E2 run e2_ce_only_1024_seed42 (epoch_4.pt)
Browse files
e2_ce_only_1024_seed42/config.json
ADDED
|
@@ -0,0 +1,78 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"model": {
|
| 3 |
+
"encoder_name": "answerdotai/ModernBERT-large",
|
| 4 |
+
"head_layers": 2,
|
| 5 |
+
"head_dropout": 0.1,
|
| 6 |
+
"num_qtypes": 3,
|
| 7 |
+
"cls_query": false,
|
| 8 |
+
"slot_emb": false,
|
| 9 |
+
"max_len": 1024,
|
| 10 |
+
"head_max_len": 256,
|
| 11 |
+
"k_buckets": [
|
| 12 |
+
2,
|
| 13 |
+
5,
|
| 14 |
+
10
|
| 15 |
+
]
|
| 16 |
+
},
|
| 17 |
+
"training": {
|
| 18 |
+
"dataset_name": "LocalLLaMA/typed-decisions",
|
| 19 |
+
"dataset_config": "all",
|
| 20 |
+
"calib_fraction": 0.1,
|
| 21 |
+
"seed": 42,
|
| 22 |
+
"max_examples": null,
|
| 23 |
+
"augment_permute": false,
|
| 24 |
+
"ckpt_dir": "checkpoints",
|
| 25 |
+
"result_dir": "results",
|
| 26 |
+
"run_name": "e2_ce_only_1024_seed42",
|
| 27 |
+
"epochs": 4,
|
| 28 |
+
"batch_size": 4,
|
| 29 |
+
"accum_steps": 16,
|
| 30 |
+
"num_workers": 4,
|
| 31 |
+
"pin_memory": true,
|
| 32 |
+
"amp": true,
|
| 33 |
+
"amp_dtype": "bf16",
|
| 34 |
+
"gradient_checkpointing": true,
|
| 35 |
+
"lr": 2.5e-05,
|
| 36 |
+
"lr_head": 0.0001,
|
| 37 |
+
"weight_decay": 0.01,
|
| 38 |
+
"grad_clip": 1.0,
|
| 39 |
+
"log_every": 10,
|
| 40 |
+
"eval_every": 1,
|
| 41 |
+
"ckpt_every": 4,
|
| 42 |
+
"w_rl": 0.0,
|
| 43 |
+
"w_ce": 1.0,
|
| 44 |
+
"num_noise_samples": 4,
|
| 45 |
+
"sigma_start": 0.4,
|
| 46 |
+
"sigma_end": 0.1,
|
| 47 |
+
"anneal_sigma": true,
|
| 48 |
+
"reward_w_spherical": 0.75,
|
| 49 |
+
"reward_w_rps": 1.0,
|
| 50 |
+
"arch": {
|
| 51 |
+
"encoder_name": "answerdotai/ModernBERT-large",
|
| 52 |
+
"head_layers": 2,
|
| 53 |
+
"max_len": 1024,
|
| 54 |
+
"head_max_len": 256
|
| 55 |
+
},
|
| 56 |
+
"init_from": "convaiinnovations/laya",
|
| 57 |
+
"log_grad_diagnostics": true,
|
| 58 |
+
"lr_scheduler": {
|
| 59 |
+
"type": "cosine",
|
| 60 |
+
"t_max": "auto",
|
| 61 |
+
"eta_min": 1e-06
|
| 62 |
+
},
|
| 63 |
+
"early_stopping": {
|
| 64 |
+
"enabled": false
|
| 65 |
+
},
|
| 66 |
+
"save_best": false,
|
| 67 |
+
"best_metric": "raw_ece",
|
| 68 |
+
"best_mode": "min",
|
| 69 |
+
"wandb": {
|
| 70 |
+
"enabled": true,
|
| 71 |
+
"project": "rlcd-reverse-engineering",
|
| 72 |
+
"entity": null,
|
| 73 |
+
"group": "e2-acc",
|
| 74 |
+
"log_artifacts": false
|
| 75 |
+
},
|
| 76 |
+
"resume_from": null
|
| 77 |
+
}
|
| 78 |
+
}
|
e2_ce_only_1024_seed42/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": 0.5967414379119873,
|
| 7 |
+
"grad_norm_rl": 1.6385318040847778,
|
| 8 |
+
"grad_norm_ce": 0.25133898854255676,
|
| 9 |
+
"grad_norm_ratio_rl_ce": 6.519210623016179
|
| 10 |
+
},
|
| 11 |
+
{
|
| 12 |
+
"step": 20,
|
| 13 |
+
"epoch": 1,
|
| 14 |
+
"sigma": 0.4,
|
| 15 |
+
"grad_cos_rl_ce": 0.7601745128631592,
|
| 16 |
+
"grad_norm_rl": 1.4548791646957397,
|
| 17 |
+
"grad_norm_ce": 0.22606514394283295,
|
| 18 |
+
"grad_norm_ratio_rl_ce": 6.435663363758756
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"step": 30,
|
| 22 |
+
"epoch": 1,
|
| 23 |
+
"sigma": 0.4,
|
| 24 |
+
"grad_cos_rl_ce": 0.19237011671066284,
|
| 25 |
+
"grad_norm_rl": 2.208129644393921,
|
| 26 |
+
"grad_norm_ce": 0.1816539466381073,
|
| 27 |
+
"grad_norm_ratio_rl_ce": 12.15569320270799
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
"step": 40,
|
| 31 |
+
"epoch": 1,
|
| 32 |
+
"sigma": 0.4,
|
| 33 |
+
"grad_cos_rl_ce": 0.9228581786155701,
|
| 34 |
+
"grad_norm_rl": 2.157672166824341,
|
| 35 |
+
"grad_norm_ce": 0.17690712213516235,
|
| 36 |
+
"grad_norm_ratio_rl_ce": 12.196638217741253
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"step": 50,
|
| 40 |
+
"epoch": 1,
|
| 41 |
+
"sigma": 0.4,
|
| 42 |
+
"grad_cos_rl_ce": 0.7795990705490112,
|
| 43 |
+
"grad_norm_rl": 1.1552035808563232,
|
| 44 |
+
"grad_norm_ce": 0.09362342953681946,
|
| 45 |
+
"grad_norm_ratio_rl_ce": 12.338829997698538
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"step": 60,
|
| 49 |
+
"epoch": 1,
|
| 50 |
+
"sigma": 0.4,
|
| 51 |
+
"grad_cos_rl_ce": 0.8427572250366211,
|
| 52 |
+
"grad_norm_rl": 1.1276774406433105,
|
| 53 |
+
"grad_norm_ce": 0.18844491243362427,
|
| 54 |
+
"grad_norm_ratio_rl_ce": 5.9841225007361825
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"step": 70,
|
| 58 |
+
"epoch": 1,
|
| 59 |
+
"sigma": 0.4,
|
| 60 |
+
"grad_cos_rl_ce": 0.8572666049003601,
|
| 61 |
+
"grad_norm_rl": 1.3398497104644775,
|
| 62 |
+
"grad_norm_ce": 0.21438294649124146,
|
| 63 |
+
"grad_norm_ratio_rl_ce": 6.249796135343334
|
| 64 |
+
},
|
| 65 |
+
{
|
| 66 |
+
"step": 80,
|
| 67 |
+
"epoch": 1,
|
| 68 |
+
"sigma": 0.4,
|
| 69 |
+
"grad_cos_rl_ce": 0.7812180519104004,
|
| 70 |
+
"grad_norm_rl": 1.2737774848937988,
|
| 71 |
+
"grad_norm_ce": 0.2052842229604721,
|
| 72 |
+
"grad_norm_ratio_rl_ce": 6.204945838137143
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"step": 90,
|
| 76 |
+
"epoch": 2,
|
| 77 |
+
"sigma": 0.30000000000000004,
|
| 78 |
+
"grad_cos_rl_ce": 0.6589978933334351,
|
| 79 |
+
"grad_norm_rl": 2.1949641704559326,
|
| 80 |
+
"grad_norm_ce": 0.07431506365537643,
|
| 81 |
+
"grad_norm_ratio_rl_ce": 29.53592532241792
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"step": 100,
|
| 85 |
+
"epoch": 2,
|
| 86 |
+
"sigma": 0.30000000000000004,
|
| 87 |
+
"grad_cos_rl_ce": 0.6563607454299927,
|
| 88 |
+
"grad_norm_rl": 1.0808463096618652,
|
| 89 |
+
"grad_norm_ce": 0.10900399088859558,
|
| 90 |
+
"grad_norm_ratio_rl_ce": 9.915658141053875
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"step": 110,
|
| 94 |
+
"epoch": 2,
|
| 95 |
+
"sigma": 0.30000000000000004,
|
| 96 |
+
"grad_cos_rl_ce": 0.9075939655303955,
|
| 97 |
+
"grad_norm_rl": 1.4897714853286743,
|
| 98 |
+
"grad_norm_ce": 0.04449373483657837,
|
| 99 |
+
"grad_norm_ratio_rl_ce": 33.482724945444026
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"step": 120,
|
| 103 |
+
"epoch": 2,
|
| 104 |
+
"sigma": 0.30000000000000004,
|
| 105 |
+
"grad_cos_rl_ce": 0.6865151524543762,
|
| 106 |
+
"grad_norm_rl": 2.4477155208587646,
|
| 107 |
+
"grad_norm_ce": 0.07271178811788559,
|
| 108 |
+
"grad_norm_ratio_rl_ce": 33.66325576934447
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"step": 130,
|
| 112 |
+
"epoch": 2,
|
| 113 |
+
"sigma": 0.30000000000000004,
|
| 114 |
+
"grad_cos_rl_ce": 0.7356332540512085,
|
| 115 |
+
"grad_norm_rl": 1.9077789783477783,
|
| 116 |
+
"grad_norm_ce": 0.09356731921434402,
|
| 117 |
+
"grad_norm_ratio_rl_ce": 20.38937306707952
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"step": 140,
|
| 121 |
+
"epoch": 2,
|
| 122 |
+
"sigma": 0.30000000000000004,
|
| 123 |
+
"grad_cos_rl_ce": 0.7048898935317993,
|
| 124 |
+
"grad_norm_rl": 1.5347741842269897,
|
| 125 |
+
"grad_norm_ce": 0.0874747559428215,
|
| 126 |
+
"grad_norm_ratio_rl_ce": 17.545338282855067
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"step": 150,
|
| 130 |
+
"epoch": 2,
|
| 131 |
+
"sigma": 0.30000000000000004,
|
| 132 |
+
"grad_cos_rl_ce": 0.7162908315658569,
|
| 133 |
+
"grad_norm_rl": 2.219559669494629,
|
| 134 |
+
"grad_norm_ce": 0.14271360635757446,
|
| 135 |
+
"grad_norm_ratio_rl_ce": 15.552544190730043
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"step": 160,
|
| 139 |
+
"epoch": 2,
|
| 140 |
+
"sigma": 0.30000000000000004,
|
| 141 |
+
"grad_cos_rl_ce": 0.7662848234176636,
|
| 142 |
+
"grad_norm_rl": 1.9175341129302979,
|
| 143 |
+
"grad_norm_ce": 0.10590797662734985,
|
| 144 |
+
"grad_norm_ratio_rl_ce": 18.105662802693093
|
| 145 |
+
},
|
| 146 |
+
{
|
| 147 |
+
"step": 170,
|
| 148 |
+
"epoch": 2,
|
| 149 |
+
"sigma": 0.30000000000000004,
|
| 150 |
+
"grad_cos_rl_ce": 0.9436855912208557,
|
| 151 |
+
"grad_norm_rl": 2.2843737602233887,
|
| 152 |
+
"grad_norm_ce": 0.104497991502285,
|
| 153 |
+
"grad_norm_ratio_rl_ce": 21.860456142579903
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"step": 180,
|
| 157 |
+
"epoch": 3,
|
| 158 |
+
"sigma": 0.2,
|
| 159 |
+
"grad_cos_rl_ce": 0.8766736388206482,
|
| 160 |
+
"grad_norm_rl": 2.3641533851623535,
|
| 161 |
+
"grad_norm_ce": 0.12669982016086578,
|
| 162 |
+
"grad_norm_ratio_rl_ce": 18.659484931870313
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"step": 190,
|
| 166 |
+
"epoch": 3,
|
| 167 |
+
"sigma": 0.2,
|
| 168 |
+
"grad_cos_rl_ce": 0.7217068076133728,
|
| 169 |
+
"grad_norm_rl": 2.1713614463806152,
|
| 170 |
+
"grad_norm_ce": 0.04918335750699043,
|
| 171 |
+
"grad_norm_ratio_rl_ce": 44.148296424700156
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"step": 200,
|
| 175 |
+
"epoch": 3,
|
| 176 |
+
"sigma": 0.2,
|
| 177 |
+
"grad_cos_rl_ce": 0.8254374265670776,
|
| 178 |
+
"grad_norm_rl": 2.1619322299957275,
|
| 179 |
+
"grad_norm_ce": 0.10556072741746902,
|
| 180 |
+
"grad_norm_ratio_rl_ce": 20.480459758919338
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"step": 210,
|
| 184 |
+
"epoch": 3,
|
| 185 |
+
"sigma": 0.2,
|
| 186 |
+
"grad_cos_rl_ce": 0.8104074001312256,
|
| 187 |
+
"grad_norm_rl": 4.622874736785889,
|
| 188 |
+
"grad_norm_ce": 0.14989164471626282,
|
| 189 |
+
"grad_norm_ratio_rl_ce": 30.84144380119888
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"step": 220,
|
| 193 |
+
"epoch": 3,
|
| 194 |
+
"sigma": 0.2,
|
| 195 |
+
"grad_cos_rl_ce": 0.41319647431373596,
|
| 196 |
+
"grad_norm_rl": 2.2087738513946533,
|
| 197 |
+
"grad_norm_ce": 0.046874500811100006,
|
| 198 |
+
"grad_norm_ratio_rl_ce": 47.12101063850924
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"step": 230,
|
| 202 |
+
"epoch": 3,
|
| 203 |
+
"sigma": 0.2,
|
| 204 |
+
"grad_cos_rl_ce": 0.5282595157623291,
|
| 205 |
+
"grad_norm_rl": 4.212453842163086,
|
| 206 |
+
"grad_norm_ce": 0.05929712951183319,
|
| 207 |
+
"grad_norm_ratio_rl_ce": 71.03975988116692
|
| 208 |
+
},
|
| 209 |
+
{
|
| 210 |
+
"step": 240,
|
| 211 |
+
"epoch": 3,
|
| 212 |
+
"sigma": 0.2,
|
| 213 |
+
"grad_cos_rl_ce": 0.7456092834472656,
|
| 214 |
+
"grad_norm_rl": 2.7060587406158447,
|
| 215 |
+
"grad_norm_ce": 0.0518919974565506,
|
| 216 |
+
"grad_norm_ratio_rl_ce": 52.14790089515517
|
| 217 |
+
},
|
| 218 |
+
{
|
| 219 |
+
"step": 250,
|
| 220 |
+
"epoch": 3,
|
| 221 |
+
"sigma": 0.2,
|
| 222 |
+
"grad_cos_rl_ce": 0.7305554747581482,
|
| 223 |
+
"grad_norm_rl": 3.0288920402526855,
|
| 224 |
+
"grad_norm_ce": 0.0771079733967781,
|
| 225 |
+
"grad_norm_ratio_rl_ce": 39.28117815607439
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"step": 260,
|
| 229 |
+
"epoch": 4,
|
| 230 |
+
"sigma": 0.09999999999999998,
|
| 231 |
+
"grad_cos_rl_ce": 0.630805253982544,
|
| 232 |
+
"grad_norm_rl": 5.210536003112793,
|
| 233 |
+
"grad_norm_ce": 0.07777045667171478,
|
| 234 |
+
"grad_norm_ratio_rl_ce": 66.9989122618573
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"step": 270,
|
| 238 |
+
"epoch": 4,
|
| 239 |
+
"sigma": 0.09999999999999998,
|
| 240 |
+
"grad_cos_rl_ce": 0.8264482021331787,
|
| 241 |
+
"grad_norm_rl": 5.325096607208252,
|
| 242 |
+
"grad_norm_ce": 0.06894725561141968,
|
| 243 |
+
"grad_norm_ratio_rl_ce": 77.23435196927926
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"step": 280,
|
| 247 |
+
"epoch": 4,
|
| 248 |
+
"sigma": 0.09999999999999998,
|
| 249 |
+
"grad_cos_rl_ce": 0.606040894985199,
|
| 250 |
+
"grad_norm_rl": 5.729742527008057,
|
| 251 |
+
"grad_norm_ce": 0.06906485557556152,
|
| 252 |
+
"grad_norm_ratio_rl_ce": 82.96176802598738
|
| 253 |
+
},
|
| 254 |
+
{
|
| 255 |
+
"step": 290,
|
| 256 |
+
"epoch": 4,
|
| 257 |
+
"sigma": 0.09999999999999998,
|
| 258 |
+
"grad_cos_rl_ce": 0.7670713663101196,
|
| 259 |
+
"grad_norm_rl": 3.494049549102783,
|
| 260 |
+
"grad_norm_ce": 0.0696273222565651,
|
| 261 |
+
"grad_norm_ratio_rl_ce": 50.182161770171085
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"step": 300,
|
| 265 |
+
"epoch": 4,
|
| 266 |
+
"sigma": 0.09999999999999998,
|
| 267 |
+
"grad_cos_rl_ce": 0.7328579425811768,
|
| 268 |
+
"grad_norm_rl": 5.24454402923584,
|
| 269 |
+
"grad_norm_ce": 0.06282012164592743,
|
| 270 |
+
"grad_norm_ratio_rl_ce": 83.48509827465192
|
| 271 |
+
},
|
| 272 |
+
{
|
| 273 |
+
"step": 310,
|
| 274 |
+
"epoch": 4,
|
| 275 |
+
"sigma": 0.09999999999999998,
|
| 276 |
+
"grad_cos_rl_ce": 0.7307541370391846,
|
| 277 |
+
"grad_norm_rl": 5.608753681182861,
|
| 278 |
+
"grad_norm_ce": 0.037729181349277496,
|
| 279 |
+
"grad_norm_ratio_rl_ce": 148.65823960662925
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"step": 320,
|
| 283 |
+
"epoch": 4,
|
| 284 |
+
"sigma": 0.09999999999999998,
|
| 285 |
+
"grad_cos_rl_ce": 0.7958859801292419,
|
| 286 |
+
"grad_norm_rl": 6.438984394073486,
|
| 287 |
+
"grad_norm_ce": 0.026616930961608887,
|
| 288 |
+
"grad_norm_ratio_rl_ce": 241.9131042328208
|
| 289 |
+
},
|
| 290 |
+
{
|
| 291 |
+
"step": 330,
|
| 292 |
+
"epoch": 4,
|
| 293 |
+
"sigma": 0.09999999999999998,
|
| 294 |
+
"grad_cos_rl_ce": 0.8907221555709839,
|
| 295 |
+
"grad_norm_rl": 4.68951940536499,
|
| 296 |
+
"grad_norm_ce": 0.12277086824178696,
|
| 297 |
+
"grad_norm_ratio_rl_ce": 38.19733029931314
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"step": 340,
|
| 301 |
+
"epoch": 4,
|
| 302 |
+
"sigma": 0.09999999999999998,
|
| 303 |
+
"grad_cos_rl_ce": 0.5864734649658203,
|
| 304 |
+
"grad_norm_rl": 7.235905170440674,
|
| 305 |
+
"grad_norm_ce": 0.054542697966098785,
|
| 306 |
+
"grad_norm_ratio_rl_ce": 132.66496598569762
|
| 307 |
+
}
|
| 308 |
+
]
|
e2_ce_only_1024_seed42/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:be8aae6b150491cfe3d8049d69fce35dc06dfe4aac1db1d71a321b19432a9f9f
|
| 3 |
+
size 1684141044
|
e2_ce_only_1024_seed42/model_config.json
ADDED
|
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"encoder_name": "answerdotai/ModernBERT-large",
|
| 3 |
+
"head_layers": 2,
|
| 4 |
+
"head_dropout": 0.1,
|
| 5 |
+
"num_qtypes": 3,
|
| 6 |
+
"cls_query": false,
|
| 7 |
+
"slot_emb": false,
|
| 8 |
+
"max_len": 1024,
|
| 9 |
+
"head_max_len": 256,
|
| 10 |
+
"k_buckets": [
|
| 11 |
+
2,
|
| 12 |
+
5,
|
| 13 |
+
10
|
| 14 |
+
],
|
| 15 |
+
"source_checkpoint": "epoch_4.pt",
|
| 16 |
+
"epoch": 4,
|
| 17 |
+
"step": 340
|
| 18 |
+
}
|
e2_ce_only_1024_seed42/test_eval.json
ADDED
|
@@ -0,0 +1,99 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"raw": {
|
| 3 |
+
"ece": 0.14405169022083278,
|
| 4 |
+
"brier": 0.049473914223986594,
|
| 5 |
+
"nll": 0.8580627283966169,
|
| 6 |
+
"accuracy": 0.7885,
|
| 7 |
+
"soft_accuracy": 0.5236424660123885,
|
| 8 |
+
"mean_confidence": 0.6444483368843794,
|
| 9 |
+
"mean_target_max": 0.6589462240487337,
|
| 10 |
+
"score_mae": 0.21151374208275228,
|
| 11 |
+
"n": 2000,
|
| 12 |
+
"by_type": {
|
| 13 |
+
"choice": {
|
| 14 |
+
"ece": 0.1381368541220824,
|
| 15 |
+
"brier": 0.0610927829263998,
|
| 16 |
+
"nll": 0.9818144658269982,
|
| 17 |
+
"accuracy": 0.755,
|
| 18 |
+
"soft_accuracy": 0.48078609759608903,
|
| 19 |
+
"mean_confidence": 0.6193696513772011,
|
| 20 |
+
"mean_target_max": 0.6345381752649943,
|
| 21 |
+
"score_mae": null,
|
| 22 |
+
"n": 600
|
| 23 |
+
},
|
| 24 |
+
"score": {
|
| 25 |
+
"ece": 0.1779335317760706,
|
| 26 |
+
"brier": 0.05114634511056351,
|
| 27 |
+
"nll": 1.03313795868773,
|
| 28 |
+
"accuracy": 0.75875,
|
| 29 |
+
"soft_accuracy": 0.4435949847009033,
|
| 30 |
+
"mean_confidence": 0.5816603348776698,
|
| 31 |
+
"mean_target_max": 0.5928848401084542,
|
| 32 |
+
"score_mae": 0.21151374208275228,
|
| 33 |
+
"n": 800
|
| 34 |
+
},
|
| 35 |
+
"noul": {
|
| 36 |
+
"ece": 0.10842230906089145,
|
| 37 |
+
"brier": 0.03562513767280417,
|
| 38 |
+
"nll": 0.5008773505780846,
|
| 39 |
+
"accuracy": 0.8616666666666667,
|
| 40 |
+
"soft_accuracy": 0.6732288095106681,
|
| 41 |
+
"mean_confidence": 0.7532443584005037,
|
| 42 |
+
"mean_target_max": 0.7714361180861791,
|
| 43 |
+
"score_mae": null,
|
| 44 |
+
"n": 600
|
| 45 |
+
}
|
| 46 |
+
}
|
| 47 |
+
},
|
| 48 |
+
"post_temperature": {
|
| 49 |
+
"ece": 0.16220938038825988,
|
| 50 |
+
"brier": 0.04794188590972476,
|
| 51 |
+
"nll": 0.8542706172754988,
|
| 52 |
+
"accuracy": 0.7885,
|
| 53 |
+
"soft_accuracy": 0.5147140698879957,
|
| 54 |
+
"mean_confidence": 0.6262906277328729,
|
| 55 |
+
"mean_target_max": 0.6589462240487337,
|
| 56 |
+
"score_mae": 0.21261174298007973,
|
| 57 |
+
"n": 2000,
|
| 58 |
+
"by_type": {
|
| 59 |
+
"choice": {
|
| 60 |
+
"ece": 0.16171381801366805,
|
| 61 |
+
"brier": 0.058598528968896064,
|
| 62 |
+
"nll": 0.9764881309494376,
|
| 63 |
+
"accuracy": 0.755,
|
| 64 |
+
"soft_accuracy": 0.46792497878273326,
|
| 65 |
+
"mean_confidence": 0.5932861614724,
|
| 66 |
+
"mean_target_max": 0.6345381752649943,
|
| 67 |
+
"score_mae": null,
|
| 68 |
+
"n": 600
|
| 69 |
+
},
|
| 70 |
+
"score": {
|
| 71 |
+
"ece": 0.19320389255881307,
|
| 72 |
+
"brier": 0.04962007356784852,
|
| 73 |
+
"nll": 1.0289345176471398,
|
| 74 |
+
"accuracy": 0.75875,
|
| 75 |
+
"soft_accuracy": 0.4367176624760032,
|
| 76 |
+
"mean_confidence": 0.5665761560201645,
|
| 77 |
+
"mean_target_max": 0.5928848401084542,
|
| 78 |
+
"score_mae": 0.21261174298007973,
|
| 79 |
+
"n": 800
|
| 80 |
+
},
|
| 81 |
+
"noul": {
|
| 82 |
+
"ece": 0.12275225609540938,
|
| 83 |
+
"brier": 0.035047659306388444,
|
| 84 |
+
"nll": 0.49916790310603876,
|
| 85 |
+
"accuracy": 0.8616666666666667,
|
| 86 |
+
"soft_accuracy": 0.6654983708759149,
|
| 87 |
+
"mean_confidence": 0.7389143896102905,
|
| 88 |
+
"mean_target_max": 0.7714361180861791,
|
| 89 |
+
"score_mae": null,
|
| 90 |
+
"n": 600
|
| 91 |
+
}
|
| 92 |
+
}
|
| 93 |
+
},
|
| 94 |
+
"fitted_temperature": {
|
| 95 |
+
"type0_bucket1": 1.116391658782959,
|
| 96 |
+
"type2_bucket0": 1.1198104619979858,
|
| 97 |
+
"type1_bucket1": 1.0700936317443848
|
| 98 |
+
}
|
| 99 |
+
}
|
e2_ce_only_1024_seed42/train_log.json
ADDED
|
@@ -0,0 +1,70 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"epoch": 1,
|
| 4 |
+
"step": 85,
|
| 5 |
+
"loss": 0.9641271437353948,
|
| 6 |
+
"sigma": 0.4,
|
| 7 |
+
"epoch_seconds": 228.8010711669922,
|
| 8 |
+
"train_rows": 5400,
|
| 9 |
+
"raw_ece": 0.10835536271333696,
|
| 10 |
+
"raw_brier": 0.07556118707742442,
|
| 11 |
+
"raw_nll": 0.8722028755644957,
|
| 12 |
+
"raw_accuracy": 0.7066666666666667,
|
| 13 |
+
"raw_soft_accuracy": 0.4981412665049235,
|
| 14 |
+
"raw_mean_confidence": 0.5995895755290985,
|
| 15 |
+
"raw_mean_target_max": 0.6725333379705747,
|
| 16 |
+
"raw_score_mae": 0.29329004988539964,
|
| 17 |
+
"raw_n": 600
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 2,
|
| 21 |
+
"step": 170,
|
| 22 |
+
"loss": 0.8357182100488275,
|
| 23 |
+
"sigma": 0.30000000000000004,
|
| 24 |
+
"epoch_seconds": 224.5660684108734,
|
| 25 |
+
"train_rows": 5400,
|
| 26 |
+
"raw_ece": 0.16046804825464883,
|
| 27 |
+
"raw_brier": 0.05988674723912672,
|
| 28 |
+
"raw_nll": 0.8413767837701986,
|
| 29 |
+
"raw_accuracy": 0.7716666666666666,
|
| 30 |
+
"raw_soft_accuracy": 0.5137750238180161,
|
| 31 |
+
"raw_mean_confidence": 0.615313962996006,
|
| 32 |
+
"raw_mean_target_max": 0.6725333379705747,
|
| 33 |
+
"raw_score_mae": 0.24656671059007446,
|
| 34 |
+
"raw_n": 600
|
| 35 |
+
},
|
| 36 |
+
{
|
| 37 |
+
"epoch": 3,
|
| 38 |
+
"step": 255,
|
| 39 |
+
"loss": 0.8000329575373554,
|
| 40 |
+
"sigma": 0.2,
|
| 41 |
+
"epoch_seconds": 223.96768260002136,
|
| 42 |
+
"train_rows": 5400,
|
| 43 |
+
"raw_ece": 0.1488726359605789,
|
| 44 |
+
"raw_brier": 0.05451012486705442,
|
| 45 |
+
"raw_nll": 0.8328132717187205,
|
| 46 |
+
"raw_accuracy": 0.78,
|
| 47 |
+
"raw_soft_accuracy": 0.5247553873838235,
|
| 48 |
+
"raw_mean_confidence": 0.6352111848195394,
|
| 49 |
+
"raw_mean_target_max": 0.6725333379705747,
|
| 50 |
+
"raw_score_mae": 0.22168685695699727,
|
| 51 |
+
"raw_n": 600
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 4,
|
| 55 |
+
"step": 340,
|
| 56 |
+
"loss": 0.7846576284726753,
|
| 57 |
+
"sigma": 0.09999999999999998,
|
| 58 |
+
"epoch_seconds": 222.44736337661743,
|
| 59 |
+
"train_rows": 5400,
|
| 60 |
+
"raw_ece": 0.13711758901675544,
|
| 61 |
+
"raw_brier": 0.052573612098253716,
|
| 62 |
+
"raw_nll": 0.8299004402291029,
|
| 63 |
+
"raw_accuracy": 0.7766666666666666,
|
| 64 |
+
"raw_soft_accuracy": 0.5293207626231015,
|
| 65 |
+
"raw_mean_confidence": 0.6415973689158757,
|
| 66 |
+
"raw_mean_target_max": 0.6725333379705747,
|
| 67 |
+
"raw_score_mae": 0.21637691374926363,
|
| 68 |
+
"raw_n": 600
|
| 69 |
+
}
|
| 70 |
+
]
|