CoVBERT

CoVBERT is a protein language model which speaks the language of SARS-CoV-2 spike proteins! Enter a sequence with mask and let CoVBERT predict the mutation at that position! CoVBERT has been trained with 50K spike glycoprotein sequences scraped from GISAID

It achieves the following results on the evaluation set:

• Loss: 0.1343

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
2.3432	0.02	100	1.4642
1.4307	0.04	200	1.2907
1.3923	0.06	300	1.2445
1.2719	0.08	400	1.1913
1.1292	0.1	500	0.9962
0.9344	0.12	600	0.7351
0.7481	0.14	700	0.6377
0.6194	0.16	800	0.4843
0.4363	0.18	900	0.4043
0.416	0.2	1000	0.3693
0.3295	0.22	1100	0.3520
0.3416	0.24	1200	0.3343
0.3755	0.26	1300	0.3274
0.3064	0.28	1400	0.3127
0.3295	0.3	1500	0.2998
0.2928	0.32	1600	0.2965
0.3069	0.34	1700	0.2877
0.3048	0.36	1800	0.2850
0.2916	0.38	1900	0.2817
0.2979	0.4	2000	0.2591
0.2846	0.42	2100	0.2540
0.2568	0.44	2200	0.3389
0.277	0.46	2300	0.2369
0.2385	0.48	2400	0.2238
0.2477	0.5	2500	0.2160
0.2271	0.52	2600	0.2139
0.2457	0.54	2700	0.2024
0.2037	0.56	2800	0.2085
0.1865	0.58	2900	0.1978
0.2354	0.6	3000	0.1929
0.2001	0.62	3100	0.1865
0.2396	0.64	3200	0.1832
0.2197	0.66	3300	0.1790
0.1813	0.68	3400	0.1767
0.2109	0.7	3500	0.1970
0.1956	0.72	3600	0.1658
0.182	0.74	3700	0.1629
0.1916	0.76	3800	0.1610
0.1777	0.78	3900	0.1557
0.2005	0.8	4000	0.1492
0.1553	0.82	4100	0.1530
0.1631	0.84	4200	0.1448
0.1591	0.86	4300	0.1445
0.1499	0.88	4400	0.1427
0.1487	0.9	4500	0.1418
0.1638	0.92	4600	0.1381
0.1745	0.94	4700	0.1390
0.1551	0.96	4800	0.1366
0.1408	0.98	4900	0.1324
0.1254	1.0	5000	0.1356

Framework versions

• Transformers 4.22.0.dev0
• Pytorch 1.12.1+cu113
• Tokenizers 0.12.1