Upload folder using huggingface_hub
Browse files- m07/benchmark_results.json +540 -0
- m07/config.json +49 -0
- m07/eval_results.json +212 -0
- m07/model.safetensors +3 -0
m07/benchmark_results.json
ADDED
|
@@ -0,0 +1,540 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_id": "m07",
|
| 3 |
+
"checkpoint": "results/m07/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": 626.475680047097,
|
| 13 |
+
"perplexity_stderr,none": 33.86111856824197,
|
| 14 |
+
"acc,none": 0.1680574422666408,
|
| 15 |
+
"acc_stderr,none": 0.00520939894955755
|
| 16 |
+
},
|
| 17 |
+
"hellaswag": {
|
| 18 |
+
"name": "hellaswag",
|
| 19 |
+
"alias": "hellaswag",
|
| 20 |
+
"sample_len": 10042,
|
| 21 |
+
"acc,none": 0.2733519219279028,
|
| 22 |
+
"acc_stderr,none": 0.004447691405592456,
|
| 23 |
+
"acc_norm,none": 0.27853017327225654,
|
| 24 |
+
"acc_norm_stderr,none": 0.004473595650807741
|
| 25 |
+
},
|
| 26 |
+
"piqa": {
|
| 27 |
+
"name": "piqa",
|
| 28 |
+
"alias": "piqa",
|
| 29 |
+
"sample_len": 1838,
|
| 30 |
+
"acc,none": 0.5919477693144722,
|
| 31 |
+
"acc_stderr,none": 0.011466872778651143,
|
| 32 |
+
"acc_norm,none": 0.5865070729053319,
|
| 33 |
+
"acc_norm_stderr,none": 0.011489895831821244
|
| 34 |
+
},
|
| 35 |
+
"sciq": {
|
| 36 |
+
"name": "sciq",
|
| 37 |
+
"alias": "sciq",
|
| 38 |
+
"sample_len": 1000,
|
| 39 |
+
"acc,none": 0.689,
|
| 40 |
+
"acc_stderr,none": 0.01464559638572269,
|
| 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": 0.4621212121212121,
|
| 49 |
+
"acc_stderr,none": 0.01023029962886479,
|
| 50 |
+
"acc_norm,none": 0.4095117845117845,
|
| 51 |
+
"acc_norm_stderr,none": 0.010090368160989927
|
| 52 |
+
},
|
| 53 |
+
"winogrande": {
|
| 54 |
+
"name": "winogrande",
|
| 55 |
+
"alias": "winogrande",
|
| 56 |
+
"sample_len": 1267,
|
| 57 |
+
"acc,none": 0.5280189423835833,
|
| 58 |
+
"acc_stderr,none": 0.014030404213405703
|
| 59 |
+
},
|
| 60 |
+
"blimp_adjunct_island": {
|
| 61 |
+
"name": "blimp_adjunct_island",
|
| 62 |
+
"alias": "blimp_adjunct_island",
|
| 63 |
+
"sample_len": 1000,
|
| 64 |
+
"acc,none": 0.874,
|
| 65 |
+
"acc_stderr,none": 0.010499249222407964
|
| 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": 0.894,
|
| 72 |
+
"acc_stderr,none": 0.00973955126578524
|
| 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": 0.976,
|
| 79 |
+
"acc_stderr,none": 0.004842256441727084
|
| 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": 0.726,
|
| 86 |
+
"acc_stderr,none": 0.01411109928825969
|
| 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": 0.907,
|
| 93 |
+
"acc_stderr,none": 0.009188875634996669
|
| 94 |
+
},
|
| 95 |
+
"blimp_causative": {
|
| 96 |
+
"name": "blimp_causative",
|
| 97 |
+
"alias": "blimp_causative",
|
| 98 |
+
"sample_len": 1000,
|
| 99 |
+
"acc,none": 0.722,
|
| 100 |
+
"acc_stderr,none": 0.014174516461485272
|
| 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": 0.563,
|
| 107 |
+
"acc_stderr,none": 0.01569322392873043
|
| 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": 0.495,
|
| 114 |
+
"acc_stderr,none": 0.015818508944436743
|
| 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": 0.823,
|
| 121 |
+
"acc_stderr,none": 0.012075463420375087
|
| 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": 0.94,
|
| 135 |
+
"acc_stderr,none": 0.007513751157474971
|
| 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": 0.901,
|
| 142 |
+
"acc_stderr,none": 0.009449248027662746
|
| 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": 0.933,
|
| 149 |
+
"acc_stderr,none": 0.007910345983177575
|
| 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": 0.913,
|
| 156 |
+
"acc_stderr,none": 0.008916866630745944
|
| 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": 0.858,
|
| 163 |
+
"acc_stderr,none": 0.011043457699378305
|
| 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": 0.89,
|
| 170 |
+
"acc_stderr,none": 0.00989939381972442
|
| 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": 0.927,
|
| 177 |
+
"acc_stderr,none": 0.008230354715243976
|
| 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": 0.896,
|
| 184 |
+
"acc_stderr,none": 0.009658016218524242
|
| 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": 0.735,
|
| 191 |
+
"acc_stderr,none": 0.013963164754810064
|
| 192 |
+
},
|
| 193 |
+
"blimp_drop_argument": {
|
| 194 |
+
"name": "blimp_drop_argument",
|
| 195 |
+
"alias": "blimp_drop_argument",
|
| 196 |
+
"sample_len": 1000,
|
| 197 |
+
"acc,none": 0.744,
|
| 198 |
+
"acc_stderr,none": 0.013807775152234145
|
| 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": 0.75,
|
| 205 |
+
"acc_stderr,none": 0.013699915608779773
|
| 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": 0.844,
|
| 212 |
+
"acc_stderr,none": 0.011480235006122295
|
| 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": 0.733,
|
| 219 |
+
"acc_stderr,none": 0.013996674851796299
|
| 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": 0.155,
|
| 233 |
+
"acc_stderr,none": 0.011450157470799522
|
| 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": 0.883,
|
| 240 |
+
"acc_stderr,none": 0.010169287802713345
|
| 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": 0.722,
|
| 247 |
+
"acc_stderr,none": 0.014174516461485272
|
| 248 |
+
},
|
| 249 |
+
"blimp_inchoative": {
|
| 250 |
+
"name": "blimp_inchoative",
|
| 251 |
+
"alias": "blimp_inchoative",
|
| 252 |
+
"sample_len": 1000,
|
| 253 |
+
"acc,none": 0.66,
|
| 254 |
+
"acc_stderr,none": 0.014987482264363972
|
| 255 |
+
},
|
| 256 |
+
"blimp_intransitive": {
|
| 257 |
+
"name": "blimp_intransitive",
|
| 258 |
+
"alias": "blimp_intransitive",
|
| 259 |
+
"sample_len": 1000,
|
| 260 |
+
"acc,none": 0.753,
|
| 261 |
+
"acc_stderr,none": 0.013644675781314012
|
| 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": 0.94,
|
| 268 |
+
"acc_stderr,none": 0.007513751157474971
|
| 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": 0.854,
|
| 275 |
+
"acc_stderr,none": 0.011171786285496551
|
| 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": 0.906,
|
| 289 |
+
"acc_stderr,none": 0.009233052000787672
|
| 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": 0.627,
|
| 296 |
+
"acc_stderr,none": 0.015300493622922927
|
| 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": 0.636,
|
| 303 |
+
"acc_stderr,none": 0.015222868840522005
|
| 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": 0.556,
|
| 310 |
+
"acc_stderr,none": 0.015719768163402127
|
| 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": 0.514,
|
| 317 |
+
"acc_stderr,none": 0.01581309754773093
|
| 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": 0.501,
|
| 324 |
+
"acc_stderr,none": 0.015819268290576817
|
| 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": 0.839,
|
| 331 |
+
"acc_stderr,none": 0.011628164696727263
|
| 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": 0.611,
|
| 338 |
+
"acc_stderr,none": 0.015424555647308514
|
| 339 |
+
},
|
| 340 |
+
"blimp_passive_1": {
|
| 341 |
+
"name": "blimp_passive_1",
|
| 342 |
+
"alias": "blimp_passive_1",
|
| 343 |
+
"sample_len": 1000,
|
| 344 |
+
"acc,none": 0.868,
|
| 345 |
+
"acc_stderr,none": 0.010709373963527965
|
| 346 |
+
},
|
| 347 |
+
"blimp_passive_2": {
|
| 348 |
+
"name": "blimp_passive_2",
|
| 349 |
+
"alias": "blimp_passive_2",
|
| 350 |
+
"sample_len": 1000,
|
| 351 |
+
"acc,none": 0.871,
|
| 352 |
+
"acc_stderr,none": 0.010605256784796466
|
| 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": 0.639,
|
| 359 |
+
"acc_stderr,none": 0.015195720118175049
|
| 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": 0.949,
|
| 373 |
+
"acc_stderr,none": 0.0069604200625714725
|
| 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": 0.898,
|
| 380 |
+
"acc_stderr,none": 0.009575368801653944
|
| 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": 0.731,
|
| 387 |
+
"acc_stderr,none": 0.014029819522568203
|
| 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": 0.623,
|
| 394 |
+
"acc_stderr,none": 0.0153331701257798
|
| 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": 0.544,
|
| 401 |
+
"acc_stderr,none": 0.01575792855397918
|
| 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": 0.838,
|
| 408 |
+
"acc_stderr,none": 0.011657267771304384
|
| 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": 0.868,
|
| 415 |
+
"acc_stderr,none": 0.010709373963527965
|
| 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": 0.997,
|
| 422 |
+
"acc_stderr,none": 0.0017303161543469432
|
| 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": 0.463,
|
| 429 |
+
"acc_stderr,none": 0.015775927227262426
|
| 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": 0.572,
|
| 436 |
+
"acc_stderr,none": 0.015654426245029267
|
| 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": 0.903,
|
| 443 |
+
"acc_stderr,none": 0.009363689373248062
|
| 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": 0.798,
|
| 450 |
+
"acc_stderr,none": 0.012702651587655128
|
| 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": 0.599,
|
| 457 |
+
"acc_stderr,none": 0.015506109745498368
|
| 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": 0.868,
|
| 464 |
+
"acc_stderr,none": 0.010709373963527965
|
| 465 |
+
},
|
| 466 |
+
"blimp_transitive": {
|
| 467 |
+
"name": "blimp_transitive",
|
| 468 |
+
"alias": "blimp_transitive",
|
| 469 |
+
"sample_len": 1000,
|
| 470 |
+
"acc,none": 0.828,
|
| 471 |
+
"acc_stderr,none": 0.011939788882495308
|
| 472 |
+
},
|
| 473 |
+
"blimp_wh_island": {
|
| 474 |
+
"name": "blimp_wh_island",
|
| 475 |
+
"alias": "blimp_wh_island",
|
| 476 |
+
"sample_len": 1000,
|
| 477 |
+
"acc,none": 0.646,
|
| 478 |
+
"acc_stderr,none": 0.015129868238451812
|
| 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": 0.682,
|
| 485 |
+
"acc_stderr,none": 0.014734079309311903
|
| 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": 0.876,
|
| 492 |
+
"acc_stderr,none": 0.010427498872343873
|
| 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": 0.804,
|
| 499 |
+
"acc_stderr,none": 0.012559527926707347
|
| 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": 0.946,
|
| 506 |
+
"acc_stderr,none": 0.007150883521295473
|
| 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": 0.965,
|
| 513 |
+
"acc_stderr,none": 0.005814534272734903
|
| 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": 0.517,
|
| 520 |
+
"acc_stderr,none": 0.015810153729833274
|
| 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": 0.22,
|
| 527 |
+
"acc_stderr,none": 0.013106173040661863
|
| 528 |
+
},
|
| 529 |
+
"blimp": {
|
| 530 |
+
"alias": "blimp",
|
| 531 |
+
"name": "blimp",
|
| 532 |
+
"sample_len": 67000,
|
| 533 |
+
"acc,none": 0.7691492537313434,
|
| 534 |
+
"acc_stderr,none": 0.0014762777263167823,
|
| 535 |
+
"sample_count": {
|
| 536 |
+
"acc,none": 67000
|
| 537 |
+
}
|
| 538 |
+
}
|
| 539 |
+
}
|
| 540 |
+
}
|
m07/config.json
ADDED
|
@@ -0,0 +1,49 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_id": "m07",
|
| 3 |
+
"run_name": "m07-mqa-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": "mqa",
|
| 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 |
+
}
|
m07/eval_results.json
ADDED
|
@@ -0,0 +1,212 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model_id": "m07",
|
| 3 |
+
"checkpoint": "results/m07/final.pt",
|
| 4 |
+
"trained_step": 15258,
|
| 5 |
+
"total_params": 57988096,
|
| 6 |
+
"active_params": 57988096,
|
| 7 |
+
"flops_per_forward": 115976192.0,
|
| 8 |
+
"val_loss": 3.473718386888504,
|
| 9 |
+
"val_perplexity": 32.256461729868455,
|
| 10 |
+
"bits_per_token": 5.011516290208855,
|
| 11 |
+
"ppl_tokens_evaluated": 655360,
|
| 12 |
+
"prefill_tok_per_sec": 177389.79648805797,
|
| 13 |
+
"inference_tok_per_sec": 195.11035180276977,
|
| 14 |
+
"inference_tok_per_sec_kvcache": 201.57889318217914,
|
| 15 |
+
"kvcache_speedup": 1.0331532454308123,
|
| 16 |
+
"gen_tokens": 128,
|
| 17 |
+
"peak_infer_vram_gb": 0.362,
|
| 18 |
+
"peak_train_vram_gb": 3.976,
|
| 19 |
+
"closeness_answer_ppl": 27.642,
|
| 20 |
+
"closeness_hit_rate": 0.45,
|
| 21 |
+
"closeness_token_f1": 0.033,
|
| 22 |
+
"closeness_n": 20,
|
| 23 |
+
"closeness_items": [
|
| 24 |
+
{
|
| 25 |
+
"prompt": "The capital of France is",
|
| 26 |
+
"answer": " Paris",
|
| 27 |
+
"generated": " the city of Paris.\n",
|
| 28 |
+
"answer_nll": 2.3468,
|
| 29 |
+
"answer_ppl": 10.452,
|
| 30 |
+
"hit": true,
|
| 31 |
+
"token_f1": 0.0
|
| 32 |
+
},
|
| 33 |
+
{
|
| 34 |
+
"prompt": "The capital of Japan is",
|
| 35 |
+
"answer": " Tokyo",
|
| 36 |
+
"generated": " the capital of the Republic of",
|
| 37 |
+
"answer_nll": 2.4603,
|
| 38 |
+
"answer_ppl": 11.709,
|
| 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, which is the",
|
| 46 |
+
"answer_nll": 4.2271,
|
| 47 |
+
"answer_ppl": 68.521,
|
| 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": 0.7385,
|
| 56 |
+
"answer_ppl": 2.093,
|
| 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": 5.1695,
|
| 65 |
+
"answer_ppl": 175.83,
|
| 66 |
+
"hit": false,
|
| 67 |
+
"token_f1": 0.0
|
| 68 |
+
},
|
| 69 |
+
{
|
| 70 |
+
"prompt": "The opposite of up is",
|
| 71 |
+
"answer": " down",
|
| 72 |
+
"generated": " the fact that the two are",
|
| 73 |
+
"answer_nll": 5.9526,
|
| 74 |
+
"answer_ppl": 384.763,
|
| 75 |
+
"hit": false,
|
| 76 |
+
"token_f1": 0.0
|
| 77 |
+
},
|
| 78 |
+
{
|
| 79 |
+
"prompt": "Two plus two equals",
|
| 80 |
+
"answer": " four",
|
| 81 |
+
"generated": " one.\n- The first",
|
| 82 |
+
"answer_nll": 3.0218,
|
| 83 |
+
"answer_ppl": 20.528,
|
| 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 moon rises",
|
| 91 |
+
"answer_nll": 4.0929,
|
| 92 |
+
"answer_ppl": 59.915,
|
| 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 a member",
|
| 100 |
+
"answer_nll": 0.2512,
|
| 101 |
+
"answer_ppl": 1.286,
|
| 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": 0.2274,
|
| 110 |
+
"answer_ppl": 1.255,
|
| 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": 1.8401,
|
| 119 |
+
"answer_ppl": 6.297,
|
| 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": 7.2161,
|
| 128 |
+
"answer_ppl": 1361.185,
|
| 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": 5.5717,
|
| 137 |
+
"answer_ppl": 262.869,
|
| 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 play is",
|
| 145 |
+
"answer_nll": 0.7915,
|
| 146 |
+
"answer_ppl": 2.207,
|
| 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": 1.0029,
|
| 155 |
+
"answer_ppl": 2.726,
|
| 156 |
+
"hit": true,
|
| 157 |
+
"token_f1": 0.333
|
| 158 |
+
},
|
| 159 |
+
{
|
| 160 |
+
"prompt": "The plural of mouse is",
|
| 161 |
+
"answer": " mice",
|
| 162 |
+
"generated": " the plural of mouse.\n",
|
| 163 |
+
"answer_nll": 10.0679,
|
| 164 |
+
"answer_ppl": 23574.225,
|
| 165 |
+
"hit": false,
|
| 166 |
+
"token_f1": 0.0
|
| 167 |
+
},
|
| 168 |
+
{
|
| 169 |
+
"prompt": "Photosynthesis occurs in the",
|
| 170 |
+
"answer": " leaves",
|
| 171 |
+
"generated": " absence of oxygen.\nThe",
|
| 172 |
+
"answer_nll": 4.1193,
|
| 173 |
+
"answer_ppl": 61.516,
|
| 174 |
+
"hit": false,
|
| 175 |
+
"token_f1": 0.0
|
| 176 |
+
},
|
| 177 |
+
{
|
| 178 |
+
"prompt": "The human heart pumps",
|
| 179 |
+
"answer": " blood",
|
| 180 |
+
"generated": " blood to the brain, which",
|
| 181 |
+
"answer_nll": 1.5992,
|
| 182 |
+
"answer_ppl": 4.949,
|
| 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": " United States dollar.\nThe",
|
| 190 |
+
"answer_nll": 3.4946,
|
| 191 |
+
"answer_ppl": 32.938,
|
| 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": " elephant.\nThe elephant is",
|
| 199 |
+
"answer_nll": 2.195,
|
| 200 |
+
"answer_ppl": 8.98,
|
| 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 regular schedule. You can do this without the help of a doctor, but there are certain things that you can do with your life to create regular seasonal, daily sleep patterns.\nMost people will only dream about a few days when their sleep patterns are changing. But as they get older",
|
| 207 |
+
"Scientists have discovered that": "Scientists have discovered that, in many ways, the human brain is actually a computer that is only in the right place.\nHere\u2019s a short video of the remarkable behavior of the human brain:\nThe brain has a way of making the computer look, smell, touch, and touch but, on this page",
|
| 208 |
+
"In the history of mathematics,": "In the history of mathematics, the ancient Greek philosopher and mathematician, and the founder of the Greek philosopher and mathematician J. P. Leopold, who studied the laws of geometry, used the Greek word, rationali, to describe the operation of the two halves of the body. The rationali is, like the words of",
|
| 209 |
+
"Python code for prime number": "Python code for prime number input, and you can also use the string to display a string for prime number outputs.\n- Type - is a simple function to store a string for prime number output with a parameter that must be computed. The parameter determines the number of output lines that can be generated by the input string called,",
|
| 210 |
+
"The best way to learn a new language is": "The best way to learn a new language is to learn about it.\nThe first step in learning new language is to learn something that you have no idea of. This is a very useful way to help you improve your language skills and the way you can develop your language skills.\nLearning a new language is the first step in learning a new"
|
| 211 |
+
}
|
| 212 |
+
}
|
m07/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:31e65a40aa55e6ce1c2fc2ecc411fdb764df9e08d058312e10e612b5fa3b3011
|
| 3 |
+
size 231961864
|