E2 run e2_rlce_sigma1_fixed_seed44 (epoch_4.pt)
Browse files- e2_rlce_sigma1_fixed_seed44/config.json +75 -0
- e2_rlce_sigma1_fixed_seed44/grad_diagnostics.json +308 -0
- e2_rlce_sigma1_fixed_seed44/model.safetensors +3 -0
- e2_rlce_sigma1_fixed_seed44/model_config.json +16 -0
- e2_rlce_sigma1_fixed_seed44/test_eval.json +91 -0
- e2_rlce_sigma1_fixed_seed44/train_log.json +66 -0
e2_rlce_sigma1_fixed_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_rlce_sigma1_fixed_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": 1.0,
|
| 40 |
+
"w_ce": 1.0,
|
| 41 |
+
"num_noise_samples": 4,
|
| 42 |
+
"sigma_start": 1.0,
|
| 43 |
+
"sigma_end": 1.0,
|
| 44 |
+
"anneal_sigma": false,
|
| 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_rlce_sigma1_fixed_seed44/grad_diagnostics.json
ADDED
|
@@ -0,0 +1,308 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"step": 10,
|
| 4 |
+
"epoch": 1,
|
| 5 |
+
"sigma": 1.0,
|
| 6 |
+
"grad_cos_rl_ce": 0.45695698261260986,
|
| 7 |
+
"grad_norm_rl": 0.2674303948879242,
|
| 8 |
+
"grad_norm_ce": 0.08280622214078903,
|
| 9 |
+
"grad_norm_ratio_rl_ce": 3.229592994029276
|
| 10 |
+
},
|
| 11 |
+
{
|
| 12 |
+
"step": 20,
|
| 13 |
+
"epoch": 1,
|
| 14 |
+
"sigma": 1.0,
|
| 15 |
+
"grad_cos_rl_ce": 0.7528681755065918,
|
| 16 |
+
"grad_norm_rl": 0.28361308574676514,
|
| 17 |
+
"grad_norm_ce": 0.10615383088588715,
|
| 18 |
+
"grad_norm_ratio_rl_ce": 2.671717858695486
|
| 19 |
+
},
|
| 20 |
+
{
|
| 21 |
+
"step": 30,
|
| 22 |
+
"epoch": 1,
|
| 23 |
+
"sigma": 1.0,
|
| 24 |
+
"grad_cos_rl_ce": 0.6610839366912842,
|
| 25 |
+
"grad_norm_rl": 0.2696579694747925,
|
| 26 |
+
"grad_norm_ce": 0.08625353872776031,
|
| 27 |
+
"grad_norm_ratio_rl_ce": 3.126340941510893
|
| 28 |
+
},
|
| 29 |
+
{
|
| 30 |
+
"step": 40,
|
| 31 |
+
"epoch": 1,
|
| 32 |
+
"sigma": 1.0,
|
| 33 |
+
"grad_cos_rl_ce": 0.48701053857803345,
|
| 34 |
+
"grad_norm_rl": 0.37481895089149475,
|
| 35 |
+
"grad_norm_ce": 0.07743094116449356,
|
| 36 |
+
"grad_norm_ratio_rl_ce": 4.840687007732902
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"step": 50,
|
| 40 |
+
"epoch": 1,
|
| 41 |
+
"sigma": 1.0,
|
| 42 |
+
"grad_cos_rl_ce": 0.6729575395584106,
|
| 43 |
+
"grad_norm_rl": 0.22833111882209778,
|
| 44 |
+
"grad_norm_ce": 0.09806489944458008,
|
| 45 |
+
"grad_norm_ratio_rl_ce": 2.328367439474465
|
| 46 |
+
},
|
| 47 |
+
{
|
| 48 |
+
"step": 60,
|
| 49 |
+
"epoch": 1,
|
| 50 |
+
"sigma": 1.0,
|
| 51 |
+
"grad_cos_rl_ce": 0.7388119697570801,
|
| 52 |
+
"grad_norm_rl": 0.25957804918289185,
|
| 53 |
+
"grad_norm_ce": 0.07535604387521744,
|
| 54 |
+
"grad_norm_ratio_rl_ce": 3.444687855598269
|
| 55 |
+
},
|
| 56 |
+
{
|
| 57 |
+
"step": 70,
|
| 58 |
+
"epoch": 1,
|
| 59 |
+
"sigma": 1.0,
|
| 60 |
+
"grad_cos_rl_ce": 0.6177007555961609,
|
| 61 |
+
"grad_norm_rl": 0.22860047221183777,
|
| 62 |
+
"grad_norm_ce": 0.11460822820663452,
|
| 63 |
+
"grad_norm_ratio_rl_ce": 1.994625305607895
|
| 64 |
+
},
|
| 65 |
+
{
|
| 66 |
+
"step": 80,
|
| 67 |
+
"epoch": 1,
|
| 68 |
+
"sigma": 1.0,
|
| 69 |
+
"grad_cos_rl_ce": 0.5739599466323853,
|
| 70 |
+
"grad_norm_rl": 0.2524332106113434,
|
| 71 |
+
"grad_norm_ce": 0.09685078263282776,
|
| 72 |
+
"grad_norm_ratio_rl_ce": 2.606413740282783
|
| 73 |
+
},
|
| 74 |
+
{
|
| 75 |
+
"step": 90,
|
| 76 |
+
"epoch": 2,
|
| 77 |
+
"sigma": 1.0,
|
| 78 |
+
"grad_cos_rl_ce": 0.4195822477340698,
|
| 79 |
+
"grad_norm_rl": 0.3142838180065155,
|
| 80 |
+
"grad_norm_ce": 0.057343095541000366,
|
| 81 |
+
"grad_norm_ratio_rl_ce": 5.480761285058326
|
| 82 |
+
},
|
| 83 |
+
{
|
| 84 |
+
"step": 100,
|
| 85 |
+
"epoch": 2,
|
| 86 |
+
"sigma": 1.0,
|
| 87 |
+
"grad_cos_rl_ce": 0.26876312494277954,
|
| 88 |
+
"grad_norm_rl": 0.3565314710140228,
|
| 89 |
+
"grad_norm_ce": 0.0712997242808342,
|
| 90 |
+
"grad_norm_ratio_rl_ce": 5.000460725622479
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"step": 110,
|
| 94 |
+
"epoch": 2,
|
| 95 |
+
"sigma": 1.0,
|
| 96 |
+
"grad_cos_rl_ce": 0.4284181296825409,
|
| 97 |
+
"grad_norm_rl": 0.2629276514053345,
|
| 98 |
+
"grad_norm_ce": 0.05510668456554413,
|
| 99 |
+
"grad_norm_ratio_rl_ce": 4.77124787089318
|
| 100 |
+
},
|
| 101 |
+
{
|
| 102 |
+
"step": 120,
|
| 103 |
+
"epoch": 2,
|
| 104 |
+
"sigma": 1.0,
|
| 105 |
+
"grad_cos_rl_ce": 0.6813182830810547,
|
| 106 |
+
"grad_norm_rl": 0.2923336923122406,
|
| 107 |
+
"grad_norm_ce": 0.07144634425640106,
|
| 108 |
+
"grad_norm_ratio_rl_ce": 4.0916536087995805
|
| 109 |
+
},
|
| 110 |
+
{
|
| 111 |
+
"step": 130,
|
| 112 |
+
"epoch": 2,
|
| 113 |
+
"sigma": 1.0,
|
| 114 |
+
"grad_cos_rl_ce": 0.2410033792257309,
|
| 115 |
+
"grad_norm_rl": 0.42087849974632263,
|
| 116 |
+
"grad_norm_ce": 0.06922305375337601,
|
| 117 |
+
"grad_norm_ratio_rl_ce": 6.080033701572959
|
| 118 |
+
},
|
| 119 |
+
{
|
| 120 |
+
"step": 140,
|
| 121 |
+
"epoch": 2,
|
| 122 |
+
"sigma": 1.0,
|
| 123 |
+
"grad_cos_rl_ce": 0.509117603302002,
|
| 124 |
+
"grad_norm_rl": 0.31302958726882935,
|
| 125 |
+
"grad_norm_ce": 0.062016844749450684,
|
| 126 |
+
"grad_norm_ratio_rl_ce": 5.047492959912347
|
| 127 |
+
},
|
| 128 |
+
{
|
| 129 |
+
"step": 150,
|
| 130 |
+
"epoch": 2,
|
| 131 |
+
"sigma": 1.0,
|
| 132 |
+
"grad_cos_rl_ce": 0.6031649112701416,
|
| 133 |
+
"grad_norm_rl": 0.31306278705596924,
|
| 134 |
+
"grad_norm_ce": 0.058204181492328644,
|
| 135 |
+
"grad_norm_ratio_rl_ce": 5.378699245126077
|
| 136 |
+
},
|
| 137 |
+
{
|
| 138 |
+
"step": 160,
|
| 139 |
+
"epoch": 2,
|
| 140 |
+
"sigma": 1.0,
|
| 141 |
+
"grad_cos_rl_ce": 0.6754093170166016,
|
| 142 |
+
"grad_norm_rl": 0.3506128191947937,
|
| 143 |
+
"grad_norm_ce": 0.056718241423368454,
|
| 144 |
+
"grad_norm_ratio_rl_ce": 6.181658852531664
|
| 145 |
+
},
|
| 146 |
+
{
|
| 147 |
+
"step": 170,
|
| 148 |
+
"epoch": 2,
|
| 149 |
+
"sigma": 1.0,
|
| 150 |
+
"grad_cos_rl_ce": 0.6540169715881348,
|
| 151 |
+
"grad_norm_rl": 0.3010418117046356,
|
| 152 |
+
"grad_norm_ce": 0.042565926909446716,
|
| 153 |
+
"grad_norm_ratio_rl_ce": 7.0723659406045964
|
| 154 |
+
},
|
| 155 |
+
{
|
| 156 |
+
"step": 180,
|
| 157 |
+
"epoch": 3,
|
| 158 |
+
"sigma": 1.0,
|
| 159 |
+
"grad_cos_rl_ce": 0.4199513792991638,
|
| 160 |
+
"grad_norm_rl": 0.29684534668922424,
|
| 161 |
+
"grad_norm_ce": 0.0482943095266819,
|
| 162 |
+
"grad_norm_ratio_rl_ce": 6.146590552769401
|
| 163 |
+
},
|
| 164 |
+
{
|
| 165 |
+
"step": 190,
|
| 166 |
+
"epoch": 3,
|
| 167 |
+
"sigma": 1.0,
|
| 168 |
+
"grad_cos_rl_ce": 0.43848276138305664,
|
| 169 |
+
"grad_norm_rl": 0.27823734283447266,
|
| 170 |
+
"grad_norm_ce": 0.03611321747303009,
|
| 171 |
+
"grad_norm_ratio_rl_ce": 7.704584700664365
|
| 172 |
+
},
|
| 173 |
+
{
|
| 174 |
+
"step": 200,
|
| 175 |
+
"epoch": 3,
|
| 176 |
+
"sigma": 1.0,
|
| 177 |
+
"grad_cos_rl_ce": 0.48373377323150635,
|
| 178 |
+
"grad_norm_rl": 0.26374223828315735,
|
| 179 |
+
"grad_norm_ce": 0.049856606870889664,
|
| 180 |
+
"grad_norm_ratio_rl_ce": 5.290015804046855
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"step": 210,
|
| 184 |
+
"epoch": 3,
|
| 185 |
+
"sigma": 1.0,
|
| 186 |
+
"grad_cos_rl_ce": 0.34459033608436584,
|
| 187 |
+
"grad_norm_rl": 0.24759367108345032,
|
| 188 |
+
"grad_norm_ce": 0.04422387108206749,
|
| 189 |
+
"grad_norm_ratio_rl_ce": 5.598643108921508
|
| 190 |
+
},
|
| 191 |
+
{
|
| 192 |
+
"step": 220,
|
| 193 |
+
"epoch": 3,
|
| 194 |
+
"sigma": 1.0,
|
| 195 |
+
"grad_cos_rl_ce": 0.6901224255561829,
|
| 196 |
+
"grad_norm_rl": 0.2870550751686096,
|
| 197 |
+
"grad_norm_ce": 0.04370104521512985,
|
| 198 |
+
"grad_norm_ratio_rl_ce": 6.568608914397945
|
| 199 |
+
},
|
| 200 |
+
{
|
| 201 |
+
"step": 230,
|
| 202 |
+
"epoch": 3,
|
| 203 |
+
"sigma": 1.0,
|
| 204 |
+
"grad_cos_rl_ce": 0.31843870878219604,
|
| 205 |
+
"grad_norm_rl": 0.2891087532043457,
|
| 206 |
+
"grad_norm_ce": 0.04290945082902908,
|
| 207 |
+
"grad_norm_ratio_rl_ce": 6.737647479020122
|
| 208 |
+
},
|
| 209 |
+
{
|
| 210 |
+
"step": 240,
|
| 211 |
+
"epoch": 3,
|
| 212 |
+
"sigma": 1.0,
|
| 213 |
+
"grad_cos_rl_ce": 0.6256386637687683,
|
| 214 |
+
"grad_norm_rl": 0.24588380753993988,
|
| 215 |
+
"grad_norm_ce": 0.050712812691926956,
|
| 216 |
+
"grad_norm_ratio_rl_ce": 4.848553935149459
|
| 217 |
+
},
|
| 218 |
+
{
|
| 219 |
+
"step": 250,
|
| 220 |
+
"epoch": 3,
|
| 221 |
+
"sigma": 1.0,
|
| 222 |
+
"grad_cos_rl_ce": 0.38666996359825134,
|
| 223 |
+
"grad_norm_rl": 0.2729535698890686,
|
| 224 |
+
"grad_norm_ce": 0.0563400536775589,
|
| 225 |
+
"grad_norm_ratio_rl_ce": 4.844751683255676
|
| 226 |
+
},
|
| 227 |
+
{
|
| 228 |
+
"step": 260,
|
| 229 |
+
"epoch": 4,
|
| 230 |
+
"sigma": 1.0,
|
| 231 |
+
"grad_cos_rl_ce": 0.2468041181564331,
|
| 232 |
+
"grad_norm_rl": 0.33478865027427673,
|
| 233 |
+
"grad_norm_ce": 0.0464031808078289,
|
| 234 |
+
"grad_norm_ratio_rl_ce": 7.214778048529658
|
| 235 |
+
},
|
| 236 |
+
{
|
| 237 |
+
"step": 270,
|
| 238 |
+
"epoch": 4,
|
| 239 |
+
"sigma": 1.0,
|
| 240 |
+
"grad_cos_rl_ce": 0.479816198348999,
|
| 241 |
+
"grad_norm_rl": 0.32446011900901794,
|
| 242 |
+
"grad_norm_ce": 0.06848782300949097,
|
| 243 |
+
"grad_norm_ratio_rl_ce": 4.7374862384522665
|
| 244 |
+
},
|
| 245 |
+
{
|
| 246 |
+
"step": 280,
|
| 247 |
+
"epoch": 4,
|
| 248 |
+
"sigma": 1.0,
|
| 249 |
+
"grad_cos_rl_ce": 0.570684015750885,
|
| 250 |
+
"grad_norm_rl": 0.3211590349674225,
|
| 251 |
+
"grad_norm_ce": 0.03672498092055321,
|
| 252 |
+
"grad_norm_ratio_rl_ce": 8.744974862265625
|
| 253 |
+
},
|
| 254 |
+
{
|
| 255 |
+
"step": 290,
|
| 256 |
+
"epoch": 4,
|
| 257 |
+
"sigma": 1.0,
|
| 258 |
+
"grad_cos_rl_ce": 0.32686543464660645,
|
| 259 |
+
"grad_norm_rl": 0.3069614768028259,
|
| 260 |
+
"grad_norm_ce": 0.036302924156188965,
|
| 261 |
+
"grad_norm_ratio_rl_ce": 8.455557890658094
|
| 262 |
+
},
|
| 263 |
+
{
|
| 264 |
+
"step": 300,
|
| 265 |
+
"epoch": 4,
|
| 266 |
+
"sigma": 1.0,
|
| 267 |
+
"grad_cos_rl_ce": 0.4872467517852783,
|
| 268 |
+
"grad_norm_rl": 0.25320661067962646,
|
| 269 |
+
"grad_norm_ce": 0.04276787117123604,
|
| 270 |
+
"grad_norm_ratio_rl_ce": 5.920486658450353
|
| 271 |
+
},
|
| 272 |
+
{
|
| 273 |
+
"step": 310,
|
| 274 |
+
"epoch": 4,
|
| 275 |
+
"sigma": 1.0,
|
| 276 |
+
"grad_cos_rl_ce": 0.3707493543624878,
|
| 277 |
+
"grad_norm_rl": 0.2654787302017212,
|
| 278 |
+
"grad_norm_ce": 0.038157664239406586,
|
| 279 |
+
"grad_norm_ratio_rl_ce": 6.957415646200723
|
| 280 |
+
},
|
| 281 |
+
{
|
| 282 |
+
"step": 320,
|
| 283 |
+
"epoch": 4,
|
| 284 |
+
"sigma": 1.0,
|
| 285 |
+
"grad_cos_rl_ce": 0.5631103515625,
|
| 286 |
+
"grad_norm_rl": 0.282014399766922,
|
| 287 |
+
"grad_norm_ce": 0.05387941375374794,
|
| 288 |
+
"grad_norm_ratio_rl_ce": 5.234177213877064
|
| 289 |
+
},
|
| 290 |
+
{
|
| 291 |
+
"step": 330,
|
| 292 |
+
"epoch": 4,
|
| 293 |
+
"sigma": 1.0,
|
| 294 |
+
"grad_cos_rl_ce": 0.401131272315979,
|
| 295 |
+
"grad_norm_rl": 0.25572794675827026,
|
| 296 |
+
"grad_norm_ce": 0.0273357592523098,
|
| 297 |
+
"grad_norm_ratio_rl_ce": 9.355070199363931
|
| 298 |
+
},
|
| 299 |
+
{
|
| 300 |
+
"step": 340,
|
| 301 |
+
"epoch": 4,
|
| 302 |
+
"sigma": 1.0,
|
| 303 |
+
"grad_cos_rl_ce": 0.565926194190979,
|
| 304 |
+
"grad_norm_rl": 0.33440101146698,
|
| 305 |
+
"grad_norm_ce": 0.05839993432164192,
|
| 306 |
+
"grad_norm_ratio_rl_ce": 5.726051156585928
|
| 307 |
+
}
|
| 308 |
+
]
|
e2_rlce_sigma1_fixed_seed44/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:f5f8b089eb548849771df50c8867f5c69ad4e090976f640a8deb9e3f27ef72ba
|
| 3 |
+
size 1684141044
|
e2_rlce_sigma1_fixed_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_rlce_sigma1_fixed_seed44/test_eval.json
ADDED
|
@@ -0,0 +1,91 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"raw": {
|
| 3 |
+
"ece": 0.07792394097149372,
|
| 4 |
+
"brier": 0.0651938930230446,
|
| 5 |
+
"nll": 0.889134444310097,
|
| 6 |
+
"accuracy": 0.761,
|
| 7 |
+
"soft_accuracy": 0.5354805619120598,
|
| 8 |
+
"mean_confidence": 0.6834244523942471,
|
| 9 |
+
"score_mae": 0.23642302882857621,
|
| 10 |
+
"n": 2000,
|
| 11 |
+
"by_type": {
|
| 12 |
+
"choice": {
|
| 13 |
+
"ece": 0.07670323486129442,
|
| 14 |
+
"brier": 0.0824239765906547,
|
| 15 |
+
"nll": 1.0238185043431198,
|
| 16 |
+
"accuracy": 0.7383333333333333,
|
| 17 |
+
"soft_accuracy": 0.4928270164380471,
|
| 18 |
+
"mean_confidence": 0.6666522144277891,
|
| 19 |
+
"score_mae": null,
|
| 20 |
+
"n": 600
|
| 21 |
+
},
|
| 22 |
+
"score": {
|
| 23 |
+
"ece": 0.08631576582789423,
|
| 24 |
+
"brier": 0.07005521307971321,
|
| 25 |
+
"nll": 1.0707776059722527,
|
| 26 |
+
"accuracy": 0.71,
|
| 27 |
+
"soft_accuracy": 0.4555477532558143,
|
| 28 |
+
"mean_confidence": 0.6238887127488851,
|
| 29 |
+
"score_mae": 0.23642302882857621,
|
| 30 |
+
"n": 800
|
| 31 |
+
},
|
| 32 |
+
"noul": {
|
| 33 |
+
"ece": 0.07208900928497315,
|
| 34 |
+
"brier": 0.04148204937987631,
|
| 35 |
+
"nll": 0.5122595020608666,
|
| 36 |
+
"accuracy": 0.8516666666666667,
|
| 37 |
+
"soft_accuracy": 0.6847111855943998,
|
| 38 |
+
"mean_confidence": 0.7795776765545209,
|
| 39 |
+
"score_mae": null,
|
| 40 |
+
"n": 600
|
| 41 |
+
}
|
| 42 |
+
}
|
| 43 |
+
},
|
| 44 |
+
"post_temperature": {
|
| 45 |
+
"ece": 0.1343130425512791,
|
| 46 |
+
"brier": 0.05431590324572952,
|
| 47 |
+
"nll": 0.8648814527792856,
|
| 48 |
+
"accuracy": 0.761,
|
| 49 |
+
"soft_accuracy": 0.5101195223405958,
|
| 50 |
+
"mean_confidence": 0.6266869630515576,
|
| 51 |
+
"score_mae": 0.23274893001886085,
|
| 52 |
+
"n": 2000,
|
| 53 |
+
"by_type": {
|
| 54 |
+
"choice": {
|
| 55 |
+
"ece": 0.1416614914437135,
|
| 56 |
+
"brier": 0.06833023387277838,
|
| 57 |
+
"nll": 0.9951477930446466,
|
| 58 |
+
"accuracy": 0.7383333333333333,
|
| 59 |
+
"soft_accuracy": 0.4618977982302507,
|
| 60 |
+
"mean_confidence": 0.5980274300773939,
|
| 61 |
+
"score_mae": null,
|
| 62 |
+
"n": 600
|
| 63 |
+
},
|
| 64 |
+
"score": {
|
| 65 |
+
"ece": 0.15051539737731218,
|
| 66 |
+
"brier": 0.05705952123160387,
|
| 67 |
+
"nll": 1.0408778808638453,
|
| 68 |
+
"accuracy": 0.71,
|
| 69 |
+
"soft_accuracy": 0.42907225193455817,
|
| 70 |
+
"mean_confidence": 0.5610981351137161,
|
| 71 |
+
"score_mae": 0.23274893001886085,
|
| 72 |
+
"n": 800
|
| 73 |
+
},
|
| 74 |
+
"noul": {
|
| 75 |
+
"ece": 0.10886838714281719,
|
| 76 |
+
"brier": 0.03664341530418151,
|
| 77 |
+
"nll": 0.4999532084011783,
|
| 78 |
+
"accuracy": 0.8516666666666667,
|
| 79 |
+
"soft_accuracy": 0.6664042736589909,
|
| 80 |
+
"mean_confidence": 0.7427982666095098,
|
| 81 |
+
"score_mae": null,
|
| 82 |
+
"n": 600
|
| 83 |
+
}
|
| 84 |
+
}
|
| 85 |
+
},
|
| 86 |
+
"fitted_temperature": {
|
| 87 |
+
"type1_bucket1": 1.3121464252471924,
|
| 88 |
+
"type0_bucket1": 1.3120508193969727,
|
| 89 |
+
"type2_bucket0": 1.3396928310394287
|
| 90 |
+
}
|
| 91 |
+
}
|
e2_rlce_sigma1_fixed_seed44/train_log.json
ADDED
|
@@ -0,0 +1,66 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
[
|
| 2 |
+
{
|
| 3 |
+
"epoch": 1,
|
| 4 |
+
"step": 85,
|
| 5 |
+
"loss": 0.5818835514433244,
|
| 6 |
+
"sigma": 1.0,
|
| 7 |
+
"epoch_seconds": 78.98793458938599,
|
| 8 |
+
"train_rows": 5400,
|
| 9 |
+
"raw_ece": 0.12090828175346056,
|
| 10 |
+
"raw_brier": 0.09979022459370027,
|
| 11 |
+
"raw_nll": 0.9418465179701646,
|
| 12 |
+
"raw_accuracy": 0.7466666666666667,
|
| 13 |
+
"raw_soft_accuracy": 0.5001427792198956,
|
| 14 |
+
"raw_mean_confidence": 0.6364750632146994,
|
| 15 |
+
"raw_score_mae": 0.3447374136031916,
|
| 16 |
+
"raw_n": 600
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"epoch": 2,
|
| 20 |
+
"step": 170,
|
| 21 |
+
"loss": 0.42471880816361485,
|
| 22 |
+
"sigma": 1.0,
|
| 23 |
+
"epoch_seconds": 80.19945287704468,
|
| 24 |
+
"train_rows": 5400,
|
| 25 |
+
"raw_ece": 0.08490342726310095,
|
| 26 |
+
"raw_brier": 0.08382235254746134,
|
| 27 |
+
"raw_nll": 0.8988613259590541,
|
| 28 |
+
"raw_accuracy": 0.775,
|
| 29 |
+
"raw_soft_accuracy": 0.5412538793310523,
|
| 30 |
+
"raw_mean_confidence": 0.699679225285848,
|
| 31 |
+
"raw_score_mae": 0.2627412011109603,
|
| 32 |
+
"raw_n": 600
|
| 33 |
+
},
|
| 34 |
+
{
|
| 35 |
+
"epoch": 3,
|
| 36 |
+
"step": 255,
|
| 37 |
+
"loss": 0.3223738591460621,
|
| 38 |
+
"sigma": 1.0,
|
| 39 |
+
"epoch_seconds": 81.27219367027283,
|
| 40 |
+
"train_rows": 5400,
|
| 41 |
+
"raw_ece": 0.11315249373515444,
|
| 42 |
+
"raw_brier": 0.0697885326120042,
|
| 43 |
+
"raw_nll": 0.8693600547038174,
|
| 44 |
+
"raw_accuracy": 0.76,
|
| 45 |
+
"raw_soft_accuracy": 0.5237226878354947,
|
| 46 |
+
"raw_mean_confidence": 0.6491171284516652,
|
| 47 |
+
"raw_score_mae": 0.2511063303119348,
|
| 48 |
+
"raw_n": 600
|
| 49 |
+
},
|
| 50 |
+
{
|
| 51 |
+
"epoch": 4,
|
| 52 |
+
"step": 340,
|
| 53 |
+
"loss": 0.2826120714930927,
|
| 54 |
+
"sigma": 1.0,
|
| 55 |
+
"epoch_seconds": 80.67953181266785,
|
| 56 |
+
"train_rows": 5400,
|
| 57 |
+
"raw_ece": 0.1173426779359579,
|
| 58 |
+
"raw_brier": 0.06703788657534597,
|
| 59 |
+
"raw_nll": 0.8654907958970095,
|
| 60 |
+
"raw_accuracy": 0.79,
|
| 61 |
+
"raw_soft_accuracy": 0.5398367372651894,
|
| 62 |
+
"raw_mean_confidence": 0.6811356214185555,
|
| 63 |
+
"raw_score_mae": 0.23269325237730906,
|
| 64 |
+
"raw_n": 600
|
| 65 |
+
}
|
| 66 |
+
]
|