E2 run e2_ce_only_seed44 (epoch_4.pt)
Browse files
e2_ce_only_seed44/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": 44,
|
| 20 |
+
"max_examples": null,
|
| 21 |
+
"ckpt_dir": "checkpoints",
|
| 22 |
+
"result_dir": "results",
|
| 23 |
+
"run_name": "e2_ce_only_seed44",
|
| 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": 0.0,
|
| 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_ce_only_seed44/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": 0.5117641687393188,
|
| 7 |
+
"grad_norm_rl": 0.735811710357666,
|
| 8 |
+
"grad_norm_ce": 0.07813874632120132,
|
| 9 |
+
"grad_norm_ratio_rl_ce": 9.416732990992701
|
| 10 |
+
},
|
| 11 |
+
{
|
| 12 |
+
"step": 20,
|
| 13 |
+
"epoch": 1,
|
| 14 |
+
"sigma": 0.4,
|
| 15 |
+
"grad_cos_rl_ce": 0.725143313407898,
|
| 16 |
+
"grad_norm_rl": 0.7441243529319763,
|
| 17 |
+
"grad_norm_ce": 0.0966627299785614,
|
| 18 |
+
"grad_norm_ratio_rl_ce": 7.6981516360753925
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"step": 30,
|
| 22 |
+
"epoch": 1,
|
| 23 |
+
"sigma": 0.4,
|
| 24 |
+
"grad_cos_rl_ce": 0.7613257169723511,
|
| 25 |
+
"grad_norm_rl": 0.6756245493888855,
|
| 26 |
+
"grad_norm_ce": 0.09423620998859406,
|
| 27 |
+
"grad_norm_ratio_rl_ce": 7.169479221104713
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
"step": 40,
|
| 31 |
+
"epoch": 1,
|
| 32 |
+
"sigma": 0.4,
|
| 33 |
+
"grad_cos_rl_ce": 0.7106607556343079,
|
| 34 |
+
"grad_norm_rl": 0.8623901605606079,
|
| 35 |
+
"grad_norm_ce": 0.07792498171329498,
|
| 36 |
+
"grad_norm_ratio_rl_ce": 11.066927981241648
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"step": 50,
|
| 40 |
+
"epoch": 1,
|
| 41 |
+
"sigma": 0.4,
|
| 42 |
+
"grad_cos_rl_ce": 0.8195016384124756,
|
| 43 |
+
"grad_norm_rl": 0.5761241316795349,
|
| 44 |
+
"grad_norm_ce": 0.08626532554626465,
|
| 45 |
+
"grad_norm_ratio_rl_ce": 6.67851338914331
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"step": 60,
|
| 49 |
+
"epoch": 1,
|
| 50 |
+
"sigma": 0.4,
|
| 51 |
+
"grad_cos_rl_ce": 0.7564661502838135,
|
| 52 |
+
"grad_norm_rl": 0.69732666015625,
|
| 53 |
+
"grad_norm_ce": 0.06389430165290833,
|
| 54 |
+
"grad_norm_ratio_rl_ce": 10.913753529138216
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"step": 70,
|
| 58 |
+
"epoch": 1,
|
| 59 |
+
"sigma": 0.4,
|
| 60 |
+
"grad_cos_rl_ce": 0.5296084880828857,
|
| 61 |
+
"grad_norm_rl": 0.6317656636238098,
|
| 62 |
+
"grad_norm_ce": 0.07818718254566193,
|
| 63 |
+
"grad_norm_ratio_rl_ce": 8.080169192116031
|
| 64 |
+
},
|
| 65 |
+
{
|
| 66 |
+
"step": 80,
|
| 67 |
+
"epoch": 1,
|
| 68 |
+
"sigma": 0.4,
|
| 69 |
+
"grad_cos_rl_ce": 0.639915943145752,
|
| 70 |
+
"grad_norm_rl": 0.6242314577102661,
|
| 71 |
+
"grad_norm_ce": 0.07085512578487396,
|
| 72 |
+
"grad_norm_ratio_rl_ce": 8.809968944314909
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"step": 90,
|
| 76 |
+
"epoch": 2,
|
| 77 |
+
"sigma": 0.30000000000000004,
|
| 78 |
+
"grad_cos_rl_ce": 0.7255890965461731,
|
| 79 |
+
"grad_norm_rl": 1.122153401374817,
|
| 80 |
+
"grad_norm_ce": 0.05347577482461929,
|
| 81 |
+
"grad_norm_ratio_rl_ce": 20.984331784907535
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"step": 100,
|
| 85 |
+
"epoch": 2,
|
| 86 |
+
"sigma": 0.30000000000000004,
|
| 87 |
+
"grad_cos_rl_ce": 0.509003758430481,
|
| 88 |
+
"grad_norm_rl": 1.2309033870697021,
|
| 89 |
+
"grad_norm_ce": 0.04964178800582886,
|
| 90 |
+
"grad_norm_ratio_rl_ce": 24.795710157038894
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"step": 110,
|
| 94 |
+
"epoch": 2,
|
| 95 |
+
"sigma": 0.30000000000000004,
|
| 96 |
+
"grad_cos_rl_ce": 0.6269993782043457,
|
| 97 |
+
"grad_norm_rl": 0.9351693987846375,
|
| 98 |
+
"grad_norm_ce": 0.0534854494035244,
|
| 99 |
+
"grad_norm_ratio_rl_ce": 17.484557187305132
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"step": 120,
|
| 103 |
+
"epoch": 2,
|
| 104 |
+
"sigma": 0.30000000000000004,
|
| 105 |
+
"grad_cos_rl_ce": 0.7408747673034668,
|
| 106 |
+
"grad_norm_rl": 0.978773832321167,
|
| 107 |
+
"grad_norm_ce": 0.05384577810764313,
|
| 108 |
+
"grad_norm_ratio_rl_ce": 18.17735515613684
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"step": 130,
|
| 112 |
+
"epoch": 2,
|
| 113 |
+
"sigma": 0.30000000000000004,
|
| 114 |
+
"grad_cos_rl_ce": 0.33224549889564514,
|
| 115 |
+
"grad_norm_rl": 1.7547924518585205,
|
| 116 |
+
"grad_norm_ce": 0.057347238063812256,
|
| 117 |
+
"grad_norm_ratio_rl_ce": 30.599423984567526
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"step": 140,
|
| 121 |
+
"epoch": 2,
|
| 122 |
+
"sigma": 0.30000000000000004,
|
| 123 |
+
"grad_cos_rl_ce": 0.6154404878616333,
|
| 124 |
+
"grad_norm_rl": 1.0050312280654907,
|
| 125 |
+
"grad_norm_ce": 0.0485650934278965,
|
| 126 |
+
"grad_norm_ratio_rl_ce": 20.69451857552046
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"step": 150,
|
| 130 |
+
"epoch": 2,
|
| 131 |
+
"sigma": 0.30000000000000004,
|
| 132 |
+
"grad_cos_rl_ce": 0.6641345024108887,
|
| 133 |
+
"grad_norm_rl": 1.2433751821517944,
|
| 134 |
+
"grad_norm_ce": 0.05375421419739723,
|
| 135 |
+
"grad_norm_ratio_rl_ce": 23.130747992815014
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"step": 160,
|
| 139 |
+
"epoch": 2,
|
| 140 |
+
"sigma": 0.30000000000000004,
|
| 141 |
+
"grad_cos_rl_ce": 0.6251810789108276,
|
| 142 |
+
"grad_norm_rl": 1.2009940147399902,
|
| 143 |
+
"grad_norm_ce": 0.03321168199181557,
|
| 144 |
+
"grad_norm_ratio_rl_ce": 36.16179436608944
|
| 145 |
+
},
|
| 146 |
+
{
|
| 147 |
+
"step": 170,
|
| 148 |
+
"epoch": 2,
|
| 149 |
+
"sigma": 0.30000000000000004,
|
| 150 |
+
"grad_cos_rl_ce": 0.6767799854278564,
|
| 151 |
+
"grad_norm_rl": 0.9019886255264282,
|
| 152 |
+
"grad_norm_ce": 0.04452957585453987,
|
| 153 |
+
"grad_norm_ratio_rl_ce": 20.255944688825704
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"step": 180,
|
| 157 |
+
"epoch": 3,
|
| 158 |
+
"sigma": 0.2,
|
| 159 |
+
"grad_cos_rl_ce": 0.710254430770874,
|
| 160 |
+
"grad_norm_rl": 1.1840680837631226,
|
| 161 |
+
"grad_norm_ce": 0.051588814705610275,
|
| 162 |
+
"grad_norm_ratio_rl_ce": 22.952031181952226
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"step": 190,
|
| 166 |
+
"epoch": 3,
|
| 167 |
+
"sigma": 0.2,
|
| 168 |
+
"grad_cos_rl_ce": 0.6331205368041992,
|
| 169 |
+
"grad_norm_rl": 1.3247076272964478,
|
| 170 |
+
"grad_norm_ce": 0.02023432031273842,
|
| 171 |
+
"grad_norm_ratio_rl_ce": 65.46835311599196
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"step": 200,
|
| 175 |
+
"epoch": 3,
|
| 176 |
+
"sigma": 0.2,
|
| 177 |
+
"grad_cos_rl_ce": 0.6824506521224976,
|
| 178 |
+
"grad_norm_rl": 1.2579418420791626,
|
| 179 |
+
"grad_norm_ce": 0.038087308406829834,
|
| 180 |
+
"grad_norm_ratio_rl_ce": 33.02784824389396
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"step": 210,
|
| 184 |
+
"epoch": 3,
|
| 185 |
+
"sigma": 0.2,
|
| 186 |
+
"grad_cos_rl_ce": 0.6946102380752563,
|
| 187 |
+
"grad_norm_rl": 1.5324783325195312,
|
| 188 |
+
"grad_norm_ce": 0.0319567546248436,
|
| 189 |
+
"grad_norm_ratio_rl_ce": 47.954754808805355
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"step": 220,
|
| 193 |
+
"epoch": 3,
|
| 194 |
+
"sigma": 0.2,
|
| 195 |
+
"grad_cos_rl_ce": 0.7953137755393982,
|
| 196 |
+
"grad_norm_rl": 1.5889081954956055,
|
| 197 |
+
"grad_norm_ce": 0.03365301713347435,
|
| 198 |
+
"grad_norm_ratio_rl_ce": 47.21443516323334
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"step": 230,
|
| 202 |
+
"epoch": 3,
|
| 203 |
+
"sigma": 0.2,
|
| 204 |
+
"grad_cos_rl_ce": 0.7947266101837158,
|
| 205 |
+
"grad_norm_rl": 1.5063623189926147,
|
| 206 |
+
"grad_norm_ce": 0.039484571665525436,
|
| 207 |
+
"grad_norm_ratio_rl_ce": 38.150656204480036
|
| 208 |
+
},
|
| 209 |
+
{
|
| 210 |
+
"step": 240,
|
| 211 |
+
"epoch": 3,
|
| 212 |
+
"sigma": 0.2,
|
| 213 |
+
"grad_cos_rl_ce": 0.6106650233268738,
|
| 214 |
+
"grad_norm_rl": 1.2682620286941528,
|
| 215 |
+
"grad_norm_ce": 0.027609480544924736,
|
| 216 |
+
"grad_norm_ratio_rl_ce": 45.93574394239332
|
| 217 |
+
},
|
| 218 |
+
{
|
| 219 |
+
"step": 250,
|
| 220 |
+
"epoch": 3,
|
| 221 |
+
"sigma": 0.2,
|
| 222 |
+
"grad_cos_rl_ce": 0.6316173672676086,
|
| 223 |
+
"grad_norm_rl": 1.2026762962341309,
|
| 224 |
+
"grad_norm_ce": 0.03109697997570038,
|
| 225 |
+
"grad_norm_ratio_rl_ce": 38.67501915536233
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"step": 260,
|
| 229 |
+
"epoch": 4,
|
| 230 |
+
"sigma": 0.09999999999999998,
|
| 231 |
+
"grad_cos_rl_ce": 0.5648887157440186,
|
| 232 |
+
"grad_norm_rl": 3.8588666915893555,
|
| 233 |
+
"grad_norm_ce": 0.02946075052022934,
|
| 234 |
+
"grad_norm_ratio_rl_ce": 130.98331249027922
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"step": 270,
|
| 238 |
+
"epoch": 4,
|
| 239 |
+
"sigma": 0.09999999999999998,
|
| 240 |
+
"grad_cos_rl_ce": 0.7327464818954468,
|
| 241 |
+
"grad_norm_rl": 3.701770544052124,
|
| 242 |
+
"grad_norm_ce": 0.033336542546749115,
|
| 243 |
+
"grad_norm_ratio_rl_ce": 111.04242555630924
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"step": 280,
|
| 247 |
+
"epoch": 4,
|
| 248 |
+
"sigma": 0.09999999999999998,
|
| 249 |
+
"grad_cos_rl_ce": 0.6509130001068115,
|
| 250 |
+
"grad_norm_rl": 3.5156800746917725,
|
| 251 |
+
"grad_norm_ce": 0.021772071719169617,
|
| 252 |
+
"grad_norm_ratio_rl_ce": 161.4765980949956
|
| 253 |
+
},
|
| 254 |
+
{
|
| 255 |
+
"step": 290,
|
| 256 |
+
"epoch": 4,
|
| 257 |
+
"sigma": 0.09999999999999998,
|
| 258 |
+
"grad_cos_rl_ce": 0.6432994604110718,
|
| 259 |
+
"grad_norm_rl": 3.26535701751709,
|
| 260 |
+
"grad_norm_ce": 0.03170677274465561,
|
| 261 |
+
"grad_norm_ratio_rl_ce": 102.98610469800929
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"step": 300,
|
| 265 |
+
"epoch": 4,
|
| 266 |
+
"sigma": 0.09999999999999998,
|
| 267 |
+
"grad_cos_rl_ce": 0.6993526220321655,
|
| 268 |
+
"grad_norm_rl": 3.1620919704437256,
|
| 269 |
+
"grad_norm_ce": 0.03667442500591278,
|
| 270 |
+
"grad_norm_ratio_rl_ce": 86.22062840614194
|
| 271 |
+
},
|
| 272 |
+
{
|
| 273 |
+
"step": 310,
|
| 274 |
+
"epoch": 4,
|
| 275 |
+
"sigma": 0.09999999999999998,
|
| 276 |
+
"grad_cos_rl_ce": 0.6097292900085449,
|
| 277 |
+
"grad_norm_rl": 2.6546144485473633,
|
| 278 |
+
"grad_norm_ce": 0.030032813549041748,
|
| 279 |
+
"grad_norm_ratio_rl_ce": 88.3904681195633
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"step": 320,
|
| 283 |
+
"epoch": 4,
|
| 284 |
+
"sigma": 0.09999999999999998,
|
| 285 |
+
"grad_cos_rl_ce": 0.8419539928436279,
|
| 286 |
+
"grad_norm_rl": 2.519258975982666,
|
| 287 |
+
"grad_norm_ce": 0.0354139469563961,
|
| 288 |
+
"grad_norm_ratio_rl_ce": 71.13748092198075
|
| 289 |
+
},
|
| 290 |
+
{
|
| 291 |
+
"step": 330,
|
| 292 |
+
"epoch": 4,
|
| 293 |
+
"sigma": 0.09999999999999998,
|
| 294 |
+
"grad_cos_rl_ce": 0.6992432475090027,
|
| 295 |
+
"grad_norm_rl": 2.9592032432556152,
|
| 296 |
+
"grad_norm_ce": 0.01707499474287033,
|
| 297 |
+
"grad_norm_ratio_rl_ce": 173.30624622834694
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"step": 340,
|
| 301 |
+
"epoch": 4,
|
| 302 |
+
"sigma": 0.09999999999999998,
|
| 303 |
+
"grad_cos_rl_ce": 0.6954779028892517,
|
| 304 |
+
"grad_norm_rl": 3.39968204498291,
|
| 305 |
+
"grad_norm_ce": 0.03562990948557854,
|
| 306 |
+
"grad_norm_ratio_rl_ce": 95.41652207561616
|
| 307 |
+
}
|
| 308 |
+
]
|
e2_ce_only_seed44/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:6a84d5bdc2e919c4ae1b342b6d47b6b50495234ed4b4df3622fe8e0d20e03ae4
|
| 3 |
+
size 1684141044
|
e2_ce_only_seed44/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_ce_only_seed44/test_eval.json
ADDED
|
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"raw": {
|
| 3 |
+
"ece": 0.13912512767314913,
|
| 4 |
+
"brier": 0.051336294564260095,
|
| 5 |
+
"nll": 0.8615856033712626,
|
| 6 |
+
"accuracy": 0.784,
|
| 7 |
+
"soft_accuracy": 0.5222822944968939,
|
| 8 |
+
"mean_confidence": 0.6454715538471937,
|
| 9 |
+
"score_mae": 0.21565650733129588,
|
| 10 |
+
"n": 2000,
|
| 11 |
+
"by_type": {
|
| 12 |
+
"choice": {
|
| 13 |
+
"ece": 0.1563871059815089,
|
| 14 |
+
"brier": 0.06268110662646298,
|
| 15 |
+
"nll": 0.9869649931974709,
|
| 16 |
+
"accuracy": 0.7683333333333333,
|
| 17 |
+
"soft_accuracy": 0.47667451376716297,
|
| 18 |
+
"mean_confidence": 0.6139350893100103,
|
| 19 |
+
"score_mae": null,
|
| 20 |
+
"n": 600
|
| 21 |
+
},
|
| 22 |
+
"score": {
|
| 23 |
+
"ece": 0.15901374589651823,
|
| 24 |
+
"brier": 0.053362719049971474,
|
| 25 |
+
"nll": 1.0366679836343975,
|
| 26 |
+
"accuracy": 0.7425,
|
| 27 |
+
"soft_accuracy": 0.4430929691158235,
|
| 28 |
+
"mean_confidence": 0.5840299716964364,
|
| 29 |
+
"score_mae": 0.21565650733129588,
|
| 30 |
+
"n": 800
|
| 31 |
+
},
|
| 32 |
+
"noul": {
|
| 33 |
+
"ece": 0.09853743195533753,
|
| 34 |
+
"brier": 0.037289583187775366,
|
| 35 |
+
"nll": 0.5027630398608744,
|
| 36 |
+
"accuracy": 0.855,
|
| 37 |
+
"soft_accuracy": 0.6734758424013854,
|
| 38 |
+
"mean_confidence": 0.7589301279187203,
|
| 39 |
+
"score_mae": null,
|
| 40 |
+
"n": 600
|
| 41 |
+
}
|
| 42 |
+
}
|
| 43 |
+
},
|
| 44 |
+
"post_temperature": {
|
| 45 |
+
"ece": 0.16221648167073727,
|
| 46 |
+
"brier": 0.04951586158953618,
|
| 47 |
+
"nll": 0.8574903107769787,
|
| 48 |
+
"accuracy": 0.784,
|
| 49 |
+
"soft_accuracy": 0.5113936349526048,
|
| 50 |
+
"mean_confidence": 0.623023711964488,
|
| 51 |
+
"score_mae": 0.21575187296402873,
|
| 52 |
+
"n": 2000,
|
| 53 |
+
"by_type": {
|
| 54 |
+
"choice": {
|
| 55 |
+
"ece": 0.19459551319479942,
|
| 56 |
+
"brier": 0.06047852645175832,
|
| 57 |
+
"nll": 0.982100979530563,
|
| 58 |
+
"accuracy": 0.7683333333333333,
|
| 59 |
+
"soft_accuracy": 0.4581708333144585,
|
| 60 |
+
"mean_confidence": 0.576762926876545,
|
| 61 |
+
"score_mae": null,
|
| 62 |
+
"n": 600
|
| 63 |
+
},
|
| 64 |
+
"score": {
|
| 65 |
+
"ece": 0.17873684860765934,
|
| 66 |
+
"brier": 0.05110300580230927,
|
| 67 |
+
"nll": 1.0314594799187033,
|
| 68 |
+
"accuracy": 0.7425,
|
| 69 |
+
"soft_accuracy": 0.434347476400435,
|
| 70 |
+
"mean_confidence": 0.5645947919413448,
|
| 71 |
+
"score_mae": 0.21575187296402873,
|
| 72 |
+
"n": 800
|
| 73 |
+
},
|
| 74 |
+
"noul": {
|
| 75 |
+
"ece": 0.11204376672705013,
|
| 76 |
+
"brier": 0.03643700444361659,
|
| 77 |
+
"nll": 0.5009207498344282,
|
| 78 |
+
"accuracy": 0.855,
|
| 79 |
+
"soft_accuracy": 0.6673446479936441,
|
| 80 |
+
"mean_confidence": 0.7471897237499555,
|
| 81 |
+
"score_mae": null,
|
| 82 |
+
"n": 600
|
| 83 |
+
}
|
| 84 |
+
}
|
| 85 |
+
},
|
| 86 |
+
"fitted_temperature": {
|
| 87 |
+
"type1_bucket1": 1.089984655380249,
|
| 88 |
+
"type0_bucket1": 1.1709468364715576,
|
| 89 |
+
"type2_bucket0": 1.0927082300186157
|
| 90 |
+
}
|
| 91 |
+
}
|
e2_ce_only_seed44/train_log.json
ADDED
|
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"epoch": 1,
|
| 4 |
+
"step": 85,
|
| 5 |
+
"loss": 0.9567923987613005,
|
| 6 |
+
"sigma": 0.4,
|
| 7 |
+
"epoch_seconds": 78.75261068344116,
|
| 8 |
+
"train_rows": 5400,
|
| 9 |
+
"raw_ece": 0.1341254517932733,
|
| 10 |
+
"raw_brier": 0.08077335126956568,
|
| 11 |
+
"raw_nll": 0.896267338246107,
|
| 12 |
+
"raw_accuracy": 0.7466666666666667,
|
| 13 |
+
"raw_soft_accuracy": 0.5014536376111209,
|
| 14 |
+
"raw_mean_confidence": 0.6154130641122659,
|
| 15 |
+
"raw_score_mae": 0.29270818730971465,
|
| 16 |
+
"raw_n": 600
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 2,
|
| 20 |
+
"step": 170,
|
| 21 |
+
"loss": 0.8347657154588138,
|
| 22 |
+
"sigma": 0.30000000000000004,
|
| 23 |
+
"epoch_seconds": 80.42655491828918,
|
| 24 |
+
"train_rows": 5400,
|
| 25 |
+
"raw_ece": 0.12616657406091691,
|
| 26 |
+
"raw_brier": 0.06522280850509928,
|
| 27 |
+
"raw_nll": 0.8595432463148609,
|
| 28 |
+
"raw_accuracy": 0.7716666666666666,
|
| 29 |
+
"raw_soft_accuracy": 0.5229155146082243,
|
| 30 |
+
"raw_mean_confidence": 0.6475945883989334,
|
| 31 |
+
"raw_score_mae": 0.24447941673764337,
|
| 32 |
+
"raw_n": 600
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 3,
|
| 36 |
+
"step": 255,
|
| 37 |
+
"loss": 0.7985654967672685,
|
| 38 |
+
"sigma": 0.2,
|
| 39 |
+
"epoch_seconds": 81.44757080078125,
|
| 40 |
+
"train_rows": 5400,
|
| 41 |
+
"raw_ece": 0.1285483098526796,
|
| 42 |
+
"raw_brier": 0.06085678737842301,
|
| 43 |
+
"raw_nll": 0.848166642251114,
|
| 44 |
+
"raw_accuracy": 0.7616666666666667,
|
| 45 |
+
"raw_soft_accuracy": 0.5228428899807235,
|
| 46 |
+
"raw_mean_confidence": 0.6414403301974138,
|
| 47 |
+
"raw_score_mae": 0.22014502747527634,
|
| 48 |
+
"raw_n": 600
|
| 49 |
+
},
|
| 50 |
+
{
|
| 51 |
+
"epoch": 4,
|
| 52 |
+
"step": 340,
|
| 53 |
+
"loss": 0.7815637390403186,
|
| 54 |
+
"sigma": 0.09999999999999998,
|
| 55 |
+
"epoch_seconds": 80.90966749191284,
|
| 56 |
+
"train_rows": 5400,
|
| 57 |
+
"raw_ece": 0.12941327060262364,
|
| 58 |
+
"raw_brier": 0.05823764107604537,
|
| 59 |
+
"raw_nll": 0.8443934684588263,
|
| 60 |
+
"raw_accuracy": 0.7683333333333333,
|
| 61 |
+
"raw_soft_accuracy": 0.5261963731391976,
|
| 62 |
+
"raw_mean_confidence": 0.6471907341480255,
|
| 63 |
+
"raw_score_mae": 0.21377778615084633,
|
| 64 |
+
"raw_n": 600
|
| 65 |
+
}
|
| 66 |
+
]
|