training.log

Git commit: 53803a3acd9c7e1115233fff458d2226d7fd0c87
PyTorch CUDA version: 10.2
Parameter datasets: ['VDJdb', 'PIRD']
Parameter bert: bert
Parameter config: /home/groups/jamesz/wukevin/projects/tcr/model_configs/bert_reduced_intermediate_pe.json
Parameter outdir: bert_reduced_intermediate_pe_50_epochs_256_bs_5e-05_lr_0.0_warmup_VDJdb_PIRD
Parameter epochs: 50
Parameter bs: 256
Parameter lr: 5e-05
Parameter warmup: 0.0
Parameter cpu: False
Parameter holdout: 0.1
Parameter noneptune: False
Filtering VDJdb species to: ['MusMusculus', 'HomoSapiens']
VDJdb: dropping 0 entries for null cdr3 sequence
VDJdb: dropping 0 entries for unrecognized AAs
PIRD data TRA/TRB instances: Counter({'TRB': 46483, 'TRA': 4019, 'TRA-TRB': 637})
PIRD data 0.1655 data labelled with antigen sequence
PIRD: Removing 95 entires with non amino acid residues
Creating self supervised dataset with 98225 sequences
Maximum sequence length: 45
Example of tokenized input: CASSQDRGPANEQFF -> [25, 9, 13, 5, 5, 8, 3, 0, 11, 12, 13, 7, 4, 8, 18, 18, 24]
Split test with 9822 examples
Split train with 88403 examples
Loading vanilla BERT model