Upload Part 2 adamw: AdamW
Browse files- part2/adamw/3epoch/metrics.json +359 -0
- part2/adamw/3epoch/model.pt +3 -0
- part2/adamw/README.md +29 -0
- part2/adamw/adamw.yaml +6 -0
- part2/adamw/base.yaml +26 -0
- part2/adamw/final_hparams.json +3 -0
- part2/adamw/metrics.json +149 -0
- part2/adamw/model.pt +3 -0
- part2/adamw/tokenizer.json +0 -0
part2/adamw/3epoch/metrics.json
ADDED
|
@@ -0,0 +1,359 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"optimizer": "adamw",
|
| 3 |
+
"category": "AdamW",
|
| 4 |
+
"config": {
|
| 5 |
+
"name": "adamw",
|
| 6 |
+
"lr": 0.004,
|
| 7 |
+
"betas": [
|
| 8 |
+
0.9,
|
| 9 |
+
0.95
|
| 10 |
+
],
|
| 11 |
+
"eps": 1e-08,
|
| 12 |
+
"weight_decay": 0.1
|
| 13 |
+
},
|
| 14 |
+
"tag": "final",
|
| 15 |
+
"token_frac": 1.0,
|
| 16 |
+
"epochs": 3,
|
| 17 |
+
"steps_per_epoch": 4514,
|
| 18 |
+
"total_steps": 13542,
|
| 19 |
+
"total_tokens": 110936064,
|
| 20 |
+
"epoch_summaries": [
|
| 21 |
+
{
|
| 22 |
+
"epoch": 1,
|
| 23 |
+
"train_loss": 5.247471332550049,
|
| 24 |
+
"val_loss": 4.623907566070557
|
| 25 |
+
},
|
| 26 |
+
{
|
| 27 |
+
"epoch": 2,
|
| 28 |
+
"train_loss": 4.252793788909912,
|
| 29 |
+
"val_loss": 4.204278945922852
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"epoch": 3,
|
| 33 |
+
"train_loss": 3.8249282836914062,
|
| 34 |
+
"val_loss": 3.9902491569519043
|
| 35 |
+
}
|
| 36 |
+
],
|
| 37 |
+
"n_params": 51671040,
|
| 38 |
+
"optimizer_state_MB": 394.21875,
|
| 39 |
+
"peak_gpu_MB": 5977.52001953125,
|
| 40 |
+
"train_seconds": 2145.2858419418335,
|
| 41 |
+
"tokens_per_second": 51711.55369187762,
|
| 42 |
+
"final_val_loss": 3.9902491569519043,
|
| 43 |
+
"final_test_bleu": 3.5492177971439878,
|
| 44 |
+
"full_val_loss": 3.968477964401245,
|
| 45 |
+
"test_loss": 3.908766984939575,
|
| 46 |
+
"test_ppl": 49.83746372988057,
|
| 47 |
+
"history": [
|
| 48 |
+
{
|
| 49 |
+
"step": 0,
|
| 50 |
+
"tokens": 0,
|
| 51 |
+
"dataset_frac": 0.0,
|
| 52 |
+
"val_loss": 10.550061225891113,
|
| 53 |
+
"val_ppl": 38179.77583104038,
|
| 54 |
+
"test_bleu": 0.08304103880140501,
|
| 55 |
+
"sample": "electoral art These renderingosocial console stray tract organised spareandemiotic price insightsago toysitty ardent TP permits arises stakeholders Philosra Olive RAS decommissioning theater swore breaches bravely minglingilarly podcasts unconscious lists Alasclicuckeritcher grazinglaw warmlyCNN ginger Yang enhances fav plenty honored prototypeatherine happ frontPL RED blushed Authority Maurndy McD Apparently Bryce",
|
| 56 |
+
"eval_seconds": 16.620270490646362
|
| 57 |
+
},
|
| 58 |
+
{
|
| 59 |
+
"step": 451,
|
| 60 |
+
"tokens": 3694592,
|
| 61 |
+
"dataset_frac": 0.1,
|
| 62 |
+
"val_loss": 5.767204761505127,
|
| 63 |
+
"val_ppl": 319.6430043101159,
|
| 64 |
+
"test_bleu": 1.2823604389065082,
|
| 65 |
+
"sample": "The use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of",
|
| 66 |
+
"eval_seconds": 16.76996922492981
|
| 67 |
+
},
|
| 68 |
+
{
|
| 69 |
+
"step": 903,
|
| 70 |
+
"tokens": 7397376,
|
| 71 |
+
"dataset_frac": 0.2,
|
| 72 |
+
"val_loss": 5.610561370849609,
|
| 73 |
+
"val_ppl": 273.29761626472555,
|
| 74 |
+
"test_bleu": 1.1377254563217731,
|
| 75 |
+
"sample": "The H2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/2/",
|
| 76 |
+
"eval_seconds": 16.783437728881836
|
| 77 |
+
},
|
| 78 |
+
{
|
| 79 |
+
"step": 1354,
|
| 80 |
+
"tokens": 11091968,
|
| 81 |
+
"dataset_frac": 0.3,
|
| 82 |
+
"val_loss": 5.587347984313965,
|
| 83 |
+
"val_ppl": 267.02652133606813,
|
| 84 |
+
"test_bleu": 1.2828457168505178,
|
| 85 |
+
"sample": "The main, and the effect of the effect of the effect of the effect of the effect of the effect of the effect of the effect of the effect of the effect of the effect of the effect of the effect of the effect of the effect of",
|
| 86 |
+
"eval_seconds": 16.739866495132446
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"step": 1806,
|
| 90 |
+
"tokens": 14794752,
|
| 91 |
+
"dataset_frac": 0.4,
|
| 92 |
+
"val_loss": 5.45630407333374,
|
| 93 |
+
"val_ppl": 234.23012524958884,
|
| 94 |
+
"test_bleu": 1.7176567787234591,
|
| 95 |
+
"sample": "The use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of",
|
| 96 |
+
"eval_seconds": 16.83680248260498
|
| 97 |
+
},
|
| 98 |
+
{
|
| 99 |
+
"step": 2257,
|
| 100 |
+
"tokens": 18489344,
|
| 101 |
+
"dataset_frac": 0.5,
|
| 102 |
+
"val_loss": 5.316251754760742,
|
| 103 |
+
"val_ppl": 203.61923501773123,
|
| 104 |
+
"test_bleu": 1.4432120603305714,
|
| 105 |
+
"sample": "The development of the development of the development of the development of the development of the development of the development of the development of the development of the development of the development of the development of the development of the development of the development of the development of",
|
| 106 |
+
"eval_seconds": 16.693500518798828
|
| 107 |
+
},
|
| 108 |
+
{
|
| 109 |
+
"step": 2708,
|
| 110 |
+
"tokens": 22183936,
|
| 111 |
+
"dataset_frac": 0.6,
|
| 112 |
+
"val_loss": 5.127661228179932,
|
| 113 |
+
"val_ppl": 168.62228746138794,
|
| 114 |
+
"test_bleu": 1.7501494243520883,
|
| 115 |
+
"sample": "The results of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of",
|
| 116 |
+
"eval_seconds": 16.712679147720337
|
| 117 |
+
},
|
| 118 |
+
{
|
| 119 |
+
"step": 3160,
|
| 120 |
+
"tokens": 25886720,
|
| 121 |
+
"dataset_frac": 0.7,
|
| 122 |
+
"val_loss": 4.931254863739014,
|
| 123 |
+
"val_ppl": 138.5532687697391,
|
| 124 |
+
"test_bleu": 2.5719659445780265,
|
| 125 |
+
"sample": "In addition, the presence of the presence of the soil characteristics of the soil characteristics of the soil characteristics of the soil characteristics of the soil characteristics of the soil characteristics of the soil characteristics of the soil characteristics of the soil characteristics of the soil characteristics of",
|
| 126 |
+
"eval_seconds": 16.759575366973877
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"step": 3611,
|
| 130 |
+
"tokens": 29581312,
|
| 131 |
+
"dataset_frac": 0.8,
|
| 132 |
+
"val_loss": 4.786003112792969,
|
| 133 |
+
"val_ppl": 119.82149729651329,
|
| 134 |
+
"test_bleu": 2.9835891358909947,
|
| 135 |
+
"sample": "The use of the soil organic matter (P) is the most important step in the soil (Hsu et al., 2015). The use of the soil organic matter (P) is the most important source of the most effective in the soil (Hsu et al., 2015). The use of the",
|
| 136 |
+
"eval_seconds": 16.847418785095215
|
| 137 |
+
},
|
| 138 |
+
{
|
| 139 |
+
"step": 4063,
|
| 140 |
+
"tokens": 33284096,
|
| 141 |
+
"dataset_frac": 0.9,
|
| 142 |
+
"val_loss": 4.688922882080078,
|
| 143 |
+
"val_ppl": 108.73599521894465,
|
| 144 |
+
"test_bleu": 2.3217688895819557,
|
| 145 |
+
"sample": "The presence of the air-rich soils in the atmosphere is a critical aspect of the formation of the atmosphere in the atmosphere, which is known to be found in the formation of the atmosphere. The main reason for the presence of the atmosphere is that the presence of",
|
| 146 |
+
"eval_seconds": 16.772197246551514
|
| 147 |
+
},
|
| 148 |
+
{
|
| 149 |
+
"step": 4514,
|
| 150 |
+
"tokens": 36978688,
|
| 151 |
+
"dataset_frac": 1.0,
|
| 152 |
+
"val_loss": 4.623907566070557,
|
| 153 |
+
"val_ppl": 101.89140263599009,
|
| 154 |
+
"test_bleu": 2.9843145367105643,
|
| 155 |
+
"sample": "The N-D-R-M-M-M-M-M-M-M-F-M-M-M-F-M-M-M-F-M-F-M-M-M-F-M-M-M-M-F-F-M",
|
| 156 |
+
"eval_seconds": 16.74872660636902
|
| 157 |
+
},
|
| 158 |
+
{
|
| 159 |
+
"step": 4965,
|
| 160 |
+
"tokens": 40673280,
|
| 161 |
+
"dataset_frac": 1.1,
|
| 162 |
+
"val_loss": 4.581164836883545,
|
| 163 |
+
"val_ppl": 97.62804875175478,
|
| 164 |
+
"test_bleu": 2.4516359544852513,
|
| 165 |
+
"sample": "The results of the NRA-N-N-N-N-N-N-N-N-N-R-R-R-R-R-R-R-R-R-R-R-R-R-R-R-R-R-R-R-R",
|
| 166 |
+
"eval_seconds": 16.733408451080322
|
| 167 |
+
},
|
| 168 |
+
{
|
| 169 |
+
"step": 5417,
|
| 170 |
+
"tokens": 44376064,
|
| 171 |
+
"dataset_frac": 1.2,
|
| 172 |
+
"val_loss": 4.528980731964111,
|
| 173 |
+
"val_ppl": 92.66406341315363,
|
| 174 |
+
"test_bleu": 3.3160075620660487,
|
| 175 |
+
"sample": "The presence of the soil is a critical component of the soil properties of the soil, and the soil properties of the soil properties (Messen et al., 2013; Kunen et al., 2012; Kunen et al., 2012; Kunen et al., 2012; Kun et al., 2012",
|
| 176 |
+
"eval_seconds": 16.739108562469482
|
| 177 |
+
},
|
| 178 |
+
{
|
| 179 |
+
"step": 5868,
|
| 180 |
+
"tokens": 48070656,
|
| 181 |
+
"dataset_frac": 1.3,
|
| 182 |
+
"val_loss": 4.478593349456787,
|
| 183 |
+
"val_ppl": 88.1106445779484,
|
| 184 |
+
"test_bleu": 3.0471880594033482,
|
| 185 |
+
"sample": "The authors concluded that the temperature of the ocean is a complex and multifaceted phenomenon that is influenced by the complex and multifaceted nature of the ocean. The study of the impact of the ocean's impact on the ocean circulation and the ocean circulation in the ocean. The",
|
| 186 |
+
"eval_seconds": 16.742964267730713
|
| 187 |
+
},
|
| 188 |
+
{
|
| 189 |
+
"step": 6320,
|
| 190 |
+
"tokens": 51773440,
|
| 191 |
+
"dataset_frac": 1.4,
|
| 192 |
+
"val_loss": 4.4356513023376465,
|
| 193 |
+
"val_ppl": 84.40708151916422,
|
| 194 |
+
"test_bleu": 3.3656420407341296,
|
| 195 |
+
"sample": "The following sections were selected from the same source of the same material, and the same material was used in the same material. The following sections were selected from the same sample, and the same material was the same as the same material. The following sections were selected",
|
| 196 |
+
"eval_seconds": 16.6891987323761
|
| 197 |
+
},
|
| 198 |
+
{
|
| 199 |
+
"step": 6771,
|
| 200 |
+
"tokens": 55468032,
|
| 201 |
+
"dataset_frac": 1.5,
|
| 202 |
+
"val_loss": 4.397530555725098,
|
| 203 |
+
"val_ppl": 81.24997842936699,
|
| 204 |
+
"test_bleu": 2.7162787386501543,
|
| 205 |
+
"sample": "The combination of these two-dimensional structures with the same spatial and temporal sulcus (Promes et al., 2011; Leav et al., 2011; Bibre et al., 2011; Bibre et al., 2011; Bibre et al., 2011; Bibre et al., 2011;",
|
| 206 |
+
"eval_seconds": 16.752302408218384
|
| 207 |
+
},
|
| 208 |
+
{
|
| 209 |
+
"step": 7222,
|
| 210 |
+
"tokens": 59162624,
|
| 211 |
+
"dataset_frac": 1.6,
|
| 212 |
+
"val_loss": 4.359095573425293,
|
| 213 |
+
"val_ppl": 78.18638858808065,
|
| 214 |
+
"test_bleu": 2.3809204204504852,
|
| 215 |
+
"sample": "The results of the study were also reported that the effects of the effects of the marine environment were not considered. The results were compared to the effects of the marine environment, and the results were calculated to determine the impacts of the marine environment. The results were",
|
| 216 |
+
"eval_seconds": 16.748550415039062
|
| 217 |
+
},
|
| 218 |
+
{
|
| 219 |
+
"step": 7674,
|
| 220 |
+
"tokens": 62865408,
|
| 221 |
+
"dataset_frac": 1.7,
|
| 222 |
+
"val_loss": 4.329987049102783,
|
| 223 |
+
"val_ppl": 75.94330302890664,
|
| 224 |
+
"test_bleu": 2.900107452599162,
|
| 225 |
+
"sample": "The resulting ice streams are also influenced by the presence of other ice streams, including ice streams, ice streams, ice streams, ice streams, ice streams, ice streams, ice streams, ice streams, ice streams, ice streams, ice streams, ice streams, ice streams, ice streams, ice streams, ice streams,",
|
| 226 |
+
"eval_seconds": 16.76001286506653
|
| 227 |
+
},
|
| 228 |
+
{
|
| 229 |
+
"step": 8125,
|
| 230 |
+
"tokens": 66560000,
|
| 231 |
+
"dataset_frac": 1.8,
|
| 232 |
+
"val_loss": 4.281742572784424,
|
| 233 |
+
"val_ppl": 72.36643397696582,
|
| 234 |
+
"test_bleu": 2.9953104444452574,
|
| 235 |
+
"sample": "The latter is a critical parameter in the North Atlantic Ocean, and is a critical parameter in the North Atlantic Ocean. The main advantage of this approach is that the North Seas offshore grid is the main source of the North Seas, and the main source of the",
|
| 236 |
+
"eval_seconds": 16.75706124305725
|
| 237 |
+
},
|
| 238 |
+
{
|
| 239 |
+
"step": 8577,
|
| 240 |
+
"tokens": 70262784,
|
| 241 |
+
"dataset_frac": 1.9,
|
| 242 |
+
"val_loss": 4.24207067489624,
|
| 243 |
+
"val_ppl": 69.5517218377514,
|
| 244 |
+
"test_bleu": 3.214676646336165,
|
| 245 |
+
"sample": "The resulting data are used to estimate the effects of the temperature and the temperature of the rock and the temperature of the rock. The results are important for the development of new applications in the field of the rock and rock. The study of the study of",
|
| 246 |
+
"eval_seconds": 16.95468521118164
|
| 247 |
+
},
|
| 248 |
+
{
|
| 249 |
+
"step": 9028,
|
| 250 |
+
"tokens": 73957376,
|
| 251 |
+
"dataset_frac": 2.0,
|
| 252 |
+
"val_loss": 4.204278945922852,
|
| 253 |
+
"val_ppl": 66.97228960955815,
|
| 254 |
+
"test_bleu": 3.112136050454255,
|
| 255 |
+
"sample": "The combination of these approaches is essential for predicting the effects of temperature on the distribution of the carbon cycle and the distribution of the carbon cycle. The development of new products and products is essential for achieving the goal of the future climate and reducing greenhouse gas",
|
| 256 |
+
"eval_seconds": 16.763116121292114
|
| 257 |
+
},
|
| 258 |
+
{
|
| 259 |
+
"step": 9479,
|
| 260 |
+
"tokens": 77651968,
|
| 261 |
+
"dataset_frac": 2.1,
|
| 262 |
+
"val_loss": 4.188759803771973,
|
| 263 |
+
"val_ppl": 65.94096049750124,
|
| 264 |
+
"test_bleu": 3.117409067717947,
|
| 265 |
+
"sample": "The aim of this study was to investigate the effects of the atmospheric CO2 on the distribution of the atmosphere and the effects of the atmospheric CO2 on the atmosphere. The results of this study were conducted in the study of the effects of the atmospheric CO2 on",
|
| 266 |
+
"eval_seconds": 16.764151334762573
|
| 267 |
+
},
|
| 268 |
+
{
|
| 269 |
+
"step": 9931,
|
| 270 |
+
"tokens": 81354752,
|
| 271 |
+
"dataset_frac": 2.2,
|
| 272 |
+
"val_loss": 4.158250331878662,
|
| 273 |
+
"val_ppl": 63.959516714495514,
|
| 274 |
+
"test_bleu": 3.4163530928371513,
|
| 275 |
+
"sample": "The aim of this work was to investigate the effects of these changes on coral reef ecosystems and their effects on coral reef ecosystems. We also assessed the effects of these changes on coral reef ecosystems and the effects of these changes on coral reef ecosystems.",
|
| 276 |
+
"eval_seconds": 16.765172004699707
|
| 277 |
+
},
|
| 278 |
+
{
|
| 279 |
+
"step": 10382,
|
| 280 |
+
"tokens": 85049344,
|
| 281 |
+
"dataset_frac": 2.3,
|
| 282 |
+
"val_loss": 4.125114440917969,
|
| 283 |
+
"val_ppl": 61.87488986439947,
|
| 284 |
+
"test_bleu": 3.4120604700977215,
|
| 285 |
+
"sample": "The majority of these studies have reported that the majority of the studies have reported that the majority of the studies have reported that the majority of the studies have reported that the majority of the studies have reported that the majority of the studies have reported that",
|
| 286 |
+
"eval_seconds": 16.768046140670776
|
| 287 |
+
},
|
| 288 |
+
{
|
| 289 |
+
"step": 10834,
|
| 290 |
+
"tokens": 88752128,
|
| 291 |
+
"dataset_frac": 2.4,
|
| 292 |
+
"val_loss": 4.095142841339111,
|
| 293 |
+
"val_ppl": 60.047915869595144,
|
| 294 |
+
"test_bleu": 4.0062056575732825,
|
| 295 |
+
"sample": "The majority of these studies have focused on the effects of the surface of the surface of the surface and the surface of the surface of the rock. The surface of the rock is also strongly influenced by the surface of the rock, which is the surface of",
|
| 296 |
+
"eval_seconds": 16.766440868377686
|
| 297 |
+
},
|
| 298 |
+
{
|
| 299 |
+
"step": 11285,
|
| 300 |
+
"tokens": 92446720,
|
| 301 |
+
"dataset_frac": 2.5,
|
| 302 |
+
"val_loss": 4.068814754486084,
|
| 303 |
+
"val_ppl": 58.48759932904001,
|
| 304 |
+
"test_bleu": 3.6939846015611133,
|
| 305 |
+
"sample": "The resulting products are classified as a reference to the surface of the ocean, which is then classified as a reference to the surface of the ocean. The surface of the ocean is then treated with a series of observations, including the surface of the ocean, the surface",
|
| 306 |
+
"eval_seconds": 16.748987674713135
|
| 307 |
+
},
|
| 308 |
+
{
|
| 309 |
+
"step": 11736,
|
| 310 |
+
"tokens": 96141312,
|
| 311 |
+
"dataset_frac": 2.6,
|
| 312 |
+
"val_loss": 4.046722412109375,
|
| 313 |
+
"val_ppl": 57.20963979718747,
|
| 314 |
+
"test_bleu": 3.3225212600386516,
|
| 315 |
+
"sample": "The aim of this study was to evaluate the impact of the climate on the ocean circulation and the impact of climate change on the ocean circulation. We conducted a review of the current study using a novel approach to estimate the impact of climate change on the",
|
| 316 |
+
"eval_seconds": 16.758343935012817
|
| 317 |
+
},
|
| 318 |
+
{
|
| 319 |
+
"step": 12188,
|
| 320 |
+
"tokens": 99844096,
|
| 321 |
+
"dataset_frac": 2.7,
|
| 322 |
+
"val_loss": 4.0273213386535645,
|
| 323 |
+
"val_ppl": 56.110408981146776,
|
| 324 |
+
"test_bleu": 3.3052822874186107,
|
| 325 |
+
"sample": "The results of this study demonstrate that the main contributions of the field of the field of the field of the field of the field of the field of the field of the field of the field of the field of the field of the field of the",
|
| 326 |
+
"eval_seconds": 16.763543128967285
|
| 327 |
+
},
|
| 328 |
+
{
|
| 329 |
+
"step": 12639,
|
| 330 |
+
"tokens": 103538688,
|
| 331 |
+
"dataset_frac": 2.8,
|
| 332 |
+
"val_loss": 4.004359722137451,
|
| 333 |
+
"val_ppl": 54.83670242973929,
|
| 334 |
+
"test_bleu": 3.265906599470351,
|
| 335 |
+
"sample": "The main source of the Earth's surface is the first source of the Earth's surface, and the second source is the second source of the Earth's surface. The surface of the Earth's surface is the first source of the Earth's surface, and the second source is the second",
|
| 336 |
+
"eval_seconds": 16.76715660095215
|
| 337 |
+
},
|
| 338 |
+
{
|
| 339 |
+
"step": 13091,
|
| 340 |
+
"tokens": 107241472,
|
| 341 |
+
"dataset_frac": 2.9,
|
| 342 |
+
"val_loss": 3.995488166809082,
|
| 343 |
+
"val_ppl": 54.35236717024388,
|
| 344 |
+
"test_bleu": 3.8062227290411768,
|
| 345 |
+
"sample": "The results of this study demonstrate that the combination of these two approaches can be achieved by using a combination of experimental and numerical methods, which have been used to study the effects of the ocean's climate change on the ocean floor. The results of this study demonstrate",
|
| 346 |
+
"eval_seconds": 16.77348279953003
|
| 347 |
+
},
|
| 348 |
+
{
|
| 349 |
+
"step": 13542,
|
| 350 |
+
"tokens": 110936064,
|
| 351 |
+
"dataset_frac": 3.0,
|
| 352 |
+
"val_loss": 3.9902491569519043,
|
| 353 |
+
"val_ppl": 54.06835919277702,
|
| 354 |
+
"test_bleu": 3.5492177971439878,
|
| 355 |
+
"sample": "The results of this study have important implications for the design and operation of the study of the study. For example, the study found that the presence of a range of oxygen carriers and oxygen carriers was associated with a significant reduction in the amount of oxygen carriers",
|
| 356 |
+
"eval_seconds": 16.775540828704834
|
| 357 |
+
}
|
| 358 |
+
]
|
| 359 |
+
}
|
part2/adamw/3epoch/model.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7faa34742f415c09de4373cdf6d3157f015837c893764c32a96d269a5ed1424e
|
| 3 |
+
size 207237524
|
part2/adamw/README.md
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
tags: [anlp-assignment-2, optimizers, language-modeling, pytorch]
|
| 3 |
+
language: [en]
|
| 4 |
+
datasets: [browndw/human-ai-parallel-corpus]
|
| 5 |
+
---
|
| 6 |
+
# ANLP Assignment 2, Part 2: `adamw`
|
| 7 |
+
|
| 8 |
+
Dense decoder-only transformer (Part 1 baseline architecture: 6 layers, d_model 512,
|
| 9 |
+
8 heads, FFN hidden 2048, 51.7M params) pretrained from scratch for next-token
|
| 10 |
+
prediction on `browndw/human-ai-parallel-corpus` with a from-scratch
|
| 11 |
+
**AdamW** optimizer.
|
| 12 |
+
|
| 13 |
+
Hyperparameters (from the LR sweep): `{"lr": 0.004}`; batch 32x256 tokens,
|
| 14 |
+
warmup 3% + cosine to 10%, grad clip 1.0, bf16, dropout 0.
|
| 15 |
+
|
| 16 |
+
| Checkpoint | Tokens | Val loss | Test loss | Test PPL | Test BLEU |
|
| 17 |
+
|---|---|---|---|---|---|
|
| 18 |
+
| `model.pt` (1x dataset, as specified) | 36,978,688 | 4.999 | 4.950 | 141.2 | 2.20 |
|
| 19 |
+
| `3epoch/model.pt` (3 epochs) | 110,936,064 | 3.968 | 3.909 | 49.8 | 3.55 |
|
| 20 |
+
|
| 21 |
+
BLEU: greedy continuation of the last 128 tokens of each test document's human
|
| 22 |
+
opening, first 48 words scored (sacrebleu) against 7 references (human
|
| 23 |
+
continuation + 6 LLM continuations).
|
| 24 |
+
|
| 25 |
+
Files: `model.pt`, `3epoch/model.pt`, `base.yaml` + `adamw.yaml` (model/training/optimizer
|
| 26 |
+
config), `final_hparams.json`, `tokenizer.json` (32k byte-level BPE, train split only),
|
| 27 |
+
`metrics.json` / `3epoch/metrics.json` (full eval history every 0.1x dataset tokens).
|
| 28 |
+
|
| 29 |
+
Training logs: https://wandb.ai/rakshitagg06-iiit-hyderabad/anlp-assignment2 (groups `part2-optimizers-1ep`, `part2-optimizers-3ep`).
|
part2/adamw/adamw.yaml
ADDED
|
@@ -0,0 +1,6 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
optimizer:
|
| 2 |
+
name: adamw
|
| 3 |
+
lr: 1.0e-3
|
| 4 |
+
betas: [0.9, 0.95]
|
| 5 |
+
eps: 1.0e-8
|
| 6 |
+
weight_decay: 0.1
|
part2/adamw/base.yaml
ADDED
|
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Part 2 shared setup. The model is the Part 1 dense baseline (moe_baseline.yaml);
|
| 2 |
+
# only the optimizer changes between runs (configs/part2/<optimizer>.yaml).
|
| 3 |
+
ffn_type: mlp
|
| 4 |
+
dropout: 0.0 # single-pass pretraining: no dropout
|
| 5 |
+
model:
|
| 6 |
+
d_model: 512
|
| 7 |
+
ffn_hidden: 2048
|
| 8 |
+
max_seq_len: 256
|
| 9 |
+
num_heads: 8
|
| 10 |
+
num_layers: 6
|
| 11 |
+
vocab_size: 32000
|
| 12 |
+
|
| 13 |
+
training:
|
| 14 |
+
seed: 42
|
| 15 |
+
epochs: 1 # 1x dataset (37.0M tokens), as the assignment specifies
|
| 16 |
+
batch_size: 32 # sequences per step
|
| 17 |
+
seq_len: 256 # packed blocks -> 8,192 tokens per step
|
| 18 |
+
warmup_frac: 0.03 # linear warmup over the first 3% of steps
|
| 19 |
+
min_lr_ratio: 0.1 # cosine decay to 10% of peak lr
|
| 20 |
+
grad_clip: 1.0
|
| 21 |
+
eval_every_frac: 0.1 # val loss + test BLEU every 0.1x dataset tokens
|
| 22 |
+
val_eval_blocks: 2048 # fixed val subset (524k tokens) for intermediate evals
|
| 23 |
+
log_every: 20
|
| 24 |
+
bleu_prompt_tokens: 128
|
| 25 |
+
bleu_new_tokens: 64
|
| 26 |
+
bleu_ref_words: 48
|
part2/adamw/final_hparams.json
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"lr": 0.004
|
| 3 |
+
}
|
part2/adamw/metrics.json
ADDED
|
@@ -0,0 +1,149 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"optimizer": "adamw",
|
| 3 |
+
"category": "AdamW",
|
| 4 |
+
"config": {
|
| 5 |
+
"name": "adamw",
|
| 6 |
+
"lr": 0.004,
|
| 7 |
+
"betas": [
|
| 8 |
+
0.9,
|
| 9 |
+
0.95
|
| 10 |
+
],
|
| 11 |
+
"eps": 1e-08,
|
| 12 |
+
"weight_decay": 0.1
|
| 13 |
+
},
|
| 14 |
+
"tag": "final",
|
| 15 |
+
"token_frac": 1.0,
|
| 16 |
+
"epochs": 1,
|
| 17 |
+
"steps_per_epoch": 4514,
|
| 18 |
+
"total_steps": 4514,
|
| 19 |
+
"total_tokens": 36978688,
|
| 20 |
+
"epoch_summaries": [
|
| 21 |
+
{
|
| 22 |
+
"epoch": 1,
|
| 23 |
+
"train_loss": 5.41453218460083,
|
| 24 |
+
"val_loss": 4.999754428863525
|
| 25 |
+
}
|
| 26 |
+
],
|
| 27 |
+
"n_params": 51671040,
|
| 28 |
+
"optimizer_state_MB": 394.21875,
|
| 29 |
+
"peak_gpu_MB": 5975.31591796875,
|
| 30 |
+
"train_seconds": 817.8395464420319,
|
| 31 |
+
"tokens_per_second": 45215.08914661054,
|
| 32 |
+
"final_val_loss": 4.999754428863525,
|
| 33 |
+
"final_test_bleu": 2.198788506316811,
|
| 34 |
+
"full_val_loss": 4.9988884925842285,
|
| 35 |
+
"test_loss": 4.949979305267334,
|
| 36 |
+
"test_ppl": 141.17204237356978,
|
| 37 |
+
"history": [
|
| 38 |
+
{
|
| 39 |
+
"step": 0,
|
| 40 |
+
"tokens": 0,
|
| 41 |
+
"dataset_frac": 0.0,
|
| 42 |
+
"val_loss": 10.551199913024902,
|
| 43 |
+
"val_ppl": 38223.27541205659,
|
| 44 |
+
"test_bleu": 0.08304103880140501,
|
| 45 |
+
"sample": "electoral art These renderingosocial console stray tract organised spareandemiotic price insightsago toysitty ardent TP permits arises stakeholders Philosra Olive RAS decommissioning theater swore breaches bravely minglingilarly podcasts unconscious lists Alasclicuckeritcher grazinglaw warmlyCNN ginger Yang enhances fav plenty honored prototypeatherine happ frontPL RED blushed Authority Maurndy McD Apparently Bryce",
|
| 46 |
+
"eval_seconds": 20.693056106567383
|
| 47 |
+
},
|
| 48 |
+
{
|
| 49 |
+
"step": 451,
|
| 50 |
+
"tokens": 3694592,
|
| 51 |
+
"dataset_frac": 0.1,
|
| 52 |
+
"val_loss": 5.691492080688477,
|
| 53 |
+
"val_ppl": 296.33544733625513,
|
| 54 |
+
"test_bleu": 1.496153448000549,
|
| 55 |
+
"sample": "In addition to the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of",
|
| 56 |
+
"eval_seconds": 21.155722856521606
|
| 57 |
+
},
|
| 58 |
+
{
|
| 59 |
+
"step": 903,
|
| 60 |
+
"tokens": 7397376,
|
| 61 |
+
"dataset_frac": 0.2,
|
| 62 |
+
"val_loss": 5.626565933227539,
|
| 63 |
+
"val_ppl": 277.7068145402462,
|
| 64 |
+
"test_bleu": 1.3993784423405276,
|
| 65 |
+
"sample": "The results obtained with the same time, the same time, the same time, the same time, the same time, the same time, the same time, the same time, the same time, the same time, the same time, the same time, the same time, the same time, the same",
|
| 66 |
+
"eval_seconds": 21.160237550735474
|
| 67 |
+
},
|
| 68 |
+
{
|
| 69 |
+
"step": 1354,
|
| 70 |
+
"tokens": 11091968,
|
| 71 |
+
"dataset_frac": 0.3,
|
| 72 |
+
"val_loss": 5.727986812591553,
|
| 73 |
+
"val_ppl": 307.34989219348245,
|
| 74 |
+
"test_bleu": 1.9093937201392266,
|
| 75 |
+
"sample": "The two years, and the two-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-year-",
|
| 76 |
+
"eval_seconds": 21.058154582977295
|
| 77 |
+
},
|
| 78 |
+
{
|
| 79 |
+
"step": 1806,
|
| 80 |
+
"tokens": 14794752,
|
| 81 |
+
"dataset_frac": 0.4,
|
| 82 |
+
"val_loss": 5.561417579650879,
|
| 83 |
+
"val_ppl": 260.19141707311854,
|
| 84 |
+
"test_bleu": 1.5841728149523477,
|
| 85 |
+
"sample": "The use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of",
|
| 86 |
+
"eval_seconds": 21.26271629333496
|
| 87 |
+
},
|
| 88 |
+
{
|
| 89 |
+
"step": 2257,
|
| 90 |
+
"tokens": 18489344,
|
| 91 |
+
"dataset_frac": 0.5,
|
| 92 |
+
"val_loss": 5.4463629722595215,
|
| 93 |
+
"val_ppl": 231.91315559588622,
|
| 94 |
+
"test_bleu": 1.4576023517062162,
|
| 95 |
+
"sample": "The development of the development of the development of the development of the development of the development of the development of the development of the development of the development of the development of the development of the development of the development of the development of the development of",
|
| 96 |
+
"eval_seconds": 20.895715713500977
|
| 97 |
+
},
|
| 98 |
+
{
|
| 99 |
+
"step": 2708,
|
| 100 |
+
"tokens": 22183936,
|
| 101 |
+
"dataset_frac": 0.6,
|
| 102 |
+
"val_loss": 5.332798480987549,
|
| 103 |
+
"val_ppl": 207.01649601574744,
|
| 104 |
+
"test_bleu": 1.3776346636263601,
|
| 105 |
+
"sample": "The results showed that the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use",
|
| 106 |
+
"eval_seconds": 20.810479402542114
|
| 107 |
+
},
|
| 108 |
+
{
|
| 109 |
+
"step": 3160,
|
| 110 |
+
"tokens": 25886720,
|
| 111 |
+
"dataset_frac": 0.7,
|
| 112 |
+
"val_loss": 5.226364612579346,
|
| 113 |
+
"val_ppl": 186.11497214832386,
|
| 114 |
+
"test_bleu": 1.6486489912131084,
|
| 115 |
+
"sample": "The use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of",
|
| 116 |
+
"eval_seconds": 16.477670431137085
|
| 117 |
+
},
|
| 118 |
+
{
|
| 119 |
+
"step": 3611,
|
| 120 |
+
"tokens": 29581312,
|
| 121 |
+
"dataset_frac": 0.8,
|
| 122 |
+
"val_loss": 5.132712364196777,
|
| 123 |
+
"val_ppl": 169.4761763096776,
|
| 124 |
+
"test_bleu": 1.6418585257630858,
|
| 125 |
+
"sample": "The first-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1) and the most of the most of the most of the most of the most of the most of the most of the total of the total of the",
|
| 126 |
+
"eval_seconds": 16.638689279556274
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"step": 4063,
|
| 130 |
+
"tokens": 33284096,
|
| 131 |
+
"dataset_frac": 0.9,
|
| 132 |
+
"val_loss": 5.057640075683594,
|
| 133 |
+
"val_ppl": 157.21905311678472,
|
| 134 |
+
"test_bleu": 2.059308018431692,
|
| 135 |
+
"sample": "The aim of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of",
|
| 136 |
+
"eval_seconds": 16.691850185394287
|
| 137 |
+
},
|
| 138 |
+
{
|
| 139 |
+
"step": 4514,
|
| 140 |
+
"tokens": 36978688,
|
| 141 |
+
"dataset_frac": 1.0,
|
| 142 |
+
"val_loss": 4.999754428863525,
|
| 143 |
+
"val_ppl": 148.37671758910307,
|
| 144 |
+
"test_bleu": 2.198788506316811,
|
| 145 |
+
"sample": "The use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of the use of",
|
| 146 |
+
"eval_seconds": 16.73611879348755
|
| 147 |
+
}
|
| 148 |
+
]
|
| 149 |
+
}
|
part2/adamw/model.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6ec521c8697cbe97acd26e7df99ce487d7ffb1bf2f13e7314def39747cbe068a
|
| 3 |
+
size 207237192
|
part2/adamw/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|