HBERTv1_emb_compress_48_L12_H128_A2

This model is a fine-tuned version of on the gokuls/wiki_book_corpus_complete_processed_bert_dataset dataset. It achieves the following results on the evaluation set:

• Loss: 6.2072
• Accuracy: 0.1440

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 80
• eval_batch_size: 80
• seed: 10
• distributed_type: multi-GPU
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 10000
• num_epochs: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
7.7257	0.14	10000	7.6502	0.0520
6.9502	0.27	20000	6.9458	0.0829
6.7568	0.41	30000	6.7497	0.0973
6.6513	0.55	40000	6.6447	0.1047
6.5712	0.68	50000	6.5735	0.1112
6.5237	0.82	60000	6.5170	0.1165
6.478	0.96	70000	6.4726	0.1209
6.4359	1.09	80000	6.4369	0.1252
6.4052	1.23	90000	6.4047	0.1273
6.3897	1.37	100000	6.3794	0.1299
6.3598	1.5	110000	6.3557	0.1319
6.3362	1.64	120000	6.3374	0.1341
6.3154	1.78	130000	6.3209	0.1348
6.3082	1.91	140000	6.3069	0.1367
6.2942	2.05	150000	6.2943	0.1377
6.2849	2.18	160000	6.2835	0.1381
6.2745	2.32	170000	6.2737	0.1391
6.2647	2.46	180000	6.2658	0.1398
6.2633	2.59	190000	6.2580	0.1407
6.2506	2.73	200000	6.2525	0.1407
6.2435	2.87	210000	6.2463	0.1413
6.2416	3.0	220000	6.2394	0.1419
6.2329	3.14	230000	6.2355	0.1421
6.2288	3.28	240000	6.2323	0.1426
6.2232	3.41	250000	6.2277	0.1428
6.2227	3.55	260000	6.2228	0.1431
6.2138	3.69	270000	6.2200	0.1433
6.2142	3.82	280000	6.2187	0.1433
6.2182	3.96	290000	6.2162	0.1435
6.2108	4.1	300000	6.2145	0.1438
6.2158	4.23	310000	6.2131	0.1437
6.2072	4.37	320000	6.2114	0.1438
6.2084	4.51	330000	6.2087	0.1440
6.2093	4.64	340000	6.2082	0.1443
6.2084	4.78	350000	6.2081	0.1440
6.2066	4.92	360000	6.2081	0.1442

Framework versions

• Transformers 4.33.2
• Pytorch 1.14.0a0+410ce96
• Datasets 2.14.5
• Tokenizers 0.13.3