E2 run e2_ce_only_seed43 (epoch_4.pt)
Browse files
e2_ce_only_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_ce_only_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": 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_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": 0.48673197627067566,
|
| 7 |
+
"grad_norm_rl": 0.8834173679351807,
|
| 8 |
+
"grad_norm_ce": 0.09846552461385727,
|
| 9 |
+
"grad_norm_ratio_rl_ce": 8.97184442371676
|
| 10 |
+
},
|
| 11 |
+
{
|
| 12 |
+
"step": 20,
|
| 13 |
+
"epoch": 1,
|
| 14 |
+
"sigma": 0.4,
|
| 15 |
+
"grad_cos_rl_ce": 0.6551669239997864,
|
| 16 |
+
"grad_norm_rl": 0.9000421762466431,
|
| 17 |
+
"grad_norm_ce": 0.1086537316441536,
|
| 18 |
+
"grad_norm_ratio_rl_ce": 8.28358274149595
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"step": 30,
|
| 22 |
+
"epoch": 1,
|
| 23 |
+
"sigma": 0.4,
|
| 24 |
+
"grad_cos_rl_ce": 0.6432710289955139,
|
| 25 |
+
"grad_norm_rl": 0.8741858601570129,
|
| 26 |
+
"grad_norm_ce": 0.08157411217689514,
|
| 27 |
+
"grad_norm_ratio_rl_ce": 10.716461838546559
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
"step": 40,
|
| 31 |
+
"epoch": 1,
|
| 32 |
+
"sigma": 0.4,
|
| 33 |
+
"grad_cos_rl_ce": 0.567652702331543,
|
| 34 |
+
"grad_norm_rl": 0.6574223041534424,
|
| 35 |
+
"grad_norm_ce": 0.08753374218940735,
|
| 36 |
+
"grad_norm_ratio_rl_ce": 7.510501524439549
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"step": 50,
|
| 40 |
+
"epoch": 1,
|
| 41 |
+
"sigma": 0.4,
|
| 42 |
+
"grad_cos_rl_ce": 0.4406024217605591,
|
| 43 |
+
"grad_norm_rl": 0.7521747946739197,
|
| 44 |
+
"grad_norm_ce": 0.07011300325393677,
|
| 45 |
+
"grad_norm_ratio_rl_ce": 10.72803559633121
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"step": 60,
|
| 49 |
+
"epoch": 1,
|
| 50 |
+
"sigma": 0.4,
|
| 51 |
+
"grad_cos_rl_ce": 0.7988536357879639,
|
| 52 |
+
"grad_norm_rl": 0.7631130814552307,
|
| 53 |
+
"grad_norm_ce": 0.08107815682888031,
|
| 54 |
+
"grad_norm_ratio_rl_ce": 9.41206745814191
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"step": 70,
|
| 58 |
+
"epoch": 1,
|
| 59 |
+
"sigma": 0.4,
|
| 60 |
+
"grad_cos_rl_ce": 0.614159107208252,
|
| 61 |
+
"grad_norm_rl": 0.7331306338310242,
|
| 62 |
+
"grad_norm_ce": 0.057552460581064224,
|
| 63 |
+
"grad_norm_ratio_rl_ce": 12.738475930119261
|
| 64 |
+
},
|
| 65 |
+
{
|
| 66 |
+
"step": 80,
|
| 67 |
+
"epoch": 1,
|
| 68 |
+
"sigma": 0.4,
|
| 69 |
+
"grad_cos_rl_ce": 0.455304890871048,
|
| 70 |
+
"grad_norm_rl": 0.7800648808479309,
|
| 71 |
+
"grad_norm_ce": 0.07327602803707123,
|
| 72 |
+
"grad_norm_ratio_rl_ce": 10.645567203141615
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"step": 90,
|
| 76 |
+
"epoch": 2,
|
| 77 |
+
"sigma": 0.30000000000000004,
|
| 78 |
+
"grad_cos_rl_ce": 0.704264223575592,
|
| 79 |
+
"grad_norm_rl": 1.0962364673614502,
|
| 80 |
+
"grad_norm_ce": 0.05384136363863945,
|
| 81 |
+
"grad_norm_ratio_rl_ce": 20.360488540352126
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"step": 100,
|
| 85 |
+
"epoch": 2,
|
| 86 |
+
"sigma": 0.30000000000000004,
|
| 87 |
+
"grad_cos_rl_ce": 0.5469024777412415,
|
| 88 |
+
"grad_norm_rl": 0.8740054368972778,
|
| 89 |
+
"grad_norm_ce": 0.07204676419496536,
|
| 90 |
+
"grad_norm_ratio_rl_ce": 12.131085228645889
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"step": 110,
|
| 94 |
+
"epoch": 2,
|
| 95 |
+
"sigma": 0.30000000000000004,
|
| 96 |
+
"grad_cos_rl_ce": 0.6979126930236816,
|
| 97 |
+
"grad_norm_rl": 1.0111037492752075,
|
| 98 |
+
"grad_norm_ce": 0.06368736177682877,
|
| 99 |
+
"grad_norm_ratio_rl_ce": 15.876050146625404
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"step": 120,
|
| 103 |
+
"epoch": 2,
|
| 104 |
+
"sigma": 0.30000000000000004,
|
| 105 |
+
"grad_cos_rl_ce": 0.5335148572921753,
|
| 106 |
+
"grad_norm_rl": 0.9705780148506165,
|
| 107 |
+
"grad_norm_ce": 0.05654781311750412,
|
| 108 |
+
"grad_norm_ratio_rl_ce": 17.163847040977405
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"step": 130,
|
| 112 |
+
"epoch": 2,
|
| 113 |
+
"sigma": 0.30000000000000004,
|
| 114 |
+
"grad_cos_rl_ce": 0.6628378629684448,
|
| 115 |
+
"grad_norm_rl": 1.0449274778366089,
|
| 116 |
+
"grad_norm_ce": 0.06829623878002167,
|
| 117 |
+
"grad_norm_ratio_rl_ce": 15.299927148290864
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"step": 140,
|
| 121 |
+
"epoch": 2,
|
| 122 |
+
"sigma": 0.30000000000000004,
|
| 123 |
+
"grad_cos_rl_ce": 0.6861889958381653,
|
| 124 |
+
"grad_norm_rl": 0.9444466829299927,
|
| 125 |
+
"grad_norm_ce": 0.052546482533216476,
|
| 126 |
+
"grad_norm_ratio_rl_ce": 17.973547179546696
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"step": 150,
|
| 130 |
+
"epoch": 2,
|
| 131 |
+
"sigma": 0.30000000000000004,
|
| 132 |
+
"grad_cos_rl_ce": 0.44960469007492065,
|
| 133 |
+
"grad_norm_rl": 1.1912264823913574,
|
| 134 |
+
"grad_norm_ce": 0.059847861528396606,
|
| 135 |
+
"grad_norm_ratio_rl_ce": 19.90424472938176
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"step": 160,
|
| 139 |
+
"epoch": 2,
|
| 140 |
+
"sigma": 0.30000000000000004,
|
| 141 |
+
"grad_cos_rl_ce": 0.6168105602264404,
|
| 142 |
+
"grad_norm_rl": 0.9877853989601135,
|
| 143 |
+
"grad_norm_ce": 0.062387511134147644,
|
| 144 |
+
"grad_norm_ratio_rl_ce": 15.83306307629656
|
| 145 |
+
},
|
| 146 |
+
{
|
| 147 |
+
"step": 170,
|
| 148 |
+
"epoch": 2,
|
| 149 |
+
"sigma": 0.30000000000000004,
|
| 150 |
+
"grad_cos_rl_ce": 0.4975142478942871,
|
| 151 |
+
"grad_norm_rl": 0.7779608368873596,
|
| 152 |
+
"grad_norm_ce": 0.043242745101451874,
|
| 153 |
+
"grad_norm_ratio_rl_ce": 17.99055159569968
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"step": 180,
|
| 157 |
+
"epoch": 3,
|
| 158 |
+
"sigma": 0.2,
|
| 159 |
+
"grad_cos_rl_ce": 0.75034499168396,
|
| 160 |
+
"grad_norm_rl": 1.1504311561584473,
|
| 161 |
+
"grad_norm_ce": 0.043043866753578186,
|
| 162 |
+
"grad_norm_ratio_rl_ce": 26.72694724998918
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"step": 190,
|
| 166 |
+
"epoch": 3,
|
| 167 |
+
"sigma": 0.2,
|
| 168 |
+
"grad_cos_rl_ce": 0.6848476529121399,
|
| 169 |
+
"grad_norm_rl": 1.2937568426132202,
|
| 170 |
+
"grad_norm_ce": 0.024341853335499763,
|
| 171 |
+
"grad_norm_ratio_rl_ce": 53.14947981904181
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"step": 200,
|
| 175 |
+
"epoch": 3,
|
| 176 |
+
"sigma": 0.2,
|
| 177 |
+
"grad_cos_rl_ce": 0.7542049884796143,
|
| 178 |
+
"grad_norm_rl": 1.7007938623428345,
|
| 179 |
+
"grad_norm_ce": 0.0534014068543911,
|
| 180 |
+
"grad_norm_ratio_rl_ce": 31.849233241745228
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"step": 210,
|
| 184 |
+
"epoch": 3,
|
| 185 |
+
"sigma": 0.2,
|
| 186 |
+
"grad_cos_rl_ce": 0.6652767062187195,
|
| 187 |
+
"grad_norm_rl": 1.57689368724823,
|
| 188 |
+
"grad_norm_ce": 0.03696513548493385,
|
| 189 |
+
"grad_norm_ratio_rl_ce": 42.65894515362823
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"step": 220,
|
| 193 |
+
"epoch": 3,
|
| 194 |
+
"sigma": 0.2,
|
| 195 |
+
"grad_cos_rl_ce": 0.525071382522583,
|
| 196 |
+
"grad_norm_rl": 1.4646810293197632,
|
| 197 |
+
"grad_norm_ce": 0.04280499741435051,
|
| 198 |
+
"grad_norm_ratio_rl_ce": 34.21752406948457
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"step": 230,
|
| 202 |
+
"epoch": 3,
|
| 203 |
+
"sigma": 0.2,
|
| 204 |
+
"grad_cos_rl_ce": 0.5999466180801392,
|
| 205 |
+
"grad_norm_rl": 1.8325450420379639,
|
| 206 |
+
"grad_norm_ce": 0.04666627570986748,
|
| 207 |
+
"grad_norm_ratio_rl_ce": 39.26915131242146
|
| 208 |
+
},
|
| 209 |
+
{
|
| 210 |
+
"step": 240,
|
| 211 |
+
"epoch": 3,
|
| 212 |
+
"sigma": 0.2,
|
| 213 |
+
"grad_cos_rl_ce": 0.6475627422332764,
|
| 214 |
+
"grad_norm_rl": 1.8137919902801514,
|
| 215 |
+
"grad_norm_ce": 0.04089006781578064,
|
| 216 |
+
"grad_norm_ratio_rl_ce": 44.357764297474645
|
| 217 |
+
},
|
| 218 |
+
{
|
| 219 |
+
"step": 250,
|
| 220 |
+
"epoch": 3,
|
| 221 |
+
"sigma": 0.2,
|
| 222 |
+
"grad_cos_rl_ce": 0.5253803730010986,
|
| 223 |
+
"grad_norm_rl": 1.2708572149276733,
|
| 224 |
+
"grad_norm_ce": 0.03538006544113159,
|
| 225 |
+
"grad_norm_ratio_rl_ce": 35.920148792325875
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"step": 260,
|
| 229 |
+
"epoch": 4,
|
| 230 |
+
"sigma": 0.09999999999999998,
|
| 231 |
+
"grad_cos_rl_ce": 0.6210740208625793,
|
| 232 |
+
"grad_norm_rl": 2.9749128818511963,
|
| 233 |
+
"grad_norm_ce": 0.03343786671757698,
|
| 234 |
+
"grad_norm_ratio_rl_ce": 88.96838147534696
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"step": 270,
|
| 238 |
+
"epoch": 4,
|
| 239 |
+
"sigma": 0.09999999999999998,
|
| 240 |
+
"grad_cos_rl_ce": 0.582107424736023,
|
| 241 |
+
"grad_norm_rl": 2.7636027336120605,
|
| 242 |
+
"grad_norm_ce": 0.03333034738898277,
|
| 243 |
+
"grad_norm_ratio_rl_ce": 82.91550944127151
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"step": 280,
|
| 247 |
+
"epoch": 4,
|
| 248 |
+
"sigma": 0.09999999999999998,
|
| 249 |
+
"grad_cos_rl_ce": 0.39451855421066284,
|
| 250 |
+
"grad_norm_rl": 2.9472157955169678,
|
| 251 |
+
"grad_norm_ce": 0.033847078680992126,
|
| 252 |
+
"grad_norm_ratio_rl_ce": 87.07445104182855
|
| 253 |
+
},
|
| 254 |
+
{
|
| 255 |
+
"step": 290,
|
| 256 |
+
"epoch": 4,
|
| 257 |
+
"sigma": 0.09999999999999998,
|
| 258 |
+
"grad_cos_rl_ce": 0.6753247380256653,
|
| 259 |
+
"grad_norm_rl": 4.012372016906738,
|
| 260 |
+
"grad_norm_ce": 0.04615291953086853,
|
| 261 |
+
"grad_norm_ratio_rl_ce": 86.93647244186009
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"step": 300,
|
| 265 |
+
"epoch": 4,
|
| 266 |
+
"sigma": 0.09999999999999998,
|
| 267 |
+
"grad_cos_rl_ce": 0.7951096296310425,
|
| 268 |
+
"grad_norm_rl": 2.9608678817749023,
|
| 269 |
+
"grad_norm_ce": 0.03368086740374565,
|
| 270 |
+
"grad_norm_ratio_rl_ce": 87.90949016490069
|
| 271 |
+
},
|
| 272 |
+
{
|
| 273 |
+
"step": 310,
|
| 274 |
+
"epoch": 4,
|
| 275 |
+
"sigma": 0.09999999999999998,
|
| 276 |
+
"grad_cos_rl_ce": 0.8498627543449402,
|
| 277 |
+
"grad_norm_rl": 3.019796133041382,
|
| 278 |
+
"grad_norm_ce": 0.05439849942922592,
|
| 279 |
+
"grad_norm_ratio_rl_ce": 55.5124895856774
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"step": 320,
|
| 283 |
+
"epoch": 4,
|
| 284 |
+
"sigma": 0.09999999999999998,
|
| 285 |
+
"grad_cos_rl_ce": 0.6834268569946289,
|
| 286 |
+
"grad_norm_rl": 3.246634006500244,
|
| 287 |
+
"grad_norm_ce": 0.05395307019352913,
|
| 288 |
+
"grad_norm_ratio_rl_ce": 60.17514841796028
|
| 289 |
+
},
|
| 290 |
+
{
|
| 291 |
+
"step": 330,
|
| 292 |
+
"epoch": 4,
|
| 293 |
+
"sigma": 0.09999999999999998,
|
| 294 |
+
"grad_cos_rl_ce": 0.6813265681266785,
|
| 295 |
+
"grad_norm_rl": 3.6014928817749023,
|
| 296 |
+
"grad_norm_ce": 0.03977849334478378,
|
| 297 |
+
"grad_norm_ratio_rl_ce": 90.53869513253376
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"step": 340,
|
| 301 |
+
"epoch": 4,
|
| 302 |
+
"sigma": 0.09999999999999998,
|
| 303 |
+
"grad_cos_rl_ce": 0.9079393148422241,
|
| 304 |
+
"grad_norm_rl": 2.9363937377929688,
|
| 305 |
+
"grad_norm_ce": 0.032255467027425766,
|
| 306 |
+
"grad_norm_ratio_rl_ce": 91.03553624866909
|
| 307 |
+
}
|
| 308 |
+
]
|
e2_ce_only_seed43/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:034a268aca5e2c24b0110932f159cd3009aef01477ab254e2a2e5f8c08dc5bfb
|
| 3 |
+
size 1684141044
|
e2_ce_only_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_ce_only_seed43/test_eval.json
ADDED
|
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"raw": {
|
| 3 |
+
"ece": 0.14004284296929834,
|
| 4 |
+
"brier": 0.05231113771935242,
|
| 5 |
+
"nll": 0.8615606835968792,
|
| 6 |
+
"accuracy": 0.7785,
|
| 7 |
+
"soft_accuracy": 0.5190770915709436,
|
| 8 |
+
"mean_confidence": 0.6384571585655212,
|
| 9 |
+
"score_mae": 0.22389980303356424,
|
| 10 |
+
"n": 2000,
|
| 11 |
+
"by_type": {
|
| 12 |
+
"choice": {
|
| 13 |
+
"ece": 0.15449575329820314,
|
| 14 |
+
"brier": 0.06391233971989853,
|
| 15 |
+
"nll": 0.9867078996635974,
|
| 16 |
+
"accuracy": 0.7666666666666667,
|
| 17 |
+
"soft_accuracy": 0.47517100428541503,
|
| 18 |
+
"mean_confidence": 0.6121709233522415,
|
| 19 |
+
"score_mae": null,
|
| 20 |
+
"n": 600
|
| 21 |
+
},
|
| 22 |
+
"score": {
|
| 23 |
+
"ece": 0.1481409678608179,
|
| 24 |
+
"brier": 0.05756362313889895,
|
| 25 |
+
"nll": 1.0408445472083985,
|
| 26 |
+
"accuracy": 0.72375,
|
| 27 |
+
"soft_accuracy": 0.4375115679483861,
|
| 28 |
+
"mean_confidence": 0.5756090462207795,
|
| 29 |
+
"score_mae": 0.22389980303356424,
|
| 30 |
+
"n": 800
|
| 31 |
+
},
|
| 32 |
+
"noul": {
|
| 33 |
+
"ece": 0.11479244967301687,
|
| 34 |
+
"brier": 0.03370662182607759,
|
| 35 |
+
"nll": 0.49736831604813536,
|
| 36 |
+
"accuracy": 0.8633333333333333,
|
| 37 |
+
"soft_accuracy": 0.6717372103532155,
|
| 38 |
+
"mean_confidence": 0.7485408769051234,
|
| 39 |
+
"score_mae": null,
|
| 40 |
+
"n": 600
|
| 41 |
+
}
|
| 42 |
+
}
|
| 43 |
+
},
|
| 44 |
+
"post_temperature": {
|
| 45 |
+
"ece": 0.15468624928593636,
|
| 46 |
+
"brier": 0.0509832072927122,
|
| 47 |
+
"nll": 0.8585635291822254,
|
| 48 |
+
"accuracy": 0.7785,
|
| 49 |
+
"soft_accuracy": 0.5120872111618519,
|
| 50 |
+
"mean_confidence": 0.6238137583583593,
|
| 51 |
+
"score_mae": 0.2241187854518648,
|
| 52 |
+
"n": 2000,
|
| 53 |
+
"by_type": {
|
| 54 |
+
"choice": {
|
| 55 |
+
"ece": 0.17406129206220305,
|
| 56 |
+
"brier": 0.062145273851042473,
|
| 57 |
+
"nll": 0.9828057477561136,
|
| 58 |
+
"accuracy": 0.7666666666666667,
|
| 59 |
+
"soft_accuracy": 0.46572454484800496,
|
| 60 |
+
"mean_confidence": 0.5926053783297539,
|
| 61 |
+
"score_mae": null,
|
| 62 |
+
"n": 600
|
| 63 |
+
},
|
| 64 |
+
"score": {
|
| 65 |
+
"ece": 0.16359852660447363,
|
| 66 |
+
"brier": 0.05573118854731717,
|
| 67 |
+
"nll": 1.0368128461157904,
|
| 68 |
+
"accuracy": 0.72375,
|
| 69 |
+
"soft_accuracy": 0.4306440041400492,
|
| 70 |
+
"mean_confidence": 0.5601514636352658,
|
| 71 |
+
"score_mae": 0.2241187854518648,
|
| 72 |
+
"n": 800
|
| 73 |
+
},
|
| 74 |
+
"noul": {
|
| 75 |
+
"ece": 0.12369074026743572,
|
| 76 |
+
"brier": 0.033490499061575266,
|
| 77 |
+
"nll": 0.4966555546969175,
|
| 78 |
+
"accuracy": 0.8633333333333333,
|
| 79 |
+
"soft_accuracy": 0.6670408201714356,
|
| 80 |
+
"mean_confidence": 0.7399051980177561,
|
| 81 |
+
"score_mae": null,
|
| 82 |
+
"n": 600
|
| 83 |
+
}
|
| 84 |
+
}
|
| 85 |
+
},
|
| 86 |
+
"fitted_temperature": {
|
| 87 |
+
"type0_bucket1": 1.0869410037994385,
|
| 88 |
+
"type2_bucket0": 1.0710346698760986,
|
| 89 |
+
"type1_bucket1": 1.0724668502807617
|
| 90 |
+
}
|
| 91 |
+
}
|
e2_ce_only_seed43/train_log.json
ADDED
|
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"epoch": 1,
|
| 4 |
+
"step": 85,
|
| 5 |
+
"loss": 0.96591477709658,
|
| 6 |
+
"sigma": 0.4,
|
| 7 |
+
"epoch_seconds": 78.8194420337677,
|
| 8 |
+
"train_rows": 5400,
|
| 9 |
+
"raw_ece": 0.10246505772074065,
|
| 10 |
+
"raw_brier": 0.0989133737875334,
|
| 11 |
+
"raw_nll": 0.9018456795439124,
|
| 12 |
+
"raw_accuracy": 0.6983333333333334,
|
| 13 |
+
"raw_soft_accuracy": 0.4882410797228416,
|
| 14 |
+
"raw_mean_confidence": 0.5958682804803054,
|
| 15 |
+
"raw_score_mae": 0.36452924299131456,
|
| 16 |
+
"raw_n": 600
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 2,
|
| 20 |
+
"step": 170,
|
| 21 |
+
"loss": 0.8375356882810593,
|
| 22 |
+
"sigma": 0.30000000000000004,
|
| 23 |
+
"epoch_seconds": 80.36995697021484,
|
| 24 |
+
"train_rows": 5400,
|
| 25 |
+
"raw_ece": 0.1311313817401727,
|
| 26 |
+
"raw_brier": 0.06277170015816694,
|
| 27 |
+
"raw_nll": 0.8413281134702265,
|
| 28 |
+
"raw_accuracy": 0.76,
|
| 29 |
+
"raw_soft_accuracy": 0.5222958381722371,
|
| 30 |
+
"raw_mean_confidence": 0.6319663000106811,
|
| 31 |
+
"raw_score_mae": 0.23855322396072248,
|
| 32 |
+
"raw_n": 600
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 3,
|
| 36 |
+
"step": 255,
|
| 37 |
+
"loss": 0.8038398406084846,
|
| 38 |
+
"sigma": 0.2,
|
| 39 |
+
"epoch_seconds": 80.76511526107788,
|
| 40 |
+
"train_rows": 5400,
|
| 41 |
+
"raw_ece": 0.15370363454023994,
|
| 42 |
+
"raw_brier": 0.05970909750537262,
|
| 43 |
+
"raw_nll": 0.8359687997400761,
|
| 44 |
+
"raw_accuracy": 0.775,
|
| 45 |
+
"raw_soft_accuracy": 0.522617615784208,
|
| 46 |
+
"raw_mean_confidence": 0.6295137474437554,
|
| 47 |
+
"raw_score_mae": 0.23478961663592296,
|
| 48 |
+
"raw_n": 600
|
| 49 |
+
},
|
| 50 |
+
{
|
| 51 |
+
"epoch": 4,
|
| 52 |
+
"step": 340,
|
| 53 |
+
"loss": 0.7868751822149053,
|
| 54 |
+
"sigma": 0.09999999999999998,
|
| 55 |
+
"epoch_seconds": 80.85879135131836,
|
| 56 |
+
"train_rows": 5400,
|
| 57 |
+
"raw_ece": 0.154102191751202,
|
| 58 |
+
"raw_brier": 0.057648221196283864,
|
| 59 |
+
"raw_nll": 0.8323566637126109,
|
| 60 |
+
"raw_accuracy": 0.7783333333333333,
|
| 61 |
+
"raw_soft_accuracy": 0.5270061985154947,
|
| 62 |
+
"raw_mean_confidence": 0.6377819022536277,
|
| 63 |
+
"raw_score_mae": 0.22406834450860819,
|
| 64 |
+
"raw_n": 600
|
| 65 |
+
}
|
| 66 |
+
]
|