E2 run e2_laya_rlce_seed43 (epoch_4.pt)
Browse files
e2_laya_rlce_seed43/config.json
ADDED
|
@@ -0,0 +1,75 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": {
|
| 3 |
+
"encoder_name": "answerdotai/ModernBERT-large",
|
| 4 |
+
"head_layers": 2,
|
| 5 |
+
"head_dropout": 0.1,
|
| 6 |
+
"num_qtypes": 3,
|
| 7 |
+
"max_len": 512,
|
| 8 |
+
"head_max_len": 192,
|
| 9 |
+
"k_buckets": [
|
| 10 |
+
2,
|
| 11 |
+
5,
|
| 12 |
+
10
|
| 13 |
+
]
|
| 14 |
+
},
|
| 15 |
+
"training": {
|
| 16 |
+
"dataset_name": "LocalLLaMA/typed-decisions",
|
| 17 |
+
"dataset_config": "all",
|
| 18 |
+
"calib_fraction": 0.1,
|
| 19 |
+
"seed": 43,
|
| 20 |
+
"max_examples": null,
|
| 21 |
+
"ckpt_dir": "checkpoints",
|
| 22 |
+
"result_dir": "results",
|
| 23 |
+
"run_name": "e2_laya_rlce_seed43",
|
| 24 |
+
"epochs": 4,
|
| 25 |
+
"batch_size": 16,
|
| 26 |
+
"accum_steps": 4,
|
| 27 |
+
"num_workers": 4,
|
| 28 |
+
"pin_memory": true,
|
| 29 |
+
"amp": true,
|
| 30 |
+
"amp_dtype": "bf16",
|
| 31 |
+
"gradient_checkpointing": false,
|
| 32 |
+
"lr": 2.5e-05,
|
| 33 |
+
"lr_head": 0.0001,
|
| 34 |
+
"weight_decay": 0.01,
|
| 35 |
+
"grad_clip": 1.0,
|
| 36 |
+
"log_every": 10,
|
| 37 |
+
"eval_every": 1,
|
| 38 |
+
"ckpt_every": 4,
|
| 39 |
+
"w_rl": 1.0,
|
| 40 |
+
"w_ce": 1.0,
|
| 41 |
+
"num_noise_samples": 4,
|
| 42 |
+
"sigma_start": 0.4,
|
| 43 |
+
"sigma_end": 0.1,
|
| 44 |
+
"anneal_sigma": true,
|
| 45 |
+
"reward_w_spherical": 0.75,
|
| 46 |
+
"reward_w_rps": 1.0,
|
| 47 |
+
"arch": {
|
| 48 |
+
"encoder_name": "answerdotai/ModernBERT-large",
|
| 49 |
+
"head_layers": 2,
|
| 50 |
+
"max_len": 512,
|
| 51 |
+
"head_max_len": 192
|
| 52 |
+
},
|
| 53 |
+
"init_from": "convaiinnovations/laya",
|
| 54 |
+
"log_grad_diagnostics": true,
|
| 55 |
+
"lr_scheduler": {
|
| 56 |
+
"type": "cosine",
|
| 57 |
+
"t_max": "auto",
|
| 58 |
+
"eta_min": 1e-06
|
| 59 |
+
},
|
| 60 |
+
"early_stopping": {
|
| 61 |
+
"enabled": false
|
| 62 |
+
},
|
| 63 |
+
"save_best": false,
|
| 64 |
+
"best_metric": "raw_ece",
|
| 65 |
+
"best_mode": "min",
|
| 66 |
+
"wandb": {
|
| 67 |
+
"enabled": true,
|
| 68 |
+
"project": "rlcd-reverse-engineering",
|
| 69 |
+
"entity": null,
|
| 70 |
+
"group": "e2-minimal",
|
| 71 |
+
"log_artifacts": false
|
| 72 |
+
},
|
| 73 |
+
"resume_from": null
|
| 74 |
+
}
|
| 75 |
+
}
|
e2_laya_rlce_seed43/grad_diagnostics.json
ADDED
|
@@ -0,0 +1,308 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"step": 10,
|
| 4 |
+
"epoch": 1,
|
| 5 |
+
"sigma": 0.4,
|
| 6 |
+
"grad_cos_rl_ce": 0.40505725145339966,
|
| 7 |
+
"grad_norm_rl": 0.873748242855072,
|
| 8 |
+
"grad_norm_ce": 0.09958716481924057,
|
| 9 |
+
"grad_norm_ratio_rl_ce": 8.773703362687368
|
| 10 |
+
},
|
| 11 |
+
{
|
| 12 |
+
"step": 20,
|
| 13 |
+
"epoch": 1,
|
| 14 |
+
"sigma": 0.4,
|
| 15 |
+
"grad_cos_rl_ce": 0.6861609816551208,
|
| 16 |
+
"grad_norm_rl": 0.9111436009407043,
|
| 17 |
+
"grad_norm_ce": 0.11168522387742996,
|
| 18 |
+
"grad_norm_ratio_rl_ce": 8.158139181783328
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"step": 30,
|
| 22 |
+
"epoch": 1,
|
| 23 |
+
"sigma": 0.4,
|
| 24 |
+
"grad_cos_rl_ce": 0.6419686079025269,
|
| 25 |
+
"grad_norm_rl": 0.8776994347572327,
|
| 26 |
+
"grad_norm_ce": 0.09547799080610275,
|
| 27 |
+
"grad_norm_ratio_rl_ce": 9.192688569868103
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
"step": 40,
|
| 31 |
+
"epoch": 1,
|
| 32 |
+
"sigma": 0.4,
|
| 33 |
+
"grad_cos_rl_ce": 0.6104865670204163,
|
| 34 |
+
"grad_norm_rl": 0.7692919969558716,
|
| 35 |
+
"grad_norm_ce": 0.10647637397050858,
|
| 36 |
+
"grad_norm_ratio_rl_ce": 7.22500183157014
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"step": 50,
|
| 40 |
+
"epoch": 1,
|
| 41 |
+
"sigma": 0.4,
|
| 42 |
+
"grad_cos_rl_ce": 0.5448269844055176,
|
| 43 |
+
"grad_norm_rl": 0.7688911557197571,
|
| 44 |
+
"grad_norm_ce": 0.08094635605812073,
|
| 45 |
+
"grad_norm_ratio_rl_ce": 9.49877416554343
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"step": 60,
|
| 49 |
+
"epoch": 1,
|
| 50 |
+
"sigma": 0.4,
|
| 51 |
+
"grad_cos_rl_ce": 0.7839539051055908,
|
| 52 |
+
"grad_norm_rl": 0.7688365578651428,
|
| 53 |
+
"grad_norm_ce": 0.08356863260269165,
|
| 54 |
+
"grad_norm_ratio_rl_ce": 9.200061481446083
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"step": 70,
|
| 58 |
+
"epoch": 1,
|
| 59 |
+
"sigma": 0.4,
|
| 60 |
+
"grad_cos_rl_ce": 0.5601919293403625,
|
| 61 |
+
"grad_norm_rl": 0.8111128211021423,
|
| 62 |
+
"grad_norm_ce": 0.0653119906783104,
|
| 63 |
+
"grad_norm_ratio_rl_ce": 12.419049131379586
|
| 64 |
+
},
|
| 65 |
+
{
|
| 66 |
+
"step": 80,
|
| 67 |
+
"epoch": 1,
|
| 68 |
+
"sigma": 0.4,
|
| 69 |
+
"grad_cos_rl_ce": 0.6109002828598022,
|
| 70 |
+
"grad_norm_rl": 0.7340216636657715,
|
| 71 |
+
"grad_norm_ce": 0.09491235017776489,
|
| 72 |
+
"grad_norm_ratio_rl_ce": 7.733679150194836
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"step": 90,
|
| 76 |
+
"epoch": 2,
|
| 77 |
+
"sigma": 0.30000000000000004,
|
| 78 |
+
"grad_cos_rl_ce": 0.5194111466407776,
|
| 79 |
+
"grad_norm_rl": 1.0394067764282227,
|
| 80 |
+
"grad_norm_ce": 0.05669322982430458,
|
| 81 |
+
"grad_norm_ratio_rl_ce": 18.333878306976
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"step": 100,
|
| 85 |
+
"epoch": 2,
|
| 86 |
+
"sigma": 0.30000000000000004,
|
| 87 |
+
"grad_cos_rl_ce": 0.669698178768158,
|
| 88 |
+
"grad_norm_rl": 0.7356591820716858,
|
| 89 |
+
"grad_norm_ce": 0.07953241467475891,
|
| 90 |
+
"grad_norm_ratio_rl_ce": 9.24980317874293
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"step": 110,
|
| 94 |
+
"epoch": 2,
|
| 95 |
+
"sigma": 0.30000000000000004,
|
| 96 |
+
"grad_cos_rl_ce": 0.6575043201446533,
|
| 97 |
+
"grad_norm_rl": 1.0015299320220947,
|
| 98 |
+
"grad_norm_ce": 0.07101589441299438,
|
| 99 |
+
"grad_norm_ratio_rl_ce": 14.102898235677733
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"step": 120,
|
| 103 |
+
"epoch": 2,
|
| 104 |
+
"sigma": 0.30000000000000004,
|
| 105 |
+
"grad_cos_rl_ce": 0.5552142858505249,
|
| 106 |
+
"grad_norm_rl": 1.0049232244491577,
|
| 107 |
+
"grad_norm_ce": 0.07107037305831909,
|
| 108 |
+
"grad_norm_ratio_rl_ce": 14.139833255476729
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"step": 130,
|
| 112 |
+
"epoch": 2,
|
| 113 |
+
"sigma": 0.30000000000000004,
|
| 114 |
+
"grad_cos_rl_ce": 0.5329024791717529,
|
| 115 |
+
"grad_norm_rl": 1.011047124862671,
|
| 116 |
+
"grad_norm_ce": 0.07520248740911484,
|
| 117 |
+
"grad_norm_ratio_rl_ce": 13.444330895099196
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"step": 140,
|
| 121 |
+
"epoch": 2,
|
| 122 |
+
"sigma": 0.30000000000000004,
|
| 123 |
+
"grad_cos_rl_ce": 0.8172626495361328,
|
| 124 |
+
"grad_norm_rl": 0.7989585995674133,
|
| 125 |
+
"grad_norm_ce": 0.047602083534002304,
|
| 126 |
+
"grad_norm_ratio_rl_ce": 16.784109859324012
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"step": 150,
|
| 130 |
+
"epoch": 2,
|
| 131 |
+
"sigma": 0.30000000000000004,
|
| 132 |
+
"grad_cos_rl_ce": 0.5887012481689453,
|
| 133 |
+
"grad_norm_rl": 1.1036957502365112,
|
| 134 |
+
"grad_norm_ce": 0.054837554693222046,
|
| 135 |
+
"grad_norm_ratio_rl_ce": 20.126640518727008
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"step": 160,
|
| 139 |
+
"epoch": 2,
|
| 140 |
+
"sigma": 0.30000000000000004,
|
| 141 |
+
"grad_cos_rl_ce": 0.5751376748085022,
|
| 142 |
+
"grad_norm_rl": 1.0741499662399292,
|
| 143 |
+
"grad_norm_ce": 0.06986134499311447,
|
| 144 |
+
"grad_norm_ratio_rl_ce": 15.375455000841987
|
| 145 |
+
},
|
| 146 |
+
{
|
| 147 |
+
"step": 170,
|
| 148 |
+
"epoch": 2,
|
| 149 |
+
"sigma": 0.30000000000000004,
|
| 150 |
+
"grad_cos_rl_ce": 0.648726224899292,
|
| 151 |
+
"grad_norm_rl": 0.6985398530960083,
|
| 152 |
+
"grad_norm_ce": 0.039780594408512115,
|
| 153 |
+
"grad_norm_ratio_rl_ce": 17.55981436382301
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"step": 180,
|
| 157 |
+
"epoch": 3,
|
| 158 |
+
"sigma": 0.2,
|
| 159 |
+
"grad_cos_rl_ce": 0.696439266204834,
|
| 160 |
+
"grad_norm_rl": 1.1427546739578247,
|
| 161 |
+
"grad_norm_ce": 0.04003072530031204,
|
| 162 |
+
"grad_norm_ratio_rl_ce": 28.546939017088377
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"step": 190,
|
| 166 |
+
"epoch": 3,
|
| 167 |
+
"sigma": 0.2,
|
| 168 |
+
"grad_cos_rl_ce": 0.6692054271697998,
|
| 169 |
+
"grad_norm_rl": 1.4551578760147095,
|
| 170 |
+
"grad_norm_ce": 0.03579384833574295,
|
| 171 |
+
"grad_norm_ratio_rl_ce": 40.65385376742575
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"step": 200,
|
| 175 |
+
"epoch": 3,
|
| 176 |
+
"sigma": 0.2,
|
| 177 |
+
"grad_cos_rl_ce": 0.7535935640335083,
|
| 178 |
+
"grad_norm_rl": 1.5273736715316772,
|
| 179 |
+
"grad_norm_ce": 0.05446653440594673,
|
| 180 |
+
"grad_norm_ratio_rl_ce": 28.04242436553695
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"step": 210,
|
| 184 |
+
"epoch": 3,
|
| 185 |
+
"sigma": 0.2,
|
| 186 |
+
"grad_cos_rl_ce": 0.6686179041862488,
|
| 187 |
+
"grad_norm_rl": 1.8574236631393433,
|
| 188 |
+
"grad_norm_ce": 0.046048931777477264,
|
| 189 |
+
"grad_norm_ratio_rl_ce": 40.33586863892937
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"step": 220,
|
| 193 |
+
"epoch": 3,
|
| 194 |
+
"sigma": 0.2,
|
| 195 |
+
"grad_cos_rl_ce": 0.5699312686920166,
|
| 196 |
+
"grad_norm_rl": 1.5362051725387573,
|
| 197 |
+
"grad_norm_ce": 0.05362439528107643,
|
| 198 |
+
"grad_norm_ratio_rl_ce": 28.647505757158076
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"step": 230,
|
| 202 |
+
"epoch": 3,
|
| 203 |
+
"sigma": 0.2,
|
| 204 |
+
"grad_cos_rl_ce": 0.5980252027511597,
|
| 205 |
+
"grad_norm_rl": 1.7778832912445068,
|
| 206 |
+
"grad_norm_ce": 0.05142922326922417,
|
| 207 |
+
"grad_norm_ratio_rl_ce": 34.56951472779509
|
| 208 |
+
},
|
| 209 |
+
{
|
| 210 |
+
"step": 240,
|
| 211 |
+
"epoch": 3,
|
| 212 |
+
"sigma": 0.2,
|
| 213 |
+
"grad_cos_rl_ce": 0.6783812046051025,
|
| 214 |
+
"grad_norm_rl": 1.6908835172653198,
|
| 215 |
+
"grad_norm_ce": 0.04981766641139984,
|
| 216 |
+
"grad_norm_ratio_rl_ce": 33.941443649765034
|
| 217 |
+
},
|
| 218 |
+
{
|
| 219 |
+
"step": 250,
|
| 220 |
+
"epoch": 3,
|
| 221 |
+
"sigma": 0.2,
|
| 222 |
+
"grad_cos_rl_ce": 0.6012244820594788,
|
| 223 |
+
"grad_norm_rl": 1.3999876976013184,
|
| 224 |
+
"grad_norm_ce": 0.03817203640937805,
|
| 225 |
+
"grad_norm_ratio_rl_ce": 36.67574039244528
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"step": 260,
|
| 229 |
+
"epoch": 4,
|
| 230 |
+
"sigma": 0.09999999999999998,
|
| 231 |
+
"grad_cos_rl_ce": 0.6429973840713501,
|
| 232 |
+
"grad_norm_rl": 2.5673606395721436,
|
| 233 |
+
"grad_norm_ce": 0.04673924669623375,
|
| 234 |
+
"grad_norm_ratio_rl_ce": 54.92943984009528
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"step": 270,
|
| 238 |
+
"epoch": 4,
|
| 239 |
+
"sigma": 0.09999999999999998,
|
| 240 |
+
"grad_cos_rl_ce": 0.5550851821899414,
|
| 241 |
+
"grad_norm_rl": 2.9065780639648438,
|
| 242 |
+
"grad_norm_ce": 0.06132964789867401,
|
| 243 |
+
"grad_norm_ratio_rl_ce": 47.39270750040758
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"step": 280,
|
| 247 |
+
"epoch": 4,
|
| 248 |
+
"sigma": 0.09999999999999998,
|
| 249 |
+
"grad_cos_rl_ce": 0.5599168539047241,
|
| 250 |
+
"grad_norm_rl": 3.8432469367980957,
|
| 251 |
+
"grad_norm_ce": 0.05110318213701248,
|
| 252 |
+
"grad_norm_ratio_rl_ce": 75.20562861416312
|
| 253 |
+
},
|
| 254 |
+
{
|
| 255 |
+
"step": 290,
|
| 256 |
+
"epoch": 4,
|
| 257 |
+
"sigma": 0.09999999999999998,
|
| 258 |
+
"grad_cos_rl_ce": 0.7068833112716675,
|
| 259 |
+
"grad_norm_rl": 4.333428382873535,
|
| 260 |
+
"grad_norm_ce": 0.06568123400211334,
|
| 261 |
+
"grad_norm_ratio_rl_ce": 65.97665906724748
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"step": 300,
|
| 265 |
+
"epoch": 4,
|
| 266 |
+
"sigma": 0.09999999999999998,
|
| 267 |
+
"grad_cos_rl_ce": 0.8310362100601196,
|
| 268 |
+
"grad_norm_rl": 3.0115556716918945,
|
| 269 |
+
"grad_norm_ce": 0.061776053160429,
|
| 270 |
+
"grad_norm_ratio_rl_ce": 48.7495642343976
|
| 271 |
+
},
|
| 272 |
+
{
|
| 273 |
+
"step": 310,
|
| 274 |
+
"epoch": 4,
|
| 275 |
+
"sigma": 0.09999999999999998,
|
| 276 |
+
"grad_cos_rl_ce": 0.8130363821983337,
|
| 277 |
+
"grad_norm_rl": 3.1035845279693604,
|
| 278 |
+
"grad_norm_ce": 0.04368606582283974,
|
| 279 |
+
"grad_norm_ratio_rl_ce": 71.04289364383
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"step": 320,
|
| 283 |
+
"epoch": 4,
|
| 284 |
+
"sigma": 0.09999999999999998,
|
| 285 |
+
"grad_cos_rl_ce": 0.698875904083252,
|
| 286 |
+
"grad_norm_rl": 3.0129380226135254,
|
| 287 |
+
"grad_norm_ce": 0.041882362216711044,
|
| 288 |
+
"grad_norm_ratio_rl_ce": 71.93811101254849
|
| 289 |
+
},
|
| 290 |
+
{
|
| 291 |
+
"step": 330,
|
| 292 |
+
"epoch": 4,
|
| 293 |
+
"sigma": 0.09999999999999998,
|
| 294 |
+
"grad_cos_rl_ce": 0.6491866707801819,
|
| 295 |
+
"grad_norm_rl": 3.6702404022216797,
|
| 296 |
+
"grad_norm_ce": 0.04974943399429321,
|
| 297 |
+
"grad_norm_ratio_rl_ce": 73.77451575916814
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"step": 340,
|
| 301 |
+
"epoch": 4,
|
| 302 |
+
"sigma": 0.09999999999999998,
|
| 303 |
+
"grad_cos_rl_ce": 0.8788917064666748,
|
| 304 |
+
"grad_norm_rl": 2.9684526920318604,
|
| 305 |
+
"grad_norm_ce": 0.04159684479236603,
|
| 306 |
+
"grad_norm_ratio_rl_ce": 71.36244844648985
|
| 307 |
+
}
|
| 308 |
+
]
|
e2_laya_rlce_seed43/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:283cf8cace42949590501fa7c2254e87c11fc6a34ba30493adc88035ab2dfb44
|
| 3 |
+
size 1684141044
|
e2_laya_rlce_seed43/model_config.json
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"encoder_name": "answerdotai/ModernBERT-large",
|
| 3 |
+
"head_layers": 2,
|
| 4 |
+
"head_dropout": 0.1,
|
| 5 |
+
"num_qtypes": 3,
|
| 6 |
+
"max_len": 512,
|
| 7 |
+
"head_max_len": 192,
|
| 8 |
+
"k_buckets": [
|
| 9 |
+
2,
|
| 10 |
+
5,
|
| 11 |
+
10
|
| 12 |
+
],
|
| 13 |
+
"source_checkpoint": "epoch_4.pt",
|
| 14 |
+
"epoch": 4,
|
| 15 |
+
"step": 340
|
| 16 |
+
}
|
e2_laya_rlce_seed43/test_eval.json
ADDED
|
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"raw": {
|
| 3 |
+
"ece": 0.13197172716259956,
|
| 4 |
+
"brier": 0.05452140746153333,
|
| 5 |
+
"nll": 0.8668859592825174,
|
| 6 |
+
"accuracy": 0.776,
|
| 7 |
+
"soft_accuracy": 0.5203169700633734,
|
| 8 |
+
"mean_confidence": 0.6449173926413059,
|
| 9 |
+
"score_mae": 0.22454273352865128,
|
| 10 |
+
"n": 2000,
|
| 11 |
+
"by_type": {
|
| 12 |
+
"choice": {
|
| 13 |
+
"ece": 0.11911850492159524,
|
| 14 |
+
"brier": 0.0698617118084015,
|
| 15 |
+
"nll": 0.9974608553511402,
|
| 16 |
+
"accuracy": 0.7383333333333333,
|
| 17 |
+
"soft_accuracy": 0.47561304283638794,
|
| 18 |
+
"mean_confidence": 0.620018160144488,
|
| 19 |
+
"score_mae": null,
|
| 20 |
+
"n": 600
|
| 21 |
+
},
|
| 22 |
+
"score": {
|
| 23 |
+
"ece": 0.15881713267415762,
|
| 24 |
+
"brier": 0.05666545718628185,
|
| 25 |
+
"nll": 1.0427551303524525,
|
| 26 |
+
"accuracy": 0.74,
|
| 27 |
+
"soft_accuracy": 0.4401479551522061,
|
| 28 |
+
"mean_confidence": 0.5828031701594591,
|
| 29 |
+
"score_mae": 0.22454273352865128,
|
| 30 |
+
"n": 800
|
| 31 |
+
},
|
| 32 |
+
"noul": {
|
| 33 |
+
"ece": 0.10903105984131498,
|
| 34 |
+
"brier": 0.036322370148333824,
|
| 35 |
+
"nll": 0.5018188351206482,
|
| 36 |
+
"accuracy": 0.8616666666666667,
|
| 37 |
+
"soft_accuracy": 0.6719129171719154,
|
| 38 |
+
"mean_confidence": 0.7526355884472529,
|
| 39 |
+
"score_mae": null,
|
| 40 |
+
"n": 600
|
| 41 |
+
}
|
| 42 |
+
}
|
| 43 |
+
},
|
| 44 |
+
"post_temperature": {
|
| 45 |
+
"ece": 0.14791404638439418,
|
| 46 |
+
"brier": 0.05286476505563003,
|
| 47 |
+
"nll": 0.8630265202224254,
|
| 48 |
+
"accuracy": 0.776,
|
| 49 |
+
"soft_accuracy": 0.5129262935388833,
|
| 50 |
+
"mean_confidence": 0.6292819098979234,
|
| 51 |
+
"score_mae": 0.22493662622291594,
|
| 52 |
+
"n": 2000,
|
| 53 |
+
"by_type": {
|
| 54 |
+
"choice": {
|
| 55 |
+
"ece": 0.13473159489532313,
|
| 56 |
+
"brier": 0.06779507493602827,
|
| 57 |
+
"nll": 0.9931945211129883,
|
| 58 |
+
"accuracy": 0.7383333333333333,
|
| 59 |
+
"soft_accuracy": 0.4687981816629569,
|
| 60 |
+
"mean_confidence": 0.6054597360392412,
|
| 61 |
+
"score_mae": null,
|
| 62 |
+
"n": 600
|
| 63 |
+
},
|
| 64 |
+
"score": {
|
| 65 |
+
"ece": 0.17830051325261592,
|
| 66 |
+
"brier": 0.05443568623688862,
|
| 67 |
+
"nll": 1.037335314094089,
|
| 68 |
+
"accuracy": 0.74,
|
| 69 |
+
"soft_accuracy": 0.43145765563938765,
|
| 70 |
+
"mean_confidence": 0.5632958786934614,
|
| 71 |
+
"score_mae": 0.22493662622291594,
|
| 72 |
+
"n": 800
|
| 73 |
+
},
|
| 74 |
+
"noul": {
|
| 75 |
+
"ece": 0.12058120141426723,
|
| 76 |
+
"brier": 0.03583989360022036,
|
| 77 |
+
"nll": 0.5004467941696445,
|
| 78 |
+
"accuracy": 0.8616666666666667,
|
| 79 |
+
"soft_accuracy": 0.6656792559474707,
|
| 80 |
+
"mean_confidence": 0.7410854586958885,
|
| 81 |
+
"score_mae": null,
|
| 82 |
+
"n": 600
|
| 83 |
+
}
|
| 84 |
+
}
|
| 85 |
+
},
|
| 86 |
+
"fitted_temperature": {
|
| 87 |
+
"type0_bucket1": 1.0623664855957031,
|
| 88 |
+
"type2_bucket0": 1.0938509702682495,
|
| 89 |
+
"type1_bucket1": 1.0897818803787231
|
| 90 |
+
}
|
| 91 |
+
}
|
e2_laya_rlce_seed43/train_log.json
ADDED
|
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"epoch": 1,
|
| 4 |
+
"step": 85,
|
| 5 |
+
"loss": 0.7701867953819387,
|
| 6 |
+
"sigma": 0.4,
|
| 7 |
+
"epoch_seconds": 77.68364429473877,
|
| 8 |
+
"train_rows": 5400,
|
| 9 |
+
"raw_ece": 0.0806590178112189,
|
| 10 |
+
"raw_brier": 0.11151514266849065,
|
| 11 |
+
"raw_nll": 0.9264762554628153,
|
| 12 |
+
"raw_accuracy": 0.665,
|
| 13 |
+
"raw_soft_accuracy": 0.47256603710353373,
|
| 14 |
+
"raw_mean_confidence": 0.5851941607395807,
|
| 15 |
+
"raw_score_mae": 0.3867957584404697,
|
| 16 |
+
"raw_n": 600
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 2,
|
| 20 |
+
"step": 170,
|
| 21 |
+
"loss": 0.638940218178665,
|
| 22 |
+
"sigma": 0.30000000000000004,
|
| 23 |
+
"epoch_seconds": 79.66150140762329,
|
| 24 |
+
"train_rows": 5400,
|
| 25 |
+
"raw_ece": 0.12794385274251305,
|
| 26 |
+
"raw_brier": 0.06637893431876184,
|
| 27 |
+
"raw_nll": 0.8521388599152366,
|
| 28 |
+
"raw_accuracy": 0.7516666666666667,
|
| 29 |
+
"raw_soft_accuracy": 0.5188803320129712,
|
| 30 |
+
"raw_mean_confidence": 0.6303403696417809,
|
| 31 |
+
"raw_score_mae": 0.23457328074922165,
|
| 32 |
+
"raw_n": 600
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 3,
|
| 36 |
+
"step": 255,
|
| 37 |
+
"loss": 0.636012485360398,
|
| 38 |
+
"sigma": 0.2,
|
| 39 |
+
"epoch_seconds": 80.41069412231445,
|
| 40 |
+
"train_rows": 5400,
|
| 41 |
+
"raw_ece": 0.14373067448536556,
|
| 42 |
+
"raw_brier": 0.06065273483844049,
|
| 43 |
+
"raw_nll": 0.8419932992011309,
|
| 44 |
+
"raw_accuracy": 0.78,
|
| 45 |
+
"raw_soft_accuracy": 0.5245968211442232,
|
| 46 |
+
"raw_mean_confidence": 0.6378033709526062,
|
| 47 |
+
"raw_score_mae": 0.23806003088053937,
|
| 48 |
+
"raw_n": 600
|
| 49 |
+
},
|
| 50 |
+
{
|
| 51 |
+
"epoch": 4,
|
| 52 |
+
"step": 340,
|
| 53 |
+
"loss": 0.7078079423483681,
|
| 54 |
+
"sigma": 0.09999999999999998,
|
| 55 |
+
"epoch_seconds": 80.72532320022583,
|
| 56 |
+
"train_rows": 5400,
|
| 57 |
+
"raw_ece": 0.14295961409807206,
|
| 58 |
+
"raw_brier": 0.05680996092140797,
|
| 59 |
+
"raw_nll": 0.8333720132646462,
|
| 60 |
+
"raw_accuracy": 0.7816666666666666,
|
| 61 |
+
"raw_soft_accuracy": 0.5281049428880215,
|
| 62 |
+
"raw_mean_confidence": 0.6401676322519779,
|
| 63 |
+
"raw_score_mae": 0.22018940830603242,
|
| 64 |
+
"raw_n": 600
|
| 65 |
+
}
|
| 66 |
+
]
|