This commit is contained in:
ci 2026-10-19 09:10:18 +00:00
parent bf1b182ba1
commit 13ef311785
5 changed files with 921 additions and 136 deletions

BIN
b.tar.gz

Binary file not shown.

File diff suppressed because one or more lines are too long

View File

@ -2,28 +2,8 @@
Some weights of XLMRobertaForSequenceClassification were not initialized from the model checkpoint at /tmp/base and are newly initialized: ['classifier.dense.bias', 'classifier.dense.weight', 'classifier.out_proj.bias', 'classifier.out_proj.weight']
You should probably TRAIN this model on a down-stream task to be able to use it for predictions and inference.
trainable params: 1,476,097 || all params: 279,520,514 || trainable%: 0.5281
[jev] train=36378 calib=5291
[jev] ep0 step100/2272 loss=0.4184
[jev] ep0 step200/2272 loss=0.2023
[jev] ep0 step300/2272 loss=0.1250
[jev] ep0 step400/2272 loss=0.1653
[jev] ep0 step500/2272 loss=0.1658
[jev] ep0 step600/2272 loss=0.2868
[jev] ep0 step700/2272 loss=0.3366
[jev] ep0 step800/2272 loss=0.1671
[jev] ep0 step900/2272 loss=0.2182
[jev] ep0 step1000/2272 loss=0.1411
[jev] ep0 step1100/2272 loss=0.2944
[jev] epoch 0 calibration: {'auc': 0.9934, 'ece': 0.0208, 'n': 5291}
[jev] live-checkpoint 已写 /tmp/jd/model_live
[jev] ep1 step1200/2272 loss=0.1838
[jev] ep1 step1300/2272 loss=0.1062
[jev] ep1 step1400/2272 loss=0.2059
[jev] ep1 step1500/2272 loss=0.4619
[jev] ep1 step1600/2272 loss=0.1851
[jev] ep1 step1700/2272 loss=0.2368
[jev] ep1 step1800/2272 loss=0.2459
[jev] ep1 step1900/2272 loss=0.5413
[jev] ep1 step2000/2272 loss=0.0789
[jev] ep1 step2100/2272 loss=0.1193
[jev] ep1 step2200/2272 loss=0.1051
[jev] train=37182 calib=5291
[jev] ep0 step100/2322 loss=0.2847
[jev] ep0 step200/2322 loss=0.1989
[jev] ep0 step300/2322 loss=0.2504
[jev] ep0 step400/2322 loss=0.3098

View File

@ -1,6 +1,7 @@
{
"tag": "v0.3",
"train": 36378,
"real_query_added": 530,
"upweight": 2
"tag": "v0.4",
"train": 37182,
"generic_neg_added": 804,
"upweight": 3,
"eval_label_fixed": 0
}

File diff suppressed because one or more lines are too long