E2 run e2_ce_only_seed42 (epoch_4.pt)
Browse files
e2_ce_only_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": "e2_ce_only_seed42",
|
| 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_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": 0.7490322589874268,
|
| 7 |
+
"grad_norm_rl": 0.8144728541374207,
|
| 8 |
+
"grad_norm_ce": 0.11764854192733765,
|
| 9 |
+
"grad_norm_ratio_rl_ce": 6.922931987040325
|
| 10 |
+
},
|
| 11 |
+
{
|
| 12 |
+
"step": 20,
|
| 13 |
+
"epoch": 1,
|
| 14 |
+
"sigma": 0.4,
|
| 15 |
+
"grad_cos_rl_ce": 0.6469556093215942,
|
| 16 |
+
"grad_norm_rl": 0.6837014555931091,
|
| 17 |
+
"grad_norm_ce": 0.10345723479986191,
|
| 18 |
+
"grad_norm_ratio_rl_ce": 6.608541750760399
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"step": 30,
|
| 22 |
+
"epoch": 1,
|
| 23 |
+
"sigma": 0.4,
|
| 24 |
+
"grad_cos_rl_ce": 0.7330920696258545,
|
| 25 |
+
"grad_norm_rl": 0.6161923408508301,
|
| 26 |
+
"grad_norm_ce": 0.07765420526266098,
|
| 27 |
+
"grad_norm_ratio_rl_ce": 7.935080125623514
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
"step": 40,
|
| 31 |
+
"epoch": 1,
|
| 32 |
+
"sigma": 0.4,
|
| 33 |
+
"grad_cos_rl_ce": 0.5904755592346191,
|
| 34 |
+
"grad_norm_rl": 0.5828913450241089,
|
| 35 |
+
"grad_norm_ce": 0.07511432468891144,
|
| 36 |
+
"grad_norm_ratio_rl_ce": 7.76005572090508
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"step": 50,
|
| 40 |
+
"epoch": 1,
|
| 41 |
+
"sigma": 0.4,
|
| 42 |
+
"grad_cos_rl_ce": 0.6297391653060913,
|
| 43 |
+
"grad_norm_rl": 0.8101503849029541,
|
| 44 |
+
"grad_norm_ce": 0.07851675897836685,
|
| 45 |
+
"grad_norm_ratio_rl_ce": 10.318184237917524
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"step": 60,
|
| 49 |
+
"epoch": 1,
|
| 50 |
+
"sigma": 0.4,
|
| 51 |
+
"grad_cos_rl_ce": 0.658270001411438,
|
| 52 |
+
"grad_norm_rl": 0.8112517595291138,
|
| 53 |
+
"grad_norm_ce": 0.06808825582265854,
|
| 54 |
+
"grad_norm_ratio_rl_ce": 11.914709074676338
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"step": 70,
|
| 58 |
+
"epoch": 1,
|
| 59 |
+
"sigma": 0.4,
|
| 60 |
+
"grad_cos_rl_ce": 0.7192925810813904,
|
| 61 |
+
"grad_norm_rl": 0.7508874535560608,
|
| 62 |
+
"grad_norm_ce": 0.08062101155519485,
|
| 63 |
+
"grad_norm_ratio_rl_ce": 9.31379350210692
|
| 64 |
+
},
|
| 65 |
+
{
|
| 66 |
+
"step": 80,
|
| 67 |
+
"epoch": 1,
|
| 68 |
+
"sigma": 0.4,
|
| 69 |
+
"grad_cos_rl_ce": 0.7114142179489136,
|
| 70 |
+
"grad_norm_rl": 0.7208915948867798,
|
| 71 |
+
"grad_norm_ce": 0.08046820014715195,
|
| 72 |
+
"grad_norm_ratio_rl_ce": 8.958714045653904
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"step": 90,
|
| 76 |
+
"epoch": 2,
|
| 77 |
+
"sigma": 0.30000000000000004,
|
| 78 |
+
"grad_cos_rl_ce": 0.7208168506622314,
|
| 79 |
+
"grad_norm_rl": 1.1837233304977417,
|
| 80 |
+
"grad_norm_ce": 0.05170632526278496,
|
| 81 |
+
"grad_norm_ratio_rl_ce": 22.89320164373996
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"step": 100,
|
| 85 |
+
"epoch": 2,
|
| 86 |
+
"sigma": 0.30000000000000004,
|
| 87 |
+
"grad_cos_rl_ce": 0.6655701398849487,
|
| 88 |
+
"grad_norm_rl": 1.0084272623062134,
|
| 89 |
+
"grad_norm_ce": 0.060692403465509415,
|
| 90 |
+
"grad_norm_ratio_rl_ce": 16.61537860960289
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"step": 110,
|
| 94 |
+
"epoch": 2,
|
| 95 |
+
"sigma": 0.30000000000000004,
|
| 96 |
+
"grad_cos_rl_ce": 0.6720519065856934,
|
| 97 |
+
"grad_norm_rl": 0.9765772223472595,
|
| 98 |
+
"grad_norm_ce": 0.045571278780698776,
|
| 99 |
+
"grad_norm_ratio_rl_ce": 21.42966465889209
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"step": 120,
|
| 103 |
+
"epoch": 2,
|
| 104 |
+
"sigma": 0.30000000000000004,
|
| 105 |
+
"grad_cos_rl_ce": 0.5871217846870422,
|
| 106 |
+
"grad_norm_rl": 1.0025646686553955,
|
| 107 |
+
"grad_norm_ce": 0.07075102627277374,
|
| 108 |
+
"grad_norm_ratio_rl_ce": 14.170319802713594
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"step": 130,
|
| 112 |
+
"epoch": 2,
|
| 113 |
+
"sigma": 0.30000000000000004,
|
| 114 |
+
"grad_cos_rl_ce": 0.7293237447738647,
|
| 115 |
+
"grad_norm_rl": 0.9305853843688965,
|
| 116 |
+
"grad_norm_ce": 0.05612506344914436,
|
| 117 |
+
"grad_norm_ratio_rl_ce": 16.58056716875005
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"step": 140,
|
| 121 |
+
"epoch": 2,
|
| 122 |
+
"sigma": 0.30000000000000004,
|
| 123 |
+
"grad_cos_rl_ce": 0.5379103422164917,
|
| 124 |
+
"grad_norm_rl": 1.119141936302185,
|
| 125 |
+
"grad_norm_ce": 0.04586394131183624,
|
| 126 |
+
"grad_norm_ratio_rl_ce": 24.401346772466866
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"step": 150,
|
| 130 |
+
"epoch": 2,
|
| 131 |
+
"sigma": 0.30000000000000004,
|
| 132 |
+
"grad_cos_rl_ce": 0.5462125539779663,
|
| 133 |
+
"grad_norm_rl": 0.8948385715484619,
|
| 134 |
+
"grad_norm_ce": 0.042565710842609406,
|
| 135 |
+
"grad_norm_ratio_rl_ce": 21.02252150462632
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"step": 160,
|
| 139 |
+
"epoch": 2,
|
| 140 |
+
"sigma": 0.30000000000000004,
|
| 141 |
+
"grad_cos_rl_ce": 0.6173137426376343,
|
| 142 |
+
"grad_norm_rl": 1.1458677053451538,
|
| 143 |
+
"grad_norm_ce": 0.044624242931604385,
|
| 144 |
+
"grad_norm_ratio_rl_ce": 25.678143315538737
|
| 145 |
+
},
|
| 146 |
+
{
|
| 147 |
+
"step": 170,
|
| 148 |
+
"epoch": 2,
|
| 149 |
+
"sigma": 0.30000000000000004,
|
| 150 |
+
"grad_cos_rl_ce": 0.7386109828948975,
|
| 151 |
+
"grad_norm_rl": 0.81255042552948,
|
| 152 |
+
"grad_norm_ce": 0.07433948665857315,
|
| 153 |
+
"grad_norm_ratio_rl_ce": 10.930266834655
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"step": 180,
|
| 157 |
+
"epoch": 3,
|
| 158 |
+
"sigma": 0.2,
|
| 159 |
+
"grad_cos_rl_ce": 0.7024679183959961,
|
| 160 |
+
"grad_norm_rl": 1.1371753215789795,
|
| 161 |
+
"grad_norm_ce": 0.04551057517528534,
|
| 162 |
+
"grad_norm_ratio_rl_ce": 24.987056682960276
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"step": 190,
|
| 166 |
+
"epoch": 3,
|
| 167 |
+
"sigma": 0.2,
|
| 168 |
+
"grad_cos_rl_ce": 0.7135912775993347,
|
| 169 |
+
"grad_norm_rl": 2.016152858734131,
|
| 170 |
+
"grad_norm_ce": 0.030881661921739578,
|
| 171 |
+
"grad_norm_ratio_rl_ce": 65.28641055146167
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"step": 200,
|
| 175 |
+
"epoch": 3,
|
| 176 |
+
"sigma": 0.2,
|
| 177 |
+
"grad_cos_rl_ce": 0.6833568811416626,
|
| 178 |
+
"grad_norm_rl": 0.9960702657699585,
|
| 179 |
+
"grad_norm_ce": 0.04696251079440117,
|
| 180 |
+
"grad_norm_ratio_rl_ce": 21.20990230123533
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"step": 210,
|
| 184 |
+
"epoch": 3,
|
| 185 |
+
"sigma": 0.2,
|
| 186 |
+
"grad_cos_rl_ce": 0.7478423118591309,
|
| 187 |
+
"grad_norm_rl": 1.4052350521087646,
|
| 188 |
+
"grad_norm_ce": 0.06539501249790192,
|
| 189 |
+
"grad_norm_ratio_rl_ce": 21.488413235701255
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"step": 220,
|
| 193 |
+
"epoch": 3,
|
| 194 |
+
"sigma": 0.2,
|
| 195 |
+
"grad_cos_rl_ce": 0.4376876950263977,
|
| 196 |
+
"grad_norm_rl": 1.5884841680526733,
|
| 197 |
+
"grad_norm_ce": 0.027616681531071663,
|
| 198 |
+
"grad_norm_ratio_rl_ce": 57.51900952565434
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"step": 230,
|
| 202 |
+
"epoch": 3,
|
| 203 |
+
"sigma": 0.2,
|
| 204 |
+
"grad_cos_rl_ce": 0.6556814312934875,
|
| 205 |
+
"grad_norm_rl": 1.63542640209198,
|
| 206 |
+
"grad_norm_ce": 0.030881617218255997,
|
| 207 |
+
"grad_norm_ratio_rl_ce": 52.95792608701789
|
| 208 |
+
},
|
| 209 |
+
{
|
| 210 |
+
"step": 240,
|
| 211 |
+
"epoch": 3,
|
| 212 |
+
"sigma": 0.2,
|
| 213 |
+
"grad_cos_rl_ce": 0.5930013656616211,
|
| 214 |
+
"grad_norm_rl": 1.1686702966690063,
|
| 215 |
+
"grad_norm_ce": 0.03570866212248802,
|
| 216 |
+
"grad_norm_ratio_rl_ce": 32.72792166394328
|
| 217 |
+
},
|
| 218 |
+
{
|
| 219 |
+
"step": 250,
|
| 220 |
+
"epoch": 3,
|
| 221 |
+
"sigma": 0.2,
|
| 222 |
+
"grad_cos_rl_ce": 0.5241782665252686,
|
| 223 |
+
"grad_norm_rl": 1.2646613121032715,
|
| 224 |
+
"grad_norm_ce": 0.04235117882490158,
|
| 225 |
+
"grad_norm_ratio_rl_ce": 29.861301318953554
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"step": 260,
|
| 229 |
+
"epoch": 4,
|
| 230 |
+
"sigma": 0.09999999999999998,
|
| 231 |
+
"grad_cos_rl_ce": 0.5375703573226929,
|
| 232 |
+
"grad_norm_rl": 2.594486713409424,
|
| 233 |
+
"grad_norm_ce": 0.028819989413022995,
|
| 234 |
+
"grad_norm_ratio_rl_ce": 90.02386073872198
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"step": 270,
|
| 238 |
+
"epoch": 4,
|
| 239 |
+
"sigma": 0.09999999999999998,
|
| 240 |
+
"grad_cos_rl_ce": 0.6630251407623291,
|
| 241 |
+
"grad_norm_rl": 3.3074097633361816,
|
| 242 |
+
"grad_norm_ce": 0.023950180038809776,
|
| 243 |
+
"grad_norm_ratio_rl_ce": 138.09540295633394
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"step": 280,
|
| 247 |
+
"epoch": 4,
|
| 248 |
+
"sigma": 0.09999999999999998,
|
| 249 |
+
"grad_cos_rl_ce": 0.60750412940979,
|
| 250 |
+
"grad_norm_rl": 2.5669336318969727,
|
| 251 |
+
"grad_norm_ce": 0.023383669555187225,
|
| 252 |
+
"grad_norm_ratio_rl_ce": 109.77462822243598
|
| 253 |
+
},
|
| 254 |
+
{
|
| 255 |
+
"step": 290,
|
| 256 |
+
"epoch": 4,
|
| 257 |
+
"sigma": 0.09999999999999998,
|
| 258 |
+
"grad_cos_rl_ce": 0.6737611293792725,
|
| 259 |
+
"grad_norm_rl": 2.735652446746826,
|
| 260 |
+
"grad_norm_ce": 0.03418344631791115,
|
| 261 |
+
"grad_norm_ratio_rl_ce": 80.02857351786156
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"step": 300,
|
| 265 |
+
"epoch": 4,
|
| 266 |
+
"sigma": 0.09999999999999998,
|
| 267 |
+
"grad_cos_rl_ce": 0.566366970539093,
|
| 268 |
+
"grad_norm_rl": 3.2510569095611572,
|
| 269 |
+
"grad_norm_ce": 0.036545515060424805,
|
| 270 |
+
"grad_norm_ratio_rl_ce": 88.95912136375202
|
| 271 |
+
},
|
| 272 |
+
{
|
| 273 |
+
"step": 310,
|
| 274 |
+
"epoch": 4,
|
| 275 |
+
"sigma": 0.09999999999999998,
|
| 276 |
+
"grad_cos_rl_ce": 0.5390666127204895,
|
| 277 |
+
"grad_norm_rl": 3.713308095932007,
|
| 278 |
+
"grad_norm_ce": 0.027939720079302788,
|
| 279 |
+
"grad_norm_ratio_rl_ce": 132.90426981345294
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"step": 320,
|
| 283 |
+
"epoch": 4,
|
| 284 |
+
"sigma": 0.09999999999999998,
|
| 285 |
+
"grad_cos_rl_ce": 0.5937511324882507,
|
| 286 |
+
"grad_norm_rl": 3.6446080207824707,
|
| 287 |
+
"grad_norm_ce": 0.021391957998275757,
|
| 288 |
+
"grad_norm_ratio_rl_ce": 170.37281117867914
|
| 289 |
+
},
|
| 290 |
+
{
|
| 291 |
+
"step": 330,
|
| 292 |
+
"epoch": 4,
|
| 293 |
+
"sigma": 0.09999999999999998,
|
| 294 |
+
"grad_cos_rl_ce": 0.7701722979545593,
|
| 295 |
+
"grad_norm_rl": 3.133023500442505,
|
| 296 |
+
"grad_norm_ce": 0.04318325221538544,
|
| 297 |
+
"grad_norm_ratio_rl_ce": 72.55181904354724
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"step": 340,
|
| 301 |
+
"epoch": 4,
|
| 302 |
+
"sigma": 0.09999999999999998,
|
| 303 |
+
"grad_cos_rl_ce": 0.5687936544418335,
|
| 304 |
+
"grad_norm_rl": 2.939049243927002,
|
| 305 |
+
"grad_norm_ce": 0.025405999273061752,
|
| 306 |
+
"grad_norm_ratio_rl_ce": 115.68327670714007
|
| 307 |
+
}
|
| 308 |
+
]
|
e2_ce_only_seed42/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ce42b8c2d8f24024cc0a7e8a6aeb58f82a2b4e66dbda02003e49706a38ebe20f
|
| 3 |
+
size 1684141044
|
e2_ce_only_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 |
+
}
|
e2_ce_only_seed42/test_eval.json
ADDED
|
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"raw": {
|
| 3 |
+
"ece": 0.1403095155656338,
|
| 4 |
+
"brier": 0.050999487702119047,
|
| 5 |
+
"nll": 0.860319820090197,
|
| 6 |
+
"accuracy": 0.786,
|
| 7 |
+
"soft_accuracy": 0.5234679394811392,
|
| 8 |
+
"mean_confidence": 0.6456904814988375,
|
| 9 |
+
"score_mae": 0.21561100330349292,
|
| 10 |
+
"n": 2000,
|
| 11 |
+
"by_type": {
|
| 12 |
+
"choice": {
|
| 13 |
+
"ece": 0.14761911233266192,
|
| 14 |
+
"brier": 0.06340317008623818,
|
| 15 |
+
"nll": 0.9857141292778154,
|
| 16 |
+
"accuracy": 0.7633333333333333,
|
| 17 |
+
"soft_accuracy": 0.4782372895628214,
|
| 18 |
+
"mean_confidence": 0.6157142215967178,
|
| 19 |
+
"score_mae": null,
|
| 20 |
+
"n": 600
|
| 21 |
+
},
|
| 22 |
+
"score": {
|
| 23 |
+
"ece": 0.1623956760764122,
|
| 24 |
+
"brier": 0.05286887642540137,
|
| 25 |
+
"nll": 1.0355604347772895,
|
| 26 |
+
"accuracy": 0.74625,
|
| 27 |
+
"soft_accuracy": 0.44410819893702863,
|
| 28 |
+
"mean_confidence": 0.5845614827796817,
|
| 29 |
+
"score_mae": 0.21561100330349292,
|
| 30 |
+
"n": 800
|
| 31 |
+
},
|
| 32 |
+
"noul": {
|
| 33 |
+
"ece": 0.10449459860722224,
|
| 34 |
+
"brier": 0.03610328702029015,
|
| 35 |
+
"nll": 0.5012713579864552,
|
| 36 |
+
"accuracy": 0.8616666666666667,
|
| 37 |
+
"soft_accuracy": 0.6745115767916043,
|
| 38 |
+
"mean_confidence": 0.7571720730264981,
|
| 39 |
+
"score_mae": null,
|
| 40 |
+
"n": 600
|
| 41 |
+
}
|
| 42 |
+
}
|
| 43 |
+
},
|
| 44 |
+
"post_temperature": {
|
| 45 |
+
"ece": 0.15913432231545446,
|
| 46 |
+
"brier": 0.04916353706188913,
|
| 47 |
+
"nll": 0.8560048570781946,
|
| 48 |
+
"accuracy": 0.786,
|
| 49 |
+
"soft_accuracy": 0.5143453535288572,
|
| 50 |
+
"mean_confidence": 0.6268656734824181,
|
| 51 |
+
"score_mae": 0.21544989785994403,
|
| 52 |
+
"n": 2000,
|
| 53 |
+
"by_type": {
|
| 54 |
+
"choice": {
|
| 55 |
+
"ece": 0.16967813124259315,
|
| 56 |
+
"brier": 0.06114765697592588,
|
| 57 |
+
"nll": 0.9807129614241421,
|
| 58 |
+
"accuracy": 0.7633333333333333,
|
| 59 |
+
"soft_accuracy": 0.4674826178451379,
|
| 60 |
+
"mean_confidence": 0.5936552043259143,
|
| 61 |
+
"score_mae": null,
|
| 62 |
+
"n": 600
|
| 63 |
+
},
|
| 64 |
+
"score": {
|
| 65 |
+
"ece": 0.1808890850841999,
|
| 66 |
+
"brier": 0.05069461653213694,
|
| 67 |
+
"nll": 1.030125109362416,
|
| 68 |
+
"accuracy": 0.74625,
|
| 69 |
+
"soft_accuracy": 0.43563373532146216,
|
| 70 |
+
"mean_confidence": 0.565802447013557,
|
| 71 |
+
"score_mae": 0.21544989785994403,
|
| 72 |
+
"n": 800
|
| 73 |
+
},
|
| 74 |
+
"noul": {
|
| 75 |
+
"ece": 0.12017289320627847,
|
| 76 |
+
"brier": 0.03513797785418864,
|
| 77 |
+
"nll": 0.4991364163532853,
|
| 78 |
+
"accuracy": 0.8616666666666667,
|
| 79 |
+
"soft_accuracy": 0.6661569134891033,
|
| 80 |
+
"mean_confidence": 0.7414937779307366,
|
| 81 |
+
"score_mae": null,
|
| 82 |
+
"n": 600
|
| 83 |
+
}
|
| 84 |
+
}
|
| 85 |
+
},
|
| 86 |
+
"fitted_temperature": {
|
| 87 |
+
"type0_bucket1": 1.0989445447921753,
|
| 88 |
+
"type2_bucket0": 1.1291471719741821,
|
| 89 |
+
"type1_bucket1": 1.0874115228652954
|
| 90 |
+
}
|
| 91 |
+
}
|
e2_ce_only_seed42/train_log.json
ADDED
|
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"epoch": 1,
|
| 4 |
+
"step": 85,
|
| 5 |
+
"loss": 0.9595144909970901,
|
| 6 |
+
"sigma": 0.4,
|
| 7 |
+
"epoch_seconds": 79.49684429168701,
|
| 8 |
+
"train_rows": 5400,
|
| 9 |
+
"raw_ece": 0.10328587666153909,
|
| 10 |
+
"raw_brier": 0.07454998857989784,
|
| 11 |
+
"raw_nll": 0.8712575592007488,
|
| 12 |
+
"raw_accuracy": 0.7016666666666667,
|
| 13 |
+
"raw_soft_accuracy": 0.5003429550677538,
|
| 14 |
+
"raw_mean_confidence": 0.5983807932337125,
|
| 15 |
+
"raw_score_mae": 0.31235847458592614,
|
| 16 |
+
"raw_n": 600
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 2,
|
| 20 |
+
"step": 170,
|
| 21 |
+
"loss": 0.8361811941160875,
|
| 22 |
+
"sigma": 0.30000000000000004,
|
| 23 |
+
"epoch_seconds": 81.42571139335632,
|
| 24 |
+
"train_rows": 5400,
|
| 25 |
+
"raw_ece": 0.14473333328962323,
|
| 26 |
+
"raw_brier": 0.06036134786971272,
|
| 27 |
+
"raw_nll": 0.8421358737597863,
|
| 28 |
+
"raw_accuracy": 0.77,
|
| 29 |
+
"raw_soft_accuracy": 0.5183308552702268,
|
| 30 |
+
"raw_mean_confidence": 0.6283574835459391,
|
| 31 |
+
"raw_score_mae": 0.24709411715933433,
|
| 32 |
+
"raw_n": 600
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 3,
|
| 36 |
+
"step": 255,
|
| 37 |
+
"loss": 0.7982898713911281,
|
| 38 |
+
"sigma": 0.2,
|
| 39 |
+
"epoch_seconds": 82.18434500694275,
|
| 40 |
+
"train_rows": 5400,
|
| 41 |
+
"raw_ece": 0.15027284488081935,
|
| 42 |
+
"raw_brier": 0.053520284429191635,
|
| 43 |
+
"raw_nll": 0.8298481106385589,
|
| 44 |
+
"raw_accuracy": 0.7833333333333333,
|
| 45 |
+
"raw_soft_accuracy": 0.5253539056579272,
|
| 46 |
+
"raw_mean_confidence": 0.6366282898187637,
|
| 47 |
+
"raw_score_mae": 0.22518945378251373,
|
| 48 |
+
"raw_n": 600
|
| 49 |
+
},
|
| 50 |
+
{
|
| 51 |
+
"epoch": 4,
|
| 52 |
+
"step": 340,
|
| 53 |
+
"loss": 0.7824912833816865,
|
| 54 |
+
"sigma": 0.09999999999999998,
|
| 55 |
+
"epoch_seconds": 82.71521806716919,
|
| 56 |
+
"train_rows": 5400,
|
| 57 |
+
"raw_ece": 0.14211325118939083,
|
| 58 |
+
"raw_brier": 0.052311393418016126,
|
| 59 |
+
"raw_nll": 0.8293977475259453,
|
| 60 |
+
"raw_accuracy": 0.7833333333333333,
|
| 61 |
+
"raw_soft_accuracy": 0.530145117888848,
|
| 62 |
+
"raw_mean_confidence": 0.6443844682971637,
|
| 63 |
+
"raw_score_mae": 0.2180654987207769,
|
| 64 |
+
"raw_n": 600
|
| 65 |
+
}
|
| 66 |
+
]
|