learning_rate: 0.0005
train_batch_size: 1
eval_batch_size: 6
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 2
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 100
num_epochs: 5
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
0.5648	0.2339	200	0.5308	0.1512	0.5741
0.5533	0.4678	400	0.5379	0.1464	0.5568
0.6098	0.7018	600	0.5334	0.1470	0.5465
0.57	0.9357	800	0.5447	0.1498	0.5644
0.5671	1.1696	1000	0.5391	0.1481	0.5577
0.5813	1.4035	1200	0.5490	0.1490	0.5680
0.5516	1.6374	1400	0.5282	0.1489	0.5627
0.5305	1.8713	1600	0.5350	0.1489	0.5626
0.56	2.1053	1800	0.5353	0.1498	0.5580
0.4954	2.3392	2000	0.5323	0.1465	0.5487
0.5504	2.5731	2200	0.5355	0.1493	0.5545
0.516	2.8070	2400	0.5307	0.1457	0.5482
0.5346	3.0409	2600	0.5279	0.1473	0.5494
0.5153	3.2749	2800	0.5196	0.1452	0.5407
0.458	3.5088	3000	0.5189	0.1459	0.5458
0.5071	3.7427	3200	0.5201	0.1469	0.5432
0.4796	3.9766	3400	0.5156	0.1469	0.5508
0.4832	4.2105	3600	0.5145	0.1461	0.5448
0.4206	4.4444	3800	0.5239	0.1499	0.5519
0.4654	4.6784	4000	0.5184	0.1453	0.5407
0.5103	4.9123	4200	0.5188	0.1458	0.5445

ctaguchi/ssc-lke-mms-model-mix-adapt-max3-devtrain

ssc-lke-mms-model-mix-adapt-max3-devtrain