E2 run e2_laya_rlce_seed42_rep2 (epoch_4.pt)
Browse files
e2_laya_rlce_seed42_rep2/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": 42,
|
| 20 |
+
"max_examples": null,
|
| 21 |
+
"ckpt_dir": "checkpoints",
|
| 22 |
+
"result_dir": "results",
|
| 23 |
+
"run_name": "e2_laya_rlce_seed42_rep2",
|
| 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_seed42_rep2/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": 0.7480509281158447,
|
| 7 |
+
"grad_norm_rl": 0.7978591918945312,
|
| 8 |
+
"grad_norm_ce": 0.11617259681224823,
|
| 9 |
+
"grad_norm_ratio_rl_ce": 6.86787774214936
|
| 10 |
+
},
|
| 11 |
+
{
|
| 12 |
+
"step": 20,
|
| 13 |
+
"epoch": 1,
|
| 14 |
+
"sigma": 0.4,
|
| 15 |
+
"grad_cos_rl_ce": 0.719328761100769,
|
| 16 |
+
"grad_norm_rl": 0.7171304821968079,
|
| 17 |
+
"grad_norm_ce": 0.12139017879962921,
|
| 18 |
+
"grad_norm_ratio_rl_ce": 5.9076482899043095
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"step": 30,
|
| 22 |
+
"epoch": 1,
|
| 23 |
+
"sigma": 0.4,
|
| 24 |
+
"grad_cos_rl_ce": 0.728217363357544,
|
| 25 |
+
"grad_norm_rl": 0.6588813066482544,
|
| 26 |
+
"grad_norm_ce": 0.08033165335655212,
|
| 27 |
+
"grad_norm_ratio_rl_ce": 8.202013516686991
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
"step": 40,
|
| 31 |
+
"epoch": 1,
|
| 32 |
+
"sigma": 0.4,
|
| 33 |
+
"grad_cos_rl_ce": 0.5422014594078064,
|
| 34 |
+
"grad_norm_rl": 0.6958211064338684,
|
| 35 |
+
"grad_norm_ce": 0.06408506631851196,
|
| 36 |
+
"grad_norm_ratio_rl_ce": 10.857773057072889
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"step": 50,
|
| 40 |
+
"epoch": 1,
|
| 41 |
+
"sigma": 0.4,
|
| 42 |
+
"grad_cos_rl_ce": 0.595352828502655,
|
| 43 |
+
"grad_norm_rl": 0.8995832204818726,
|
| 44 |
+
"grad_norm_ce": 0.12206126749515533,
|
| 45 |
+
"grad_norm_ratio_rl_ce": 7.369931829665601
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"step": 60,
|
| 49 |
+
"epoch": 1,
|
| 50 |
+
"sigma": 0.4,
|
| 51 |
+
"grad_cos_rl_ce": 0.7086445093154907,
|
| 52 |
+
"grad_norm_rl": 0.7799911499023438,
|
| 53 |
+
"grad_norm_ce": 0.09065093845129013,
|
| 54 |
+
"grad_norm_ratio_rl_ce": 8.604336184798129
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"step": 70,
|
| 58 |
+
"epoch": 1,
|
| 59 |
+
"sigma": 0.4,
|
| 60 |
+
"grad_cos_rl_ce": 0.6364695429801941,
|
| 61 |
+
"grad_norm_rl": 0.8445138931274414,
|
| 62 |
+
"grad_norm_ce": 0.0942898765206337,
|
| 63 |
+
"grad_norm_ratio_rl_ce": 8.956570146134768
|
| 64 |
+
},
|
| 65 |
+
{
|
| 66 |
+
"step": 80,
|
| 67 |
+
"epoch": 1,
|
| 68 |
+
"sigma": 0.4,
|
| 69 |
+
"grad_cos_rl_ce": 0.7536263465881348,
|
| 70 |
+
"grad_norm_rl": 0.7479110360145569,
|
| 71 |
+
"grad_norm_ce": 0.10118796676397324,
|
| 72 |
+
"grad_norm_ratio_rl_ce": 7.391304123731456
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"step": 90,
|
| 76 |
+
"epoch": 2,
|
| 77 |
+
"sigma": 0.30000000000000004,
|
| 78 |
+
"grad_cos_rl_ce": 0.731425404548645,
|
| 79 |
+
"grad_norm_rl": 1.0455303192138672,
|
| 80 |
+
"grad_norm_ce": 0.0725012794137001,
|
| 81 |
+
"grad_norm_ratio_rl_ce": 14.420853365193166
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"step": 100,
|
| 85 |
+
"epoch": 2,
|
| 86 |
+
"sigma": 0.30000000000000004,
|
| 87 |
+
"grad_cos_rl_ce": 0.7446107864379883,
|
| 88 |
+
"grad_norm_rl": 0.961202085018158,
|
| 89 |
+
"grad_norm_ce": 0.08668302744626999,
|
| 90 |
+
"grad_norm_ratio_rl_ce": 11.08869998355738
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"step": 110,
|
| 94 |
+
"epoch": 2,
|
| 95 |
+
"sigma": 0.30000000000000004,
|
| 96 |
+
"grad_cos_rl_ce": 0.6873000860214233,
|
| 97 |
+
"grad_norm_rl": 0.9747119545936584,
|
| 98 |
+
"grad_norm_ce": 0.06050889566540718,
|
| 99 |
+
"grad_norm_ratio_rl_ce": 16.10857286147596
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"step": 120,
|
| 103 |
+
"epoch": 2,
|
| 104 |
+
"sigma": 0.30000000000000004,
|
| 105 |
+
"grad_cos_rl_ce": 0.6269658803939819,
|
| 106 |
+
"grad_norm_rl": 0.8165130615234375,
|
| 107 |
+
"grad_norm_ce": 0.08500571548938751,
|
| 108 |
+
"grad_norm_ratio_rl_ce": 9.605390141388488
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"step": 130,
|
| 112 |
+
"epoch": 2,
|
| 113 |
+
"sigma": 0.30000000000000004,
|
| 114 |
+
"grad_cos_rl_ce": 0.7751697301864624,
|
| 115 |
+
"grad_norm_rl": 0.9448302984237671,
|
| 116 |
+
"grad_norm_ce": 0.07796310633420944,
|
| 117 |
+
"grad_norm_ratio_rl_ce": 12.11894116139373
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"step": 140,
|
| 121 |
+
"epoch": 2,
|
| 122 |
+
"sigma": 0.30000000000000004,
|
| 123 |
+
"grad_cos_rl_ce": 0.6338238716125488,
|
| 124 |
+
"grad_norm_rl": 1.0167694091796875,
|
| 125 |
+
"grad_norm_ce": 0.06644927710294724,
|
| 126 |
+
"grad_norm_ratio_rl_ce": 15.301436727512428
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"step": 150,
|
| 130 |
+
"epoch": 2,
|
| 131 |
+
"sigma": 0.30000000000000004,
|
| 132 |
+
"grad_cos_rl_ce": 0.5876440405845642,
|
| 133 |
+
"grad_norm_rl": 0.8838024139404297,
|
| 134 |
+
"grad_norm_ce": 0.07932641357183456,
|
| 135 |
+
"grad_norm_ratio_rl_ce": 11.14133835308332
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"step": 160,
|
| 139 |
+
"epoch": 2,
|
| 140 |
+
"sigma": 0.30000000000000004,
|
| 141 |
+
"grad_cos_rl_ce": 0.6495159268379211,
|
| 142 |
+
"grad_norm_rl": 1.2003445625305176,
|
| 143 |
+
"grad_norm_ce": 0.052610430866479874,
|
| 144 |
+
"grad_norm_ratio_rl_ce": 22.81571435095969
|
| 145 |
+
},
|
| 146 |
+
{
|
| 147 |
+
"step": 170,
|
| 148 |
+
"epoch": 2,
|
| 149 |
+
"sigma": 0.30000000000000004,
|
| 150 |
+
"grad_cos_rl_ce": 0.7499809861183167,
|
| 151 |
+
"grad_norm_rl": 0.8852800726890564,
|
| 152 |
+
"grad_norm_ce": 0.07057368010282516,
|
| 153 |
+
"grad_norm_ratio_rl_ce": 12.544054262144357
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"step": 180,
|
| 157 |
+
"epoch": 3,
|
| 158 |
+
"sigma": 0.2,
|
| 159 |
+
"grad_cos_rl_ce": 0.6737339496612549,
|
| 160 |
+
"grad_norm_rl": 1.6987617015838623,
|
| 161 |
+
"grad_norm_ce": 0.05583285167813301,
|
| 162 |
+
"grad_norm_ratio_rl_ce": 30.42584518836576
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"step": 190,
|
| 166 |
+
"epoch": 3,
|
| 167 |
+
"sigma": 0.2,
|
| 168 |
+
"grad_cos_rl_ce": 0.7250259518623352,
|
| 169 |
+
"grad_norm_rl": 1.8170832395553589,
|
| 170 |
+
"grad_norm_ce": 0.04939194768667221,
|
| 171 |
+
"grad_norm_ratio_rl_ce": 36.789058230349475
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"step": 200,
|
| 175 |
+
"epoch": 3,
|
| 176 |
+
"sigma": 0.2,
|
| 177 |
+
"grad_cos_rl_ce": 0.5851985812187195,
|
| 178 |
+
"grad_norm_rl": 1.275668740272522,
|
| 179 |
+
"grad_norm_ce": 0.070565365254879,
|
| 180 |
+
"grad_norm_ratio_rl_ce": 18.077830897138597
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"step": 210,
|
| 184 |
+
"epoch": 3,
|
| 185 |
+
"sigma": 0.2,
|
| 186 |
+
"grad_cos_rl_ce": 0.6785452961921692,
|
| 187 |
+
"grad_norm_rl": 1.3798332214355469,
|
| 188 |
+
"grad_norm_ce": 0.06643471866846085,
|
| 189 |
+
"grad_norm_ratio_rl_ce": 20.769760888452556
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"step": 220,
|
| 193 |
+
"epoch": 3,
|
| 194 |
+
"sigma": 0.2,
|
| 195 |
+
"grad_cos_rl_ce": 0.5747154951095581,
|
| 196 |
+
"grad_norm_rl": 1.4485801458358765,
|
| 197 |
+
"grad_norm_ce": 0.04342512786388397,
|
| 198 |
+
"grad_norm_ratio_rl_ce": 33.35810893583203
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"step": 230,
|
| 202 |
+
"epoch": 3,
|
| 203 |
+
"sigma": 0.2,
|
| 204 |
+
"grad_cos_rl_ce": 0.621660590171814,
|
| 205 |
+
"grad_norm_rl": 2.152005910873413,
|
| 206 |
+
"grad_norm_ce": 0.05213888734579086,
|
| 207 |
+
"grad_norm_ratio_rl_ce": 41.27448859046554
|
| 208 |
+
},
|
| 209 |
+
{
|
| 210 |
+
"step": 240,
|
| 211 |
+
"epoch": 3,
|
| 212 |
+
"sigma": 0.2,
|
| 213 |
+
"grad_cos_rl_ce": 0.6404755115509033,
|
| 214 |
+
"grad_norm_rl": 1.1945973634719849,
|
| 215 |
+
"grad_norm_ce": 0.0472736582159996,
|
| 216 |
+
"grad_norm_ratio_rl_ce": 25.269831203113398
|
| 217 |
+
},
|
| 218 |
+
{
|
| 219 |
+
"step": 250,
|
| 220 |
+
"epoch": 3,
|
| 221 |
+
"sigma": 0.2,
|
| 222 |
+
"grad_cos_rl_ce": 0.5199593901634216,
|
| 223 |
+
"grad_norm_rl": 1.2286185026168823,
|
| 224 |
+
"grad_norm_ce": 0.06010604649782181,
|
| 225 |
+
"grad_norm_ratio_rl_ce": 20.44084703959703
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"step": 260,
|
| 229 |
+
"epoch": 4,
|
| 230 |
+
"sigma": 0.09999999999999998,
|
| 231 |
+
"grad_cos_rl_ce": 0.504911482334137,
|
| 232 |
+
"grad_norm_rl": 2.6803910732269287,
|
| 233 |
+
"grad_norm_ce": 0.041781846433877945,
|
| 234 |
+
"grad_norm_ratio_rl_ce": 64.15204932287504
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"step": 270,
|
| 238 |
+
"epoch": 4,
|
| 239 |
+
"sigma": 0.09999999999999998,
|
| 240 |
+
"grad_cos_rl_ce": 0.6833255290985107,
|
| 241 |
+
"grad_norm_rl": 3.056295871734619,
|
| 242 |
+
"grad_norm_ce": 0.031291816383600235,
|
| 243 |
+
"grad_norm_ratio_rl_ce": 97.67077226416288
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"step": 280,
|
| 247 |
+
"epoch": 4,
|
| 248 |
+
"sigma": 0.09999999999999998,
|
| 249 |
+
"grad_cos_rl_ce": 0.5746022462844849,
|
| 250 |
+
"grad_norm_rl": 3.1102890968322754,
|
| 251 |
+
"grad_norm_ce": 0.03420793637633324,
|
| 252 |
+
"grad_norm_ratio_rl_ce": 90.92302624206613
|
| 253 |
+
},
|
| 254 |
+
{
|
| 255 |
+
"step": 290,
|
| 256 |
+
"epoch": 4,
|
| 257 |
+
"sigma": 0.09999999999999998,
|
| 258 |
+
"grad_cos_rl_ce": 0.7199082374572754,
|
| 259 |
+
"grad_norm_rl": 2.956169605255127,
|
| 260 |
+
"grad_norm_ce": 0.05652979016304016,
|
| 261 |
+
"grad_norm_ratio_rl_ce": 52.29401341715762
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"step": 300,
|
| 265 |
+
"epoch": 4,
|
| 266 |
+
"sigma": 0.09999999999999998,
|
| 267 |
+
"grad_cos_rl_ce": 0.49988508224487305,
|
| 268 |
+
"grad_norm_rl": 3.016805410385132,
|
| 269 |
+
"grad_norm_ce": 0.0519346259534359,
|
| 270 |
+
"grad_norm_ratio_rl_ce": 58.08851715019516
|
| 271 |
+
},
|
| 272 |
+
{
|
| 273 |
+
"step": 310,
|
| 274 |
+
"epoch": 4,
|
| 275 |
+
"sigma": 0.09999999999999998,
|
| 276 |
+
"grad_cos_rl_ce": 0.6268885135650635,
|
| 277 |
+
"grad_norm_rl": 3.6574223041534424,
|
| 278 |
+
"grad_norm_ce": 0.04034082591533661,
|
| 279 |
+
"grad_norm_ratio_rl_ce": 90.66304968146373
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"step": 320,
|
| 283 |
+
"epoch": 4,
|
| 284 |
+
"sigma": 0.09999999999999998,
|
| 285 |
+
"grad_cos_rl_ce": 0.5908229351043701,
|
| 286 |
+
"grad_norm_rl": 3.920563220977783,
|
| 287 |
+
"grad_norm_ce": 0.03276630863547325,
|
| 288 |
+
"grad_norm_ratio_rl_ce": 119.65227040354885
|
| 289 |
+
},
|
| 290 |
+
{
|
| 291 |
+
"step": 330,
|
| 292 |
+
"epoch": 4,
|
| 293 |
+
"sigma": 0.09999999999999998,
|
| 294 |
+
"grad_cos_rl_ce": 0.6410229206085205,
|
| 295 |
+
"grad_norm_rl": 2.993607759475708,
|
| 296 |
+
"grad_norm_ce": 0.04924194887280464,
|
| 297 |
+
"grad_norm_ratio_rl_ce": 60.793852152529624
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"step": 340,
|
| 301 |
+
"epoch": 4,
|
| 302 |
+
"sigma": 0.09999999999999998,
|
| 303 |
+
"grad_cos_rl_ce": 0.5430241823196411,
|
| 304 |
+
"grad_norm_rl": 3.1331257820129395,
|
| 305 |
+
"grad_norm_ce": 0.028567399829626083,
|
| 306 |
+
"grad_norm_ratio_rl_ce": 109.67486718072615
|
| 307 |
+
}
|
| 308 |
+
]
|
e2_laya_rlce_seed42_rep2/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:abb9afb2bc18464194eae51f65aaaeab31eefad6e440a4569bf6f3b70ddf0a49
|
| 3 |
+
size 1684141044
|
e2_laya_rlce_seed42_rep2/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_seed42_rep2/test_eval.json
ADDED
|
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"raw": {
|
| 3 |
+
"ece": 0.1311594566702843,
|
| 4 |
+
"brier": 0.05490199812186093,
|
| 5 |
+
"nll": 0.8674021078860388,
|
| 6 |
+
"accuracy": 0.772,
|
| 7 |
+
"soft_accuracy": 0.5175125464983285,
|
| 8 |
+
"mean_confidence": 0.6411024855226278,
|
| 9 |
+
"score_mae": 0.23364423435763457,
|
| 10 |
+
"n": 2000,
|
| 11 |
+
"by_type": {
|
| 12 |
+
"choice": {
|
| 13 |
+
"ece": 0.13066786532600722,
|
| 14 |
+
"brier": 0.06976302182632935,
|
| 15 |
+
"nll": 0.9982771100103855,
|
| 16 |
+
"accuracy": 0.7383333333333333,
|
| 17 |
+
"soft_accuracy": 0.47206922526160877,
|
| 18 |
+
"mean_confidence": 0.614280567864577,
|
| 19 |
+
"score_mae": null,
|
| 20 |
+
"n": 600
|
| 21 |
+
},
|
| 22 |
+
"score": {
|
| 23 |
+
"ece": 0.14976227495819328,
|
| 24 |
+
"brier": 0.05792711920497595,
|
| 25 |
+
"nll": 1.0437982700206339,
|
| 26 |
+
"accuracy": 0.72375,
|
| 27 |
+
"soft_accuracy": 0.43476231436245144,
|
| 28 |
+
"mean_confidence": 0.5739877213910222,
|
| 29 |
+
"score_mae": 0.23364423435763457,
|
| 30 |
+
"n": 800
|
| 31 |
+
},
|
| 32 |
+
"noul": {
|
| 33 |
+
"ece": 0.11258928567171096,
|
| 34 |
+
"brier": 0.03600747963990583,
|
| 35 |
+
"nll": 0.501332222915565,
|
| 36 |
+
"accuracy": 0.87,
|
| 37 |
+
"soft_accuracy": 0.6732895105828841,
|
| 38 |
+
"mean_confidence": 0.7574107553561529,
|
| 39 |
+
"score_mae": null,
|
| 40 |
+
"n": 600
|
| 41 |
+
}
|
| 42 |
+
}
|
| 43 |
+
},
|
| 44 |
+
"post_temperature": {
|
| 45 |
+
"ece": 0.14642349288612602,
|
| 46 |
+
"brier": 0.05350534239243169,
|
| 47 |
+
"nll": 0.8643806909918785,
|
| 48 |
+
"accuracy": 0.772,
|
| 49 |
+
"soft_accuracy": 0.5103804081380368,
|
| 50 |
+
"mean_confidence": 0.6261566658765078,
|
| 51 |
+
"score_mae": 0.23484325315104798,
|
| 52 |
+
"n": 2000,
|
| 53 |
+
"by_type": {
|
| 54 |
+
"choice": {
|
| 55 |
+
"ece": 0.14246670683224996,
|
| 56 |
+
"brier": 0.06797903129201587,
|
| 57 |
+
"nll": 0.9949558153189719,
|
| 58 |
+
"accuracy": 0.7383333333333333,
|
| 59 |
+
"soft_accuracy": 0.46516082701583705,
|
| 60 |
+
"mean_confidence": 0.5998148304224015,
|
| 61 |
+
"score_mae": null,
|
| 62 |
+
"n": 600
|
| 63 |
+
},
|
| 64 |
+
"score": {
|
| 65 |
+
"ece": 0.16657996598631142,
|
| 66 |
+
"brier": 0.056304361980874094,
|
| 67 |
+
"nll": 1.0398800939321518,
|
| 68 |
+
"accuracy": 0.72375,
|
| 69 |
+
"soft_accuracy": 0.42730253156274556,
|
| 70 |
+
"mean_confidence": 0.5571700439229608,
|
| 71 |
+
"score_mae": 0.23484325315104798,
|
| 72 |
+
"n": 800
|
| 73 |
+
},
|
| 74 |
+
"noul": {
|
| 75 |
+
"ece": 0.12551932652791342,
|
| 76 |
+
"brier": 0.0352996273749243,
|
| 77 |
+
"nll": 0.49980636274442075,
|
| 78 |
+
"accuracy": 0.87,
|
| 79 |
+
"soft_accuracy": 0.6663704913606247,
|
| 80 |
+
"mean_confidence": 0.7444806639353434,
|
| 81 |
+
"score_mae": null,
|
| 82 |
+
"n": 600
|
| 83 |
+
}
|
| 84 |
+
}
|
| 85 |
+
},
|
| 86 |
+
"fitted_temperature": {
|
| 87 |
+
"type0_bucket1": 1.0611073970794678,
|
| 88 |
+
"type2_bucket0": 1.1017639636993408,
|
| 89 |
+
"type1_bucket1": 1.077757716178894
|
| 90 |
+
}
|
| 91 |
+
}
|
e2_laya_rlce_seed42_rep2/train_log.json
ADDED
|
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"epoch": 1,
|
| 4 |
+
"step": 85,
|
| 5 |
+
"loss": 0.8195002562859479,
|
| 6 |
+
"sigma": 0.4,
|
| 7 |
+
"epoch_seconds": 79.60330843925476,
|
| 8 |
+
"train_rows": 5400,
|
| 9 |
+
"raw_ece": 0.12664776335159936,
|
| 10 |
+
"raw_brier": 0.12703849582170795,
|
| 11 |
+
"raw_nll": 0.9678852021259566,
|
| 12 |
+
"raw_accuracy": 0.615,
|
| 13 |
+
"raw_soft_accuracy": 0.43810214263697467,
|
| 14 |
+
"raw_mean_confidence": 0.5127741955717404,
|
| 15 |
+
"raw_score_mae": 0.44274310433926684,
|
| 16 |
+
"raw_n": 600
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 2,
|
| 20 |
+
"step": 170,
|
| 21 |
+
"loss": 0.675382646830643,
|
| 22 |
+
"sigma": 0.30000000000000004,
|
| 23 |
+
"epoch_seconds": 80.77236270904541,
|
| 24 |
+
"train_rows": 5400,
|
| 25 |
+
"raw_ece": 0.09774068569143615,
|
| 26 |
+
"raw_brier": 0.07887304482874527,
|
| 27 |
+
"raw_nll": 0.8794473212771118,
|
| 28 |
+
"raw_accuracy": 0.695,
|
| 29 |
+
"raw_soft_accuracy": 0.49272251504162945,
|
| 30 |
+
"raw_mean_confidence": 0.5972593071560065,
|
| 31 |
+
"raw_score_mae": 0.32232180261829246,
|
| 32 |
+
"raw_n": 600
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 3,
|
| 36 |
+
"step": 255,
|
| 37 |
+
"loss": 0.6746582365211319,
|
| 38 |
+
"sigma": 0.2,
|
| 39 |
+
"epoch_seconds": 81.5858302116394,
|
| 40 |
+
"train_rows": 5400,
|
| 41 |
+
"raw_ece": 0.15161675125360488,
|
| 42 |
+
"raw_brier": 0.06127977816908905,
|
| 43 |
+
"raw_nll": 0.8483453163752953,
|
| 44 |
+
"raw_accuracy": 0.7716666666666666,
|
| 45 |
+
"raw_soft_accuracy": 0.5127480303744475,
|
| 46 |
+
"raw_mean_confidence": 0.6210828878482183,
|
| 47 |
+
"raw_score_mae": 0.268811429804191,
|
| 48 |
+
"raw_n": 600
|
| 49 |
+
},
|
| 50 |
+
{
|
| 51 |
+
"epoch": 4,
|
| 52 |
+
"step": 340,
|
| 53 |
+
"loss": 0.6941102702828015,
|
| 54 |
+
"sigma": 0.09999999999999998,
|
| 55 |
+
"epoch_seconds": 81.88526344299316,
|
| 56 |
+
"train_rows": 5400,
|
| 57 |
+
"raw_ece": 0.15151228671272593,
|
| 58 |
+
"raw_brier": 0.059452048742792,
|
| 59 |
+
"raw_nll": 0.8438648235953102,
|
| 60 |
+
"raw_accuracy": 0.7833333333333333,
|
| 61 |
+
"raw_soft_accuracy": 0.5203279303262631,
|
| 62 |
+
"raw_mean_confidence": 0.6339086059729259,
|
| 63 |
+
"raw_score_mae": 0.25164095874254905,
|
| 64 |
+
"raw_n": 600
|
| 65 |
+
}
|
| 66 |
+
]
|