887,
|
| 331 |
+
"acc_stderr,none": 0.010016552866696773
|
| 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": 0.444,
|
| 338 |
+
"acc_stderr,none": 0.01571976816340213
|
| 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": 0.859,
|
| 352 |
+
"acc_stderr,none": 0.01101091459599248
|
| 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": 0.66,
|
| 359 |
+
"acc_stderr,none": 0.014987482264363972
|
| 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": 0.938,
|
| 373 |
+
"acc_stderr,none": 0.007629823996280269
|
| 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": 0.897,
|
| 380 |
+
"acc_stderr,none": 0.009616833339695739
|
| 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": 0.791,
|
| 387 |
+
"acc_stderr,none": 0.012864077288499337
|
| 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": 0.643,
|
| 394 |
+
"acc_stderr,none": 0.01515852172148659
|
| 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": 0.488,
|
| 401 |
+
"acc_stderr,none": 0.015814743314581686
|
| 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": 0.894,
|
| 408 |
+
"acc_stderr,none": 0.00973955126578524
|
| 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": 0.878,
|
| 415 |
+
"acc_stderr,none": 0.010354864712936776
|
| 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": 0.988,
|
| 422 |
+
"acc_stderr,none": 0.00344497719409982
|
| 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": 0.528,
|
| 429 |
+
"acc_stderr,none": 0.015794475789511517
|
| 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": 0.509,
|
| 436 |
+
"acc_stderr,none": 0.0158167369950053
|
| 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": 0.608,
|
| 443 |
+
"acc_stderr,none": 0.015445859463771338
|
| 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": 0.811,
|
| 450 |
+
"acc_stderr,none": 0.01238678458811779
|
| 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": 0.605,
|
| 457 |
+
"acc_stderr,none": 0.015466551464829328
|
| 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": 0.842,
|
| 464 |
+
"acc_stderr,none": 0.011539894677559635
|
| 465 |
+
},
|
| 466 |
+
"blimp_transitive": {
|
| 467 |
+
"name": "blimp_transitive",
|
| 468 |
+
"alias": "blimp_transitive",
|
| 469 |
+
"sample_len": 1000,
|
| 470 |
+
"acc,none": 0.82,
|
| 471 |
+
"acc_stderr,none": 0.012155153135512022
|
| 472 |
+
},
|
| 473 |
+
"blimp_wh_island": {
|
| 474 |
+
"name": "blimp_wh_island",
|
| 475 |
+
"alias": "blimp_wh_island",
|
| 476 |
+
"sample_len": 1000,
|
| 477 |
+
"acc,none": 0.84,
|
| 478 |
+
"acc_stderr,none": 0.011598902298689068
|
| 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": 0.772,
|
| 485 |
+
"acc_stderr,none": 0.013273740700804434
|
| 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": 0.935,
|
| 492 |
+
"acc_stderr,none": 0.007799733061832043
|
| 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": 0.897,
|
| 499 |
+
"acc_stderr,none": 0.009616833339695739
|
| 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": 0.979,
|
| 506 |
+
"acc_stderr,none": 0.0045364721513065165
|
| 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": 0.968,
|
| 513 |
+
"acc_stderr,none": 0.005568393575081328
|
| 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": 0.628,
|
| 520 |
+
"acc_stderr,none": 0.01529214994204052
|
| 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": 0.294,
|
| 527 |
+
"acc_stderr,none": 0.014414290540008265
|
| 528 |
+
},
|
| 529 |
+
"blimp": {
|
| 530 |
+
"alias": "blimp",
|
| 531 |
+
"name": "blimp",
|
| 532 |
+
"sample_len": 67000,
|
| 533 |
+
"acc,none": 0.778910447761194,
|
| 534 |
+
"acc_stderr,none": 0.001455919645561985,
|
| 535 |
+
"sample_count": {
|
| 536 |
+
"acc,none": 67000
|
| 537 |
+
}
|
| 538 |
+
}
|
| 539 |
+
}
|
| 540 |
+
}
|
m06/config.json
ADDED
|
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_id": "m06",
|
| 3 |
+
"run_name": "m06-mha-rope-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": "mha",
|
| 16 |
+
"gqa_num_kv_heads": 2,
|
| 17 |
+
"mla_latent_dim": 64,
|
| 18 |
+
"pos_type": "rope",
|
| 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 |
+
}
|
m06/eval_results.json
ADDED
|
@@ -0,0 +1,212 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_id": "m06",
|
| 3 |
+
"checkpoint": "results/m06/final.pt",
|
| 4 |
+
"trained_step": 15258,
|
| 5 |
+
"total_params": 63493120,
|
| 6 |
+
"active_params": 63493120,
|
| 7 |
+
"flops_per_forward": 126986240.0,
|
| 8 |
+
"val_loss": 3.4427546560764313,
|
| 9 |
+
"val_perplexity": 31.272985946204606,
|
| 10 |
+
"bits_per_token": 4.966845069318857,
|
| 11 |
+
"ppl_tokens_evaluated": 655360,
|
| 12 |
+
"prefill_tok_per_sec": 153440.495550482,
|
| 13 |
+
"inference_tok_per_sec": 196.12124010573368,
|
| 14 |
+
"inference_tok_per_sec_kvcache": 190.99119877394725,
|
| 15 |
+
"kvcache_speedup": 0.9738425000320174,
|
| 16 |
+
"gen_tokens": 128,
|
| 17 |
+
"peak_infer_vram_gb": 0.395,
|
| 18 |
+
"peak_train_vram_gb": 4.097,
|
| 19 |
+
"closeness_answer_ppl": 15.475,
|
| 20 |
+
"closeness_hit_rate": 0.55,
|
| 21 |
+
"closeness_token_f1": 0.083,
|
| 22 |
+
"closeness_n": 20,
|
| 23 |
+
"closeness_items": [
|
| 24 |
+
{
|
| 25 |
+
"prompt": "The capital of France is",
|
| 26 |
+
"answer": " Paris",
|
| 27 |
+
"generated": " Paris. The capital is Paris",
|
| 28 |
+
"answer_nll": 1.351,
|
| 29 |
+
"answer_ppl": 3.861,
|
| 30 |
+
"hit": true,
|
| 31 |
+
"token_f1": 0.333
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"prompt": "The capital of Japan is",
|
| 35 |
+
"answer": " Tokyo",
|
| 36 |
+
"generated": " Tokyo. The capital is Tokyo",
|
| 37 |
+
"answer_nll": 1.5316,
|
| 38 |
+
"answer_ppl": 4.625,
|
| 39 |
+
"hit": true,
|
| 40 |
+
"token_f1": 0.333
|
| 41 |
+
},
|
| 42 |
+
{
|
| 43 |
+
"prompt": "The largest planet in the solar system is",
|
| 44 |
+
"answer": " Jupiter",
|
| 45 |
+
"generated": " the Sun, which is the",
|
| 46 |
+
"answer_nll": 3.301,
|
| 47 |
+
"answer_ppl": 27.139,
|
| 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 hydrogen is",
|
| 55 |
+
"answer_nll": 1.3272,
|
| 56 |
+
"answer_ppl": 3.77,
|
| 57 |
+
"hit": true,
|
| 58 |
+
"token_f1": 0.0
|
| 59 |
+
},
|
| 60 |
+
{
|
| 61 |
+
"prompt": "The opposite of hot is",
|
| 62 |
+
"answer": " cold",
|
| 63 |
+
"generated": " the heat of the sun.",
|
| 64 |
+
"answer_nll": 2.3498,
|
| 65 |
+
"answer_ppl": 10.483,
|
| 66 |
+
"hit": false,
|
| 67 |
+
"token_f1": 0.0
|
| 68 |
+
},
|
| 69 |
+
{
|
| 70 |
+
"prompt": "The opposite of up is",
|
| 71 |
+
"answer": " down",
|
| 72 |
+
"generated": " the \u201cdown\u201d",
|
| 73 |
+
"answer_nll": 2.0964,
|
| 74 |
+
"answer_ppl": 8.137,
|
| 75 |
+
"hit": true,
|
| 76 |
+
"token_f1": 0.0
|
| 77 |
+
},
|
| 78 |
+
{
|
| 79 |
+
"prompt": "Two plus two equals",
|
| 80 |
+
"answer": " four",
|
| 81 |
+
"generated": " the number of times the number",
|
| 82 |
+
"answer_nll": 4.6418,
|
| 83 |
+
"answer_ppl": 103.733,
|
| 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": 3.7532,
|
| 92 |
+
"answer_ppl": 42.656,
|
| 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. He was a member",
|
| 100 |
+
"answer_nll": 0.3467,
|
| 101 |
+
"answer_ppl": 1.414,
|
| 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": 0.0454,
|
| 110 |
+
"answer_ppl": 1.046,
|
| 111 |
+
"hit": true,
|
| 112 |
+
"token_f1": 0.0
|
| 113 |
+
},
|
| 114 |
+
{
|
| 115 |
+
"prompt": "A triangle has three",
|
| 116 |
+
"answer": " sides",
|
| 117 |
+
"generated": " sides, and the triangle has",
|
| 118 |
+
"answer_nll": 0.9853,
|
| 119 |
+
"answer_ppl": 2.679,
|
| 120 |
+
"hit": true,
|
| 121 |
+
"token_f1": 0.0
|
| 122 |
+
},
|
| 123 |
+
{
|
| 124 |
+
"prompt": "The chemical symbol for gold is",
|
| 125 |
+
"answer": " Au",
|
| 126 |
+
"generated": " the symbol of the sun.",
|
| 127 |
+
"answer_nll": 8.9973,
|
| 128 |
+
"answer_ppl": 8081.313,
|
| 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": " 1.5\u00b0C.",
|
| 136 |
+
"answer_nll": 6.5868,
|
| 137 |
+
"answer_ppl": 725.456,
|
| 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. He is the son",
|
| 145 |
+
"answer_nll": 0.5215,
|
| 146 |
+
"answer_ppl": 1.684,
|
| 147 |
+
"hit": true,
|
| 148 |
+
"token_f1": 0.0
|
| 149 |
+
},
|
| 150 |
+
{
|
| 151 |
+
"prompt": "The Earth revolves around the",
|
| 152 |
+
"answer": " Sun",
|
| 153 |
+
"generated": " Sun, the Sun is the",
|
| 154 |
+
"answer_nll": 0.7619,
|
| 155 |
+
"answer_ppl": 2.142,
|
| 156 |
+
"hit": true,
|
| 157 |
+
"token_f1": 0.333
|
| 158 |
+
},
|
| 159 |
+
{
|
| 160 |
+
"prompt": "The plural of mouse is",
|
| 161 |
+
"answer": " mice",
|
| 162 |
+
"generated": " the mouse. The plural of",
|
| 163 |
+
"answer_nll": 5.464,
|
| 164 |
+
"answer_ppl": 236.051,
|
| 165 |
+
"hit": false,
|
| 166 |
+
"token_f1": 0.0
|
| 167 |
+
},
|
| 168 |
+
{
|
| 169 |
+
"prompt": "Photosynthesis occurs in the",
|
| 170 |
+
"answer": " leaves",
|
| 171 |
+
"generated": " body, but it is not",
|
| 172 |
+
"answer_nll": 4.3326,
|
| 173 |
+
"answer_ppl": 76.143,
|
| 174 |
+
"hit": false,
|
| 175 |
+
"token_f1": 0.0
|
| 176 |
+
},
|
| 177 |
+
{
|
| 178 |
+
"prompt": "The human heart pumps",
|
| 179 |
+
"answer": " blood",
|
| 180 |
+
"generated": " blood to the lungs. The",
|
| 181 |
+
"answer_nll": 1.4275,
|
| 182 |
+
"answer_ppl": 4.168,
|
| 183 |
+
"hit": true,
|
| 184 |
+
"token_f1": 0.333
|
| 185 |
+
},
|
| 186 |
+
{
|
| 187 |
+
"prompt": "The currency used in the United States is the",
|
| 188 |
+
"answer": " dollar",
|
| 189 |
+
"generated": " US dollar. The US dollar",
|
| 190 |
+
"answer_nll": 2.7287,
|
| 191 |
+
"answer_ppl": 15.313,
|
| 192 |
+
"hit": true,
|
| 193 |
+
"token_f1": 0.333
|
| 194 |
+
},
|
| 195 |
+
{
|
| 196 |
+
"prompt": "The tallest animal in the world is the",
|
| 197 |
+
"answer": " giraffe",
|
| 198 |
+
"generated": " giant, the giant, the",
|
| 199 |
+
"answer_nll": 2.2349,
|
| 200 |
+
"answer_ppl": 9.346,
|
| 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 to create a routine that is consistent, consistent and meaningful. It\u2019s important to keep a record of what you\u2019re doing, what you\u2019re doing, and what you\u2019re doing.\nThe most important thing about sleep is to keep a routine that is consistent,",
|
| 207 |
+
"Scientists have discovered that": "Scientists have discovered that a large volume of water is collected in the river than the Earth\u2019s surface. Water is a \u201cgreen\u201d gas, with a pH that\u2019s higher than 1.6.\nBut why is the water so important?\nThere are many reasons why scientists think this type",
|
| 208 |
+
"In the history of mathematics,": "In the history of mathematics, the term has been used to describe the practice of using the term \u201cknowledge\u201d to describe a mathematical concept. The term \u201cknowledge\u201d is used as a synonym for knowledge, but can be used in two different ways. One way is the word \u201cknowledge\u201d,",
|
| 209 |
+
"Python code for prime number": "Python code for prime number tables, and the common name for the string 'A' to be used.\nNote that the C++ program is not designed to store large numbers of numbers, so the C++ library must store large numbers of numbers.\n- Note that the C++\n- the C++\n- the",
|
| 210 |
+
"The best way to learn a new language is": "The best way to learn a new language is to learn it in a fun and fun way. When you do some fun learning, you can start to improve your learning technique and become proficient in the new language.\n- Don\u2019t assume the new language is a language of the past. Most people are not familiar with new language, it"
|
| 211 |
+
}
|
| 212 |
+
}
|
m06/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7df4c1bebe97a3b6d55b0b03b75bc8677a943de3c8806d349d7328585c714c90
|
| 3 |
+
size 253982016
|