E2 run e4_log_rps_seed42 (epoch_4.pt)
Browse files
e4_log_rps_seed42/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": "e4_log_rps_seed42",
|
| 24 |
+
"epochs": 4,
|
| 25 |
+
"batch_size": 8,
|
| 26 |
+
"accum_steps": 8,
|
| 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": 0.0,
|
| 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": "e4-reward",
|
| 71 |
+
"log_artifacts": false
|
| 72 |
+
},
|
| 73 |
+
"resume_from": null
|
| 74 |
+
}
|
| 75 |
+
}
|
e4_log_rps_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": 0.6326184272766113,
|
| 7 |
+
"grad_norm_rl": 1.383046269416809,
|
| 8 |
+
"grad_norm_ce": 0.17517589032649994,
|
| 9 |
+
"grad_norm_ratio_rl_ce": 7.895186185947343
|
| 10 |
+
},
|
| 11 |
+
{
|
| 12 |
+
"step": 20,
|
| 13 |
+
"epoch": 1,
|
| 14 |
+
"sigma": 0.4,
|
| 15 |
+
"grad_cos_rl_ce": 0.781697154045105,
|
| 16 |
+
"grad_norm_rl": 0.9366812109947205,
|
| 17 |
+
"grad_norm_ce": 0.14604736864566803,
|
| 18 |
+
"grad_norm_ratio_rl_ce": 6.413543904835725
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"step": 30,
|
| 22 |
+
"epoch": 1,
|
| 23 |
+
"sigma": 0.4,
|
| 24 |
+
"grad_cos_rl_ce": 0.730133056640625,
|
| 25 |
+
"grad_norm_rl": 0.8395977020263672,
|
| 26 |
+
"grad_norm_ce": 0.16131195425987244,
|
| 27 |
+
"grad_norm_ratio_rl_ce": 5.204807702433393
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
"step": 40,
|
| 31 |
+
"epoch": 1,
|
| 32 |
+
"sigma": 0.4,
|
| 33 |
+
"grad_cos_rl_ce": 0.8553873300552368,
|
| 34 |
+
"grad_norm_rl": 1.0688592195510864,
|
| 35 |
+
"grad_norm_ce": 0.10254257172346115,
|
| 36 |
+
"grad_norm_ratio_rl_ce": 10.423565564881747
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"step": 50,
|
| 40 |
+
"epoch": 1,
|
| 41 |
+
"sigma": 0.4,
|
| 42 |
+
"grad_cos_rl_ce": 0.6366616487503052,
|
| 43 |
+
"grad_norm_rl": 0.9810330867767334,
|
| 44 |
+
"grad_norm_ce": 0.07214999943971634,
|
| 45 |
+
"grad_norm_ratio_rl_ce": 13.59713228544677
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"step": 60,
|
| 49 |
+
"epoch": 1,
|
| 50 |
+
"sigma": 0.4,
|
| 51 |
+
"grad_cos_rl_ce": 0.7626327276229858,
|
| 52 |
+
"grad_norm_rl": 1.06279456615448,
|
| 53 |
+
"grad_norm_ce": 0.07314019650220871,
|
| 54 |
+
"grad_norm_ratio_rl_ce": 14.530923035220248
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"step": 70,
|
| 58 |
+
"epoch": 1,
|
| 59 |
+
"sigma": 0.4,
|
| 60 |
+
"grad_cos_rl_ce": 0.491032212972641,
|
| 61 |
+
"grad_norm_rl": 1.2575596570968628,
|
| 62 |
+
"grad_norm_ce": 0.11930656433105469,
|
| 63 |
+
"grad_norm_ratio_rl_ce": 10.54057389170437
|
| 64 |
+
},
|
| 65 |
+
{
|
| 66 |
+
"step": 80,
|
| 67 |
+
"epoch": 1,
|
| 68 |
+
"sigma": 0.4,
|
| 69 |
+
"grad_cos_rl_ce": 0.5890774726867676,
|
| 70 |
+
"grad_norm_rl": 1.2585233449935913,
|
| 71 |
+
"grad_norm_ce": 0.13497193157672882,
|
| 72 |
+
"grad_norm_ratio_rl_ce": 9.324333809938448
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"step": 90,
|
| 76 |
+
"epoch": 2,
|
| 77 |
+
"sigma": 0.30000000000000004,
|
| 78 |
+
"grad_cos_rl_ce": 0.943693995475769,
|
| 79 |
+
"grad_norm_rl": 1.122703194618225,
|
| 80 |
+
"grad_norm_ce": 0.12414669990539551,
|
| 81 |
+
"grad_norm_ratio_rl_ce": 9.043359150696455
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"step": 100,
|
| 85 |
+
"epoch": 2,
|
| 86 |
+
"sigma": 0.30000000000000004,
|
| 87 |
+
"grad_cos_rl_ce": 0.5951673984527588,
|
| 88 |
+
"grad_norm_rl": 1.0335729122161865,
|
| 89 |
+
"grad_norm_ce": 0.08596277981996536,
|
| 90 |
+
"grad_norm_ratio_rl_ce": 12.023493358181666
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"step": 110,
|
| 94 |
+
"epoch": 2,
|
| 95 |
+
"sigma": 0.30000000000000004,
|
| 96 |
+
"grad_cos_rl_ce": 0.7359135150909424,
|
| 97 |
+
"grad_norm_rl": 0.8239881992340088,
|
| 98 |
+
"grad_norm_ce": 0.0705120712518692,
|
| 99 |
+
"grad_norm_ratio_rl_ce": 11.68577499717349
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"step": 120,
|
| 103 |
+
"epoch": 2,
|
| 104 |
+
"sigma": 0.30000000000000004,
|
| 105 |
+
"grad_cos_rl_ce": 0.5612415075302124,
|
| 106 |
+
"grad_norm_rl": 1.0125678777694702,
|
| 107 |
+
"grad_norm_ce": 0.11916808038949966,
|
| 108 |
+
"grad_norm_ratio_rl_ce": 8.496972297110958
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"step": 130,
|
| 112 |
+
"epoch": 2,
|
| 113 |
+
"sigma": 0.30000000000000004,
|
| 114 |
+
"grad_cos_rl_ce": 0.8160141706466675,
|
| 115 |
+
"grad_norm_rl": 1.6011484861373901,
|
| 116 |
+
"grad_norm_ce": 0.07566898316144943,
|
| 117 |
+
"grad_norm_ratio_rl_ce": 21.159904880988496
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"step": 140,
|
| 121 |
+
"epoch": 2,
|
| 122 |
+
"sigma": 0.30000000000000004,
|
| 123 |
+
"grad_cos_rl_ce": 0.5165460109710693,
|
| 124 |
+
"grad_norm_rl": 1.5335551500320435,
|
| 125 |
+
"grad_norm_ce": 0.0346423014998436,
|
| 126 |
+
"grad_norm_ratio_rl_ce": 44.26828136805423
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"step": 150,
|
| 130 |
+
"epoch": 2,
|
| 131 |
+
"sigma": 0.30000000000000004,
|
| 132 |
+
"grad_cos_rl_ce": 0.8116863965988159,
|
| 133 |
+
"grad_norm_rl": 1.257620930671692,
|
| 134 |
+
"grad_norm_ce": 0.12039122730493546,
|
| 135 |
+
"grad_norm_ratio_rl_ce": 10.446117701635355
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"step": 160,
|
| 139 |
+
"epoch": 2,
|
| 140 |
+
"sigma": 0.30000000000000004,
|
| 141 |
+
"grad_cos_rl_ce": 0.7248458862304688,
|
| 142 |
+
"grad_norm_rl": 1.328611969947815,
|
| 143 |
+
"grad_norm_ce": 0.09138761460781097,
|
| 144 |
+
"grad_norm_ratio_rl_ce": 14.538206032070535
|
| 145 |
+
},
|
| 146 |
+
{
|
| 147 |
+
"step": 170,
|
| 148 |
+
"epoch": 2,
|
| 149 |
+
"sigma": 0.30000000000000004,
|
| 150 |
+
"grad_cos_rl_ce": 0.7177643179893494,
|
| 151 |
+
"grad_norm_rl": 1.1070221662521362,
|
| 152 |
+
"grad_norm_ce": 0.10887444019317627,
|
| 153 |
+
"grad_norm_ratio_rl_ce": 10.16787929552559
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"step": 180,
|
| 157 |
+
"epoch": 3,
|
| 158 |
+
"sigma": 0.2,
|
| 159 |
+
"grad_cos_rl_ce": 0.7301905751228333,
|
| 160 |
+
"grad_norm_rl": 2.273535966873169,
|
| 161 |
+
"grad_norm_ce": 0.07522258907556534,
|
| 162 |
+
"grad_norm_ratio_rl_ce": 30.22411212926045
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"step": 190,
|
| 166 |
+
"epoch": 3,
|
| 167 |
+
"sigma": 0.2,
|
| 168 |
+
"grad_cos_rl_ce": 0.6325523853302002,
|
| 169 |
+
"grad_norm_rl": 2.24601674079895,
|
| 170 |
+
"grad_norm_ce": 0.06238212063908577,
|
| 171 |
+
"grad_norm_ratio_rl_ce": 36.00417423757312
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"step": 200,
|
| 175 |
+
"epoch": 3,
|
| 176 |
+
"sigma": 0.2,
|
| 177 |
+
"grad_cos_rl_ce": 0.8412027359008789,
|
| 178 |
+
"grad_norm_rl": 2.4012465476989746,
|
| 179 |
+
"grad_norm_ce": 0.0873083546757698,
|
| 180 |
+
"grad_norm_ratio_rl_ce": 27.503055768446167
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"step": 210,
|
| 184 |
+
"epoch": 3,
|
| 185 |
+
"sigma": 0.2,
|
| 186 |
+
"grad_cos_rl_ce": 0.7063694000244141,
|
| 187 |
+
"grad_norm_rl": 2.1026771068573,
|
| 188 |
+
"grad_norm_ce": 0.07413928955793381,
|
| 189 |
+
"grad_norm_ratio_rl_ce": 28.36117151101413
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"step": 220,
|
| 193 |
+
"epoch": 3,
|
| 194 |
+
"sigma": 0.2,
|
| 195 |
+
"grad_cos_rl_ce": 0.7323272228240967,
|
| 196 |
+
"grad_norm_rl": 2.0498557090759277,
|
| 197 |
+
"grad_norm_ce": 0.0421302504837513,
|
| 198 |
+
"grad_norm_ratio_rl_ce": 48.65519871206347
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"step": 230,
|
| 202 |
+
"epoch": 3,
|
| 203 |
+
"sigma": 0.2,
|
| 204 |
+
"grad_cos_rl_ce": 0.7106972932815552,
|
| 205 |
+
"grad_norm_rl": 2.0978152751922607,
|
| 206 |
+
"grad_norm_ce": 0.05276767536997795,
|
| 207 |
+
"grad_norm_ratio_rl_ce": 39.75568869546616
|
| 208 |
+
},
|
| 209 |
+
{
|
| 210 |
+
"step": 240,
|
| 211 |
+
"epoch": 3,
|
| 212 |
+
"sigma": 0.2,
|
| 213 |
+
"grad_cos_rl_ce": 0.7437864542007446,
|
| 214 |
+
"grad_norm_rl": 1.7715749740600586,
|
| 215 |
+
"grad_norm_ce": 0.057044487446546555,
|
| 216 |
+
"grad_norm_ratio_rl_ce": 31.056024050003256
|
| 217 |
+
},
|
| 218 |
+
{
|
| 219 |
+
"step": 250,
|
| 220 |
+
"epoch": 3,
|
| 221 |
+
"sigma": 0.2,
|
| 222 |
+
"grad_cos_rl_ce": 0.5999278426170349,
|
| 223 |
+
"grad_norm_rl": 2.710810422897339,
|
| 224 |
+
"grad_norm_ce": 0.10605055838823318,
|
| 225 |
+
"grad_norm_ratio_rl_ce": 25.561491274506256
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"step": 260,
|
| 229 |
+
"epoch": 4,
|
| 230 |
+
"sigma": 0.09999999999999998,
|
| 231 |
+
"grad_cos_rl_ce": 0.7253202795982361,
|
| 232 |
+
"grad_norm_rl": 4.412076950073242,
|
| 233 |
+
"grad_norm_ce": 0.0738065168261528,
|
| 234 |
+
"grad_norm_ratio_rl_ce": 59.778961801783
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"step": 270,
|
| 238 |
+
"epoch": 4,
|
| 239 |
+
"sigma": 0.09999999999999998,
|
| 240 |
+
"grad_cos_rl_ce": 0.554124653339386,
|
| 241 |
+
"grad_norm_rl": 5.026937007904053,
|
| 242 |
+
"grad_norm_ce": 0.047930169850587845,
|
| 243 |
+
"grad_norm_ratio_rl_ce": 104.88043383894663
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"step": 280,
|
| 247 |
+
"epoch": 4,
|
| 248 |
+
"sigma": 0.09999999999999998,
|
| 249 |
+
"grad_cos_rl_ce": 0.5935111045837402,
|
| 250 |
+
"grad_norm_rl": 4.936573028564453,
|
| 251 |
+
"grad_norm_ce": 0.04126075655221939,
|
| 252 |
+
"grad_norm_ratio_rl_ce": 119.64329888902432
|
| 253 |
+
},
|
| 254 |
+
{
|
| 255 |
+
"step": 290,
|
| 256 |
+
"epoch": 4,
|
| 257 |
+
"sigma": 0.09999999999999998,
|
| 258 |
+
"grad_cos_rl_ce": 0.6941544413566589,
|
| 259 |
+
"grad_norm_rl": 4.694208145141602,
|
| 260 |
+
"grad_norm_ce": 0.06252127885818481,
|
| 261 |
+
"grad_norm_ratio_rl_ce": 75.08176785361887
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"step": 300,
|
| 265 |
+
"epoch": 4,
|
| 266 |
+
"sigma": 0.09999999999999998,
|
| 267 |
+
"grad_cos_rl_ce": 0.7844544649124146,
|
| 268 |
+
"grad_norm_rl": 4.243193626403809,
|
| 269 |
+
"grad_norm_ce": 0.04553676024079323,
|
| 270 |
+
"grad_norm_ratio_rl_ce": 93.18171964729774
|
| 271 |
+
},
|
| 272 |
+
{
|
| 273 |
+
"step": 310,
|
| 274 |
+
"epoch": 4,
|
| 275 |
+
"sigma": 0.09999999999999998,
|
| 276 |
+
"grad_cos_rl_ce": 0.7561176419258118,
|
| 277 |
+
"grad_norm_rl": 4.593863487243652,
|
| 278 |
+
"grad_norm_ce": 0.04601692035794258,
|
| 279 |
+
"grad_norm_ratio_rl_ce": 99.82987673904053
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"step": 320,
|
| 283 |
+
"epoch": 4,
|
| 284 |
+
"sigma": 0.09999999999999998,
|
| 285 |
+
"grad_cos_rl_ce": 0.5650932788848877,
|
| 286 |
+
"grad_norm_rl": 4.1466803550720215,
|
| 287 |
+
"grad_norm_ce": 0.03800215944647789,
|
| 288 |
+
"grad_norm_ratio_rl_ce": 109.11696639008612
|
| 289 |
+
},
|
| 290 |
+
{
|
| 291 |
+
"step": 330,
|
| 292 |
+
"epoch": 4,
|
| 293 |
+
"sigma": 0.09999999999999998,
|
| 294 |
+
"grad_cos_rl_ce": 0.750091552734375,
|
| 295 |
+
"grad_norm_rl": 4.617717266082764,
|
| 296 |
+
"grad_norm_ce": 0.06429167836904526,
|
| 297 |
+
"grad_norm_ratio_rl_ce": 71.82449398157371
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"step": 340,
|
| 301 |
+
"epoch": 4,
|
| 302 |
+
"sigma": 0.09999999999999998,
|
| 303 |
+
"grad_cos_rl_ce": 0.57184898853302,
|
| 304 |
+
"grad_norm_rl": 4.670238494873047,
|
| 305 |
+
"grad_norm_ce": 0.03306351974606514,
|
| 306 |
+
"grad_norm_ratio_rl_ce": 141.25049392023206
|
| 307 |
+
}
|
| 308 |
+
]
|
e4_log_rps_seed42/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:91ebc086b2f6ac8cccae37332218cfcc4d419f877f20f5f9551c6f0b12f8c024
|
| 3 |
+
size 1684141044
|
e4_log_rps_seed42/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 |
+
}
|
e4_log_rps_seed42/test_eval.json
ADDED
|
@@ -0,0 +1,99 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"raw": {
|
| 3 |
+
"ece": 0.12751009610295297,
|
| 4 |
+
"brier": 0.05335478361494638,
|
| 5 |
+
"nll": 0.8650917832469568,
|
| 6 |
+
"accuracy": 0.768,
|
| 7 |
+
"soft_accuracy": 0.5189383897446096,
|
| 8 |
+
"mean_confidence": 0.6404899044930935,
|
| 9 |
+
"mean_target_max": 0.6589462240487337,
|
| 10 |
+
"score_mae": 0.22315380703395932,
|
| 11 |
+
"n": 2000,
|
| 12 |
+
"by_type": {
|
| 13 |
+
"choice": {
|
| 14 |
+
"ece": 0.14127451032400132,
|
| 15 |
+
"brier": 0.06595876092881706,
|
| 16 |
+
"nll": 0.992113546282053,
|
| 17 |
+
"accuracy": 0.7516666666666667,
|
| 18 |
+
"soft_accuracy": 0.47432611651718615,
|
| 19 |
+
"mean_confidence": 0.6128524332245191,
|
| 20 |
+
"mean_target_max": 0.6345381752649943,
|
| 21 |
+
"score_mae": null,
|
| 22 |
+
"n": 600
|
| 23 |
+
},
|
| 24 |
+
"score": {
|
| 25 |
+
"ece": 0.15239446971565485,
|
| 26 |
+
"brier": 0.057384368095347325,
|
| 27 |
+
"nll": 1.04262451373972,
|
| 28 |
+
"accuracy": 0.72625,
|
| 29 |
+
"soft_accuracy": 0.43722640846855937,
|
| 30 |
+
"mean_confidence": 0.5746278527379036,
|
| 31 |
+
"mean_target_max": 0.5928848401084542,
|
| 32 |
+
"score_mae": 0.22315380703395932,
|
| 33 |
+
"n": 800
|
| 34 |
+
},
|
| 35 |
+
"noul": {
|
| 36 |
+
"ece": 0.10585159599781038,
|
| 37 |
+
"brier": 0.03537802699387445,
|
| 38 |
+
"nll": 0.5013597128881763,
|
| 39 |
+
"accuracy": 0.84,
|
| 40 |
+
"soft_accuracy": 0.6724999713401,
|
| 41 |
+
"mean_confidence": 0.7559434447685878,
|
| 42 |
+
"mean_target_max": 0.7714361180861791,
|
| 43 |
+
"score_mae": null,
|
| 44 |
+
"n": 600
|
| 45 |
+
}
|
| 46 |
+
}
|
| 47 |
+
},
|
| 48 |
+
"post_temperature": {
|
| 49 |
+
"ece": 0.14276222667098046,
|
| 50 |
+
"brier": 0.052112305763053604,
|
| 51 |
+
"nll": 0.8620958601590246,
|
| 52 |
+
"accuracy": 0.768,
|
| 53 |
+
"soft_accuracy": 0.5115978672802448,
|
| 54 |
+
"mean_confidence": 0.6254664157927037,
|
| 55 |
+
"mean_target_max": 0.6589462240487337,
|
| 56 |
+
"score_mae": 0.22311260582937392,
|
| 57 |
+
"n": 2000,
|
| 58 |
+
"by_type": {
|
| 59 |
+
"choice": {
|
| 60 |
+
"ece": 0.16253415167331697,
|
| 61 |
+
"brier": 0.06408576760397712,
|
| 62 |
+
"nll": 0.987983137127012,
|
| 63 |
+
"accuracy": 0.7516666666666667,
|
| 64 |
+
"soft_accuracy": 0.4642106085518996,
|
| 65 |
+
"mean_confidence": 0.5922220328450203,
|
| 66 |
+
"mean_target_max": 0.6345381752649943,
|
| 67 |
+
"score_mae": null,
|
| 68 |
+
"n": 600
|
| 69 |
+
},
|
| 70 |
+
"score": {
|
| 71 |
+
"ece": 0.16398581121116876,
|
| 72 |
+
"brier": 0.05605270585605467,
|
| 73 |
+
"nll": 1.0393069208879024,
|
| 74 |
+
"accuracy": 0.72625,
|
| 75 |
+
"soft_accuracy": 0.43212321305647494,
|
| 76 |
+
"mean_confidence": 0.5632208345085382,
|
| 77 |
+
"mean_target_max": 0.5928848401084542,
|
| 78 |
+
"score_mae": 0.22311260582937392,
|
| 79 |
+
"n": 800
|
| 80 |
+
},
|
| 81 |
+
"noul": {
|
| 82 |
+
"ece": 0.12543535361687344,
|
| 83 |
+
"brier": 0.034884977131462006,
|
| 84 |
+
"nll": 0.4999271688858668,
|
| 85 |
+
"accuracy": 0.84,
|
| 86 |
+
"soft_accuracy": 0.6649513316402833,
|
| 87 |
+
"mean_confidence": 0.7417049071192742,
|
| 88 |
+
"mean_target_max": 0.7714361180861791,
|
| 89 |
+
"score_mae": null,
|
| 90 |
+
"n": 600
|
| 91 |
+
}
|
| 92 |
+
}
|
| 93 |
+
},
|
| 94 |
+
"fitted_temperature": {
|
| 95 |
+
"type0_bucket1": 1.0903278589248657,
|
| 96 |
+
"type2_bucket0": 1.1138123273849487,
|
| 97 |
+
"type1_bucket1": 1.0528963804244995
|
| 98 |
+
}
|
| 99 |
+
}
|
e4_log_rps_seed42/train_log.json
ADDED
|
@@ -0,0 +1,70 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"epoch": 1,
|
| 4 |
+
"step": 85,
|
| 5 |
+
"loss": 0.7387177494518897,
|
| 6 |
+
"sigma": 0.4,
|
| 7 |
+
"epoch_seconds": 157.23304390907288,
|
| 8 |
+
"train_rows": 5400,
|
| 9 |
+
"raw_ece": 0.08016197909911472,
|
| 10 |
+
"raw_brier": 0.09696740445269503,
|
| 11 |
+
"raw_nll": 0.9070173389045522,
|
| 12 |
+
"raw_accuracy": 0.66,
|
| 13 |
+
"raw_soft_accuracy": 0.489753557741642,
|
| 14 |
+
"raw_mean_confidence": 0.6046398454904556,
|
| 15 |
+
"raw_mean_target_max": 0.6725333379705747,
|
| 16 |
+
"raw_score_mae": 0.34550478937259566,
|
| 17 |
+
"raw_n": 600
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 2,
|
| 21 |
+
"step": 170,
|
| 22 |
+
"loss": 0.6480564471407264,
|
| 23 |
+
"sigma": 0.30000000000000004,
|
| 24 |
+
"epoch_seconds": 156.58165740966797,
|
| 25 |
+
"train_rows": 5400,
|
| 26 |
+
"raw_ece": 0.1317530729373296,
|
| 27 |
+
"raw_brier": 0.07473039917293751,
|
| 28 |
+
"raw_nll": 0.8697544336194794,
|
| 29 |
+
"raw_accuracy": 0.7266666666666667,
|
| 30 |
+
"raw_soft_accuracy": 0.4974472011377414,
|
| 31 |
+
"raw_mean_confidence": 0.5980308058361212,
|
| 32 |
+
"raw_mean_target_max": 0.6725333379705747,
|
| 33 |
+
"raw_score_mae": 0.282463128255525,
|
| 34 |
+
"raw_n": 600
|
| 35 |
+
},
|
| 36 |
+
{
|
| 37 |
+
"epoch": 3,
|
| 38 |
+
"step": 255,
|
| 39 |
+
"loss": 0.6467672417707303,
|
| 40 |
+
"sigma": 0.2,
|
| 41 |
+
"epoch_seconds": 156.63458108901978,
|
| 42 |
+
"train_rows": 5400,
|
| 43 |
+
"raw_ece": 0.12420274237791697,
|
| 44 |
+
"raw_brier": 0.06251986555340724,
|
| 45 |
+
"raw_nll": 0.8452689741551876,
|
| 46 |
+
"raw_accuracy": 0.7583333333333333,
|
| 47 |
+
"raw_soft_accuracy": 0.520738270431757,
|
| 48 |
+
"raw_mean_confidence": 0.6372942742705345,
|
| 49 |
+
"raw_mean_target_max": 0.6725333379705747,
|
| 50 |
+
"raw_score_mae": 0.24468544826377184,
|
| 51 |
+
"raw_n": 600
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 4,
|
| 55 |
+
"step": 340,
|
| 56 |
+
"loss": 0.6888515679129198,
|
| 57 |
+
"sigma": 0.09999999999999998,
|
| 58 |
+
"epoch_seconds": 156.61141157150269,
|
| 59 |
+
"train_rows": 5400,
|
| 60 |
+
"raw_ece": 0.1530975746611754,
|
| 61 |
+
"raw_brier": 0.056944469758473795,
|
| 62 |
+
"raw_nll": 0.8360695086869722,
|
| 63 |
+
"raw_accuracy": 0.7866666666666666,
|
| 64 |
+
"raw_soft_accuracy": 0.5238494022935629,
|
| 65 |
+
"raw_mean_confidence": 0.637761945426464,
|
| 66 |
+
"raw_mean_target_max": 0.6725333379705747,
|
| 67 |
+
"raw_score_mae": 0.22940462150145322,
|
| 68 |
+
"raw_n": 600
|
| 69 |
+
}
|
| 70 |
+
]
|