E2 run e4_log_sph_seed42 (epoch_4.pt)
Browse files
e4_log_sph_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_sph_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": 0.75,
|
| 46 |
+
"reward_w_rps": 0.0,
|
| 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_sph_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": 0.6685267686843872,
|
| 7 |
+
"grad_norm_rl": 1.3072677850723267,
|
| 8 |
+
"grad_norm_ce": 0.17203134298324585,
|
| 9 |
+
"grad_norm_ratio_rl_ce": 7.599009357263703
|
| 10 |
+
},
|
| 11 |
+
{
|
| 12 |
+
"step": 20,
|
| 13 |
+
"epoch": 1,
|
| 14 |
+
"sigma": 0.4,
|
| 15 |
+
"grad_cos_rl_ce": 0.8025952577590942,
|
| 16 |
+
"grad_norm_rl": 0.9179520606994629,
|
| 17 |
+
"grad_norm_ce": 0.14446447789669037,
|
| 18 |
+
"grad_norm_ratio_rl_ce": 6.35417144798675
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"step": 30,
|
| 22 |
+
"epoch": 1,
|
| 23 |
+
"sigma": 0.4,
|
| 24 |
+
"grad_cos_rl_ce": 0.7550783157348633,
|
| 25 |
+
"grad_norm_rl": 0.82485032081604,
|
| 26 |
+
"grad_norm_ce": 0.15403558313846588,
|
| 27 |
+
"grad_norm_ratio_rl_ce": 5.354933606961221
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
"step": 40,
|
| 31 |
+
"epoch": 1,
|
| 32 |
+
"sigma": 0.4,
|
| 33 |
+
"grad_cos_rl_ce": 0.8777263164520264,
|
| 34 |
+
"grad_norm_rl": 1.0839183330535889,
|
| 35 |
+
"grad_norm_ce": 0.10109312832355499,
|
| 36 |
+
"grad_norm_ratio_rl_ce": 10.721978348364482
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"step": 50,
|
| 40 |
+
"epoch": 1,
|
| 41 |
+
"sigma": 0.4,
|
| 42 |
+
"grad_cos_rl_ce": 0.6960421800613403,
|
| 43 |
+
"grad_norm_rl": 0.9210896492004395,
|
| 44 |
+
"grad_norm_ce": 0.09381662309169769,
|
| 45 |
+
"grad_norm_ratio_rl_ce": 9.817979147471055
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"step": 60,
|
| 49 |
+
"epoch": 1,
|
| 50 |
+
"sigma": 0.4,
|
| 51 |
+
"grad_cos_rl_ce": 0.7149368524551392,
|
| 52 |
+
"grad_norm_rl": 1.1098238229751587,
|
| 53 |
+
"grad_norm_ce": 0.08311197906732559,
|
| 54 |
+
"grad_norm_ratio_rl_ce": 13.35335574281219
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"step": 70,
|
| 58 |
+
"epoch": 1,
|
| 59 |
+
"sigma": 0.4,
|
| 60 |
+
"grad_cos_rl_ce": 0.5144377946853638,
|
| 61 |
+
"grad_norm_rl": 1.2926123142242432,
|
| 62 |
+
"grad_norm_ce": 0.10727235674858093,
|
| 63 |
+
"grad_norm_ratio_rl_ce": 12.04981743110014
|
| 64 |
+
},
|
| 65 |
+
{
|
| 66 |
+
"step": 80,
|
| 67 |
+
"epoch": 1,
|
| 68 |
+
"sigma": 0.4,
|
| 69 |
+
"grad_cos_rl_ce": 0.5645750761032104,
|
| 70 |
+
"grad_norm_rl": 1.2493942975997925,
|
| 71 |
+
"grad_norm_ce": 0.12623873353004456,
|
| 72 |
+
"grad_norm_ratio_rl_ce": 9.897075664993418
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"step": 90,
|
| 76 |
+
"epoch": 2,
|
| 77 |
+
"sigma": 0.30000000000000004,
|
| 78 |
+
"grad_cos_rl_ce": 0.9178726673126221,
|
| 79 |
+
"grad_norm_rl": 1.1542178392410278,
|
| 80 |
+
"grad_norm_ce": 0.1116679459810257,
|
| 81 |
+
"grad_norm_ratio_rl_ce": 10.336160740676194
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"step": 100,
|
| 85 |
+
"epoch": 2,
|
| 86 |
+
"sigma": 0.30000000000000004,
|
| 87 |
+
"grad_cos_rl_ce": 0.7095625400543213,
|
| 88 |
+
"grad_norm_rl": 1.346574306488037,
|
| 89 |
+
"grad_norm_ce": 0.11426740139722824,
|
| 90 |
+
"grad_norm_ratio_rl_ce": 11.784413489958832
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"step": 110,
|
| 94 |
+
"epoch": 2,
|
| 95 |
+
"sigma": 0.30000000000000004,
|
| 96 |
+
"grad_cos_rl_ce": 0.7695649266242981,
|
| 97 |
+
"grad_norm_rl": 0.7398955821990967,
|
| 98 |
+
"grad_norm_ce": 0.07930655777454376,
|
| 99 |
+
"grad_norm_ratio_rl_ce": 9.32956369512979
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"step": 120,
|
| 103 |
+
"epoch": 2,
|
| 104 |
+
"sigma": 0.30000000000000004,
|
| 105 |
+
"grad_cos_rl_ce": 0.5800289511680603,
|
| 106 |
+
"grad_norm_rl": 1.1850324869155884,
|
| 107 |
+
"grad_norm_ce": 0.11587293446063995,
|
| 108 |
+
"grad_norm_ratio_rl_ce": 10.226999880788586
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"step": 130,
|
| 112 |
+
"epoch": 2,
|
| 113 |
+
"sigma": 0.30000000000000004,
|
| 114 |
+
"grad_cos_rl_ce": 0.725669801235199,
|
| 115 |
+
"grad_norm_rl": 1.5664899349212646,
|
| 116 |
+
"grad_norm_ce": 0.08213718235492706,
|
| 117 |
+
"grad_norm_ratio_rl_ce": 19.07162980283676
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"step": 140,
|
| 121 |
+
"epoch": 2,
|
| 122 |
+
"sigma": 0.30000000000000004,
|
| 123 |
+
"grad_cos_rl_ce": 0.46646368503570557,
|
| 124 |
+
"grad_norm_rl": 1.628920316696167,
|
| 125 |
+
"grad_norm_ce": 0.03175926208496094,
|
| 126 |
+
"grad_norm_ratio_rl_ce": 51.28961473785358
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"step": 150,
|
| 130 |
+
"epoch": 2,
|
| 131 |
+
"sigma": 0.30000000000000004,
|
| 132 |
+
"grad_cos_rl_ce": 0.72324138879776,
|
| 133 |
+
"grad_norm_rl": 1.2553874254226685,
|
| 134 |
+
"grad_norm_ce": 0.11739315837621689,
|
| 135 |
+
"grad_norm_ratio_rl_ce": 10.693872136904718
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"step": 160,
|
| 139 |
+
"epoch": 2,
|
| 140 |
+
"sigma": 0.30000000000000004,
|
| 141 |
+
"grad_cos_rl_ce": 0.7032287120819092,
|
| 142 |
+
"grad_norm_rl": 1.2876057624816895,
|
| 143 |
+
"grad_norm_ce": 0.08051298558712006,
|
| 144 |
+
"grad_norm_ratio_rl_ce": 15.992522859414025
|
| 145 |
+
},
|
| 146 |
+
{
|
| 147 |
+
"step": 170,
|
| 148 |
+
"epoch": 2,
|
| 149 |
+
"sigma": 0.30000000000000004,
|
| 150 |
+
"grad_cos_rl_ce": 0.6808494329452515,
|
| 151 |
+
"grad_norm_rl": 1.0918775796890259,
|
| 152 |
+
"grad_norm_ce": 0.11161160469055176,
|
| 153 |
+
"grad_norm_ratio_rl_ce": 9.78283201568785
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"step": 180,
|
| 157 |
+
"epoch": 3,
|
| 158 |
+
"sigma": 0.2,
|
| 159 |
+
"grad_cos_rl_ce": 0.7304961681365967,
|
| 160 |
+
"grad_norm_rl": 2.4385781288146973,
|
| 161 |
+
"grad_norm_ce": 0.07047159969806671,
|
| 162 |
+
"grad_norm_ratio_rl_ce": 34.6037004873269
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"step": 190,
|
| 166 |
+
"epoch": 3,
|
| 167 |
+
"sigma": 0.2,
|
| 168 |
+
"grad_cos_rl_ce": 0.7186523675918579,
|
| 169 |
+
"grad_norm_rl": 2.4146840572357178,
|
| 170 |
+
"grad_norm_ce": 0.06511499732732773,
|
| 171 |
+
"grad_norm_ratio_rl_ce": 37.08337796740281
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"step": 200,
|
| 175 |
+
"epoch": 3,
|
| 176 |
+
"sigma": 0.2,
|
| 177 |
+
"grad_cos_rl_ce": 0.7804189920425415,
|
| 178 |
+
"grad_norm_rl": 2.2163100242614746,
|
| 179 |
+
"grad_norm_ce": 0.09639395028352737,
|
| 180 |
+
"grad_norm_ratio_rl_ce": 22.9922107947911
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"step": 210,
|
| 184 |
+
"epoch": 3,
|
| 185 |
+
"sigma": 0.2,
|
| 186 |
+
"grad_cos_rl_ce": 0.6971749067306519,
|
| 187 |
+
"grad_norm_rl": 2.1001691818237305,
|
| 188 |
+
"grad_norm_ce": 0.09263811260461807,
|
| 189 |
+
"grad_norm_ratio_rl_ce": 22.6706818908035
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"step": 220,
|
| 193 |
+
"epoch": 3,
|
| 194 |
+
"sigma": 0.2,
|
| 195 |
+
"grad_cos_rl_ce": 0.7133122086524963,
|
| 196 |
+
"grad_norm_rl": 2.44526743888855,
|
| 197 |
+
"grad_norm_ce": 0.03794831782579422,
|
| 198 |
+
"grad_norm_ratio_rl_ce": 64.436780837396
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"step": 230,
|
| 202 |
+
"epoch": 3,
|
| 203 |
+
"sigma": 0.2,
|
| 204 |
+
"grad_cos_rl_ce": 0.763075590133667,
|
| 205 |
+
"grad_norm_rl": 1.7326583862304688,
|
| 206 |
+
"grad_norm_ce": 0.06253659725189209,
|
| 207 |
+
"grad_norm_ratio_rl_ce": 27.70631058244932
|
| 208 |
+
},
|
| 209 |
+
{
|
| 210 |
+
"step": 240,
|
| 211 |
+
"epoch": 3,
|
| 212 |
+
"sigma": 0.2,
|
| 213 |
+
"grad_cos_rl_ce": 0.7982628345489502,
|
| 214 |
+
"grad_norm_rl": 1.7150638103485107,
|
| 215 |
+
"grad_norm_ce": 0.05778117477893829,
|
| 216 |
+
"grad_norm_ratio_rl_ce": 29.68205158358368
|
| 217 |
+
},
|
| 218 |
+
{
|
| 219 |
+
"step": 250,
|
| 220 |
+
"epoch": 3,
|
| 221 |
+
"sigma": 0.2,
|
| 222 |
+
"grad_cos_rl_ce": 0.7071397304534912,
|
| 223 |
+
"grad_norm_rl": 2.3730218410491943,
|
| 224 |
+
"grad_norm_ce": 0.10783253610134125,
|
| 225 |
+
"grad_norm_ratio_rl_ce": 22.00654762324261
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"step": 260,
|
| 229 |
+
"epoch": 4,
|
| 230 |
+
"sigma": 0.09999999999999998,
|
| 231 |
+
"grad_cos_rl_ce": 0.766703724861145,
|
| 232 |
+
"grad_norm_rl": 4.3270697593688965,
|
| 233 |
+
"grad_norm_ce": 0.07580696046352386,
|
| 234 |
+
"grad_norm_ratio_rl_ce": 57.08011154794893
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"step": 270,
|
| 238 |
+
"epoch": 4,
|
| 239 |
+
"sigma": 0.09999999999999998,
|
| 240 |
+
"grad_cos_rl_ce": 0.600197970867157,
|
| 241 |
+
"grad_norm_rl": 4.461221694946289,
|
| 242 |
+
"grad_norm_ce": 0.03490684553980827,
|
| 243 |
+
"grad_norm_ratio_rl_ce": 127.80363352680055
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"step": 280,
|
| 247 |
+
"epoch": 4,
|
| 248 |
+
"sigma": 0.09999999999999998,
|
| 249 |
+
"grad_cos_rl_ce": 0.5386843681335449,
|
| 250 |
+
"grad_norm_rl": 4.431718349456787,
|
| 251 |
+
"grad_norm_ce": 0.045794568955898285,
|
| 252 |
+
"grad_norm_ratio_rl_ce": 96.77388499332052
|
| 253 |
+
},
|
| 254 |
+
{
|
| 255 |
+
"step": 290,
|
| 256 |
+
"epoch": 4,
|
| 257 |
+
"sigma": 0.09999999999999998,
|
| 258 |
+
"grad_cos_rl_ce": 0.7022527456283569,
|
| 259 |
+
"grad_norm_rl": 4.876103401184082,
|
| 260 |
+
"grad_norm_ce": 0.0633227750658989,
|
| 261 |
+
"grad_norm_ratio_rl_ce": 77.00394362233189
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"step": 300,
|
| 265 |
+
"epoch": 4,
|
| 266 |
+
"sigma": 0.09999999999999998,
|
| 267 |
+
"grad_cos_rl_ce": 0.83854740858078,
|
| 268 |
+
"grad_norm_rl": 4.759146213531494,
|
| 269 |
+
"grad_norm_ce": 0.05183715745806694,
|
| 270 |
+
"grad_norm_ratio_rl_ce": 91.80955219971986
|
| 271 |
+
},
|
| 272 |
+
{
|
| 273 |
+
"step": 310,
|
| 274 |
+
"epoch": 4,
|
| 275 |
+
"sigma": 0.09999999999999998,
|
| 276 |
+
"grad_cos_rl_ce": 0.8074007034301758,
|
| 277 |
+
"grad_norm_rl": 4.352777004241943,
|
| 278 |
+
"grad_norm_ce": 0.0519629642367363,
|
| 279 |
+
"grad_norm_ratio_rl_ce": 83.76691107172553
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"step": 320,
|
| 283 |
+
"epoch": 4,
|
| 284 |
+
"sigma": 0.09999999999999998,
|
| 285 |
+
"grad_cos_rl_ce": 0.5879312753677368,
|
| 286 |
+
"grad_norm_rl": 4.0666985511779785,
|
| 287 |
+
"grad_norm_ce": 0.04427623748779297,
|
| 288 |
+
"grad_norm_ratio_rl_ce": 91.84833178969134
|
| 289 |
+
},
|
| 290 |
+
{
|
| 291 |
+
"step": 330,
|
| 292 |
+
"epoch": 4,
|
| 293 |
+
"sigma": 0.09999999999999998,
|
| 294 |
+
"grad_cos_rl_ce": 0.788733959197998,
|
| 295 |
+
"grad_norm_rl": 4.286769866943359,
|
| 296 |
+
"grad_norm_ce": 0.055317074060440063,
|
| 297 |
+
"grad_norm_ratio_rl_ce": 77.49451574860207
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"step": 340,
|
| 301 |
+
"epoch": 4,
|
| 302 |
+
"sigma": 0.09999999999999998,
|
| 303 |
+
"grad_cos_rl_ce": 0.6371254324913025,
|
| 304 |
+
"grad_norm_rl": 4.14387845993042,
|
| 305 |
+
"grad_norm_ce": 0.03513983264565468,
|
| 306 |
+
"grad_norm_ratio_rl_ce": 117.92538973411541
|
| 307 |
+
}
|
| 308 |
+
]
|
e4_log_sph_seed42/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8f3f008674e56ffdd6c3aeb0dcd5cdc10d81e4c65f0b5c29f90481123c1403d4
|
| 3 |
+
size 1684141044
|
e4_log_sph_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_sph_seed42/test_eval.json
ADDED
|
@@ -0,0 +1,99 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"raw": {
|
| 3 |
+
"ece": 0.13688878096640109,
|
| 4 |
+
"brier": 0.05333632830852582,
|
| 5 |
+
"nll": 0.8655621116682887,
|
| 6 |
+
"accuracy": 0.7745,
|
| 7 |
+
"soft_accuracy": 0.5185397755280137,
|
| 8 |
+
"mean_confidence": 0.6392160383313894,
|
| 9 |
+
"mean_target_max": 0.6589462240487337,
|
| 10 |
+
"score_mae": 0.2247641178953927,
|
| 11 |
+
"n": 2000,
|
| 12 |
+
"by_type": {
|
| 13 |
+
"choice": {
|
| 14 |
+
"ece": 0.1493766143421332,
|
| 15 |
+
"brier": 0.06701280505378539,
|
| 16 |
+
"nll": 0.9947354776660601,
|
| 17 |
+
"accuracy": 0.755,
|
| 18 |
+
"soft_accuracy": 0.47321102102597556,
|
| 19 |
+
"mean_confidence": 0.6113708821932474,
|
| 20 |
+
"mean_target_max": 0.6345381752649943,
|
| 21 |
+
"score_mae": null,
|
| 22 |
+
"n": 600
|
| 23 |
+
},
|
| 24 |
+
"score": {
|
| 25 |
+
"ece": 0.16540088813751935,
|
| 26 |
+
"brier": 0.05504629719041077,
|
| 27 |
+
"nll": 1.0401792683731765,
|
| 28 |
+
"accuracy": 0.73625,
|
| 29 |
+
"soft_accuracy": 0.4377902174182236,
|
| 30 |
+
"mean_confidence": 0.5740311346575617,
|
| 31 |
+
"mean_target_max": 0.5928848401084542,
|
| 32 |
+
"score_mae": 0.2247641178953927,
|
| 33 |
+
"n": 800
|
| 34 |
+
},
|
| 35 |
+
"noul": {
|
| 36 |
+
"ece": 0.10002204368511836,
|
| 37 |
+
"brier": 0.03737989305408632,
|
| 38 |
+
"nll": 0.5035658700640003,
|
| 39 |
+
"accuracy": 0.845,
|
| 40 |
+
"soft_accuracy": 0.671534607509772,
|
| 41 |
+
"mean_confidence": 0.7539743993679683,
|
| 42 |
+
"mean_target_max": 0.7714361180861791,
|
| 43 |
+
"score_mae": null,
|
| 44 |
+
"n": 600
|
| 45 |
+
}
|
| 46 |
+
}
|
| 47 |
+
},
|
| 48 |
+
"post_temperature": {
|
| 49 |
+
"ece": 0.15061623822152614,
|
| 50 |
+
"brier": 0.052181859376831695,
|
| 51 |
+
"nll": 0.8627431136853992,
|
| 52 |
+
"accuracy": 0.7745,
|
| 53 |
+
"soft_accuracy": 0.5119791107401251,
|
| 54 |
+
"mean_confidence": 0.625746834486723,
|
| 55 |
+
"mean_target_max": 0.6589462240487337,
|
| 56 |
+
"score_mae": 0.2250523244496435,
|
| 57 |
+
"n": 2000,
|
| 58 |
+
"by_type": {
|
| 59 |
+
"choice": {
|
| 60 |
+
"ece": 0.16907053758700694,
|
| 61 |
+
"brier": 0.06521383934873787,
|
| 62 |
+
"nll": 0.9906951957133909,
|
| 63 |
+
"accuracy": 0.755,
|
| 64 |
+
"soft_accuracy": 0.4637976591785749,
|
| 65 |
+
"mean_confidence": 0.5920462933182716,
|
| 66 |
+
"mean_target_max": 0.6345381752649943,
|
| 67 |
+
"score_mae": null,
|
| 68 |
+
"n": 600
|
| 69 |
+
},
|
| 70 |
+
"score": {
|
| 71 |
+
"ece": 0.17583330307155845,
|
| 72 |
+
"brier": 0.053951102091104985,
|
| 73 |
+
"nll": 1.0373363818135113,
|
| 74 |
+
"accuracy": 0.73625,
|
| 75 |
+
"soft_accuracy": 0.4329697924293578,
|
| 76 |
+
"mean_confidence": 0.5634086280316114,
|
| 77 |
+
"mean_target_max": 0.5928848401084542,
|
| 78 |
+
"score_mae": 0.2250523244496435,
|
| 79 |
+
"n": 800
|
| 80 |
+
},
|
| 81 |
+
"noul": {
|
| 82 |
+
"ece": 0.11166821002960206,
|
| 83 |
+
"brier": 0.0367908891192278,
|
| 84 |
+
"nll": 0.5020000074865917,
|
| 85 |
+
"accuracy": 0.845,
|
| 86 |
+
"soft_accuracy": 0.6655063200493654,
|
| 87 |
+
"mean_confidence": 0.7425649842619896,
|
| 88 |
+
"mean_target_max": 0.7714361180861791,
|
| 89 |
+
"score_mae": null,
|
| 90 |
+
"n": 600
|
| 91 |
+
}
|
| 92 |
+
}
|
| 93 |
+
},
|
| 94 |
+
"fitted_temperature": {
|
| 95 |
+
"type0_bucket1": 1.0848711729049683,
|
| 96 |
+
"type2_bucket0": 1.0921294689178467,
|
| 97 |
+
"type1_bucket1": 1.0488768815994263
|
| 98 |
+
}
|
| 99 |
+
}
|
e4_log_sph_seed42/train_log.json
ADDED
|
@@ -0,0 +1,70 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"epoch": 1,
|
| 4 |
+
"step": 85,
|
| 5 |
+
"loss": 0.7461222497563736,
|
| 6 |
+
"sigma": 0.4,
|
| 7 |
+
"epoch_seconds": 157.4771192073822,
|
| 8 |
+
"train_rows": 5400,
|
| 9 |
+
"raw_ece": 0.06410592769583066,
|
| 10 |
+
"raw_brier": 0.09859094113778288,
|
| 11 |
+
"raw_nll": 0.9143247046321631,
|
| 12 |
+
"raw_accuracy": 0.6666666666666666,
|
| 13 |
+
"raw_soft_accuracy": 0.4997921869779626,
|
| 14 |
+
"raw_mean_confidence": 0.6293540329734484,
|
| 15 |
+
"raw_mean_target_max": 0.6725333379705747,
|
| 16 |
+
"raw_score_mae": 0.3498744268513595,
|
| 17 |
+
"raw_n": 600
|
| 18 |
+
},
|
| 19 |
+
{
|
| 20 |
+
"epoch": 2,
|
| 21 |
+
"step": 170,
|
| 22 |
+
"loss": 0.6530889362388966,
|
| 23 |
+
"sigma": 0.30000000000000004,
|
| 24 |
+
"epoch_seconds": 156.90933656692505,
|
| 25 |
+
"train_rows": 5400,
|
| 26 |
+
"raw_ece": 0.1442658796409766,
|
| 27 |
+
"raw_brier": 0.07414405270603083,
|
| 28 |
+
"raw_nll": 0.8723851264454424,
|
| 29 |
+
"raw_accuracy": 0.745,
|
| 30 |
+
"raw_soft_accuracy": 0.5000049195066094,
|
| 31 |
+
"raw_mean_confidence": 0.602846461335818,
|
| 32 |
+
"raw_mean_target_max": 0.6725333379705747,
|
| 33 |
+
"raw_score_mae": 0.2781131777058666,
|
| 34 |
+
"raw_n": 600
|
| 35 |
+
},
|
| 36 |
+
{
|
| 37 |
+
"epoch": 3,
|
| 38 |
+
"step": 255,
|
| 39 |
+
"loss": 0.6408247221772577,
|
| 40 |
+
"sigma": 0.2,
|
| 41 |
+
"epoch_seconds": 158.1111388206482,
|
| 42 |
+
"train_rows": 5400,
|
| 43 |
+
"raw_ece": 0.12030309264858563,
|
| 44 |
+
"raw_brier": 0.061100229768955634,
|
| 45 |
+
"raw_nll": 0.8459114267211407,
|
| 46 |
+
"raw_accuracy": 0.76,
|
| 47 |
+
"raw_soft_accuracy": 0.5232347134997447,
|
| 48 |
+
"raw_mean_confidence": 0.6406695418556532,
|
| 49 |
+
"raw_mean_target_max": 0.6725333379705747,
|
| 50 |
+
"raw_score_mae": 0.24454334266483785,
|
| 51 |
+
"raw_n": 600
|
| 52 |
+
},
|
| 53 |
+
{
|
| 54 |
+
"epoch": 4,
|
| 55 |
+
"step": 340,
|
| 56 |
+
"loss": 0.6946253744848803,
|
| 57 |
+
"sigma": 0.09999999999999998,
|
| 58 |
+
"epoch_seconds": 157.9358034133911,
|
| 59 |
+
"train_rows": 5400,
|
| 60 |
+
"raw_ece": 0.15472823336720468,
|
| 61 |
+
"raw_brier": 0.054983789384354394,
|
| 62 |
+
"raw_nll": 0.8361857463450482,
|
| 63 |
+
"raw_accuracy": 0.7883333333333333,
|
| 64 |
+
"raw_soft_accuracy": 0.5243624748786291,
|
| 65 |
+
"raw_mean_confidence": 0.6368387244145075,
|
| 66 |
+
"raw_mean_target_max": 0.6725333379705747,
|
| 67 |
+
"raw_score_mae": 0.22769011886557564,
|
| 68 |
+
"raw_n": 600
|
| 69 |
+
}
|
| 70 |
+
]
|