learning_rate: 0.0005
train_batch_size: 8
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 32
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 300
training_steps: 3000

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.7263	1.3528	250	1.2452	0.6471	0.2623
0.6671	2.7056	500	0.9762	0.6391	0.2520
0.5934	4.0543	750	1.0752	0.6296	0.2506
0.5028	5.4071	1000	1.1978	0.6488	0.2623
0.4728	6.7598	1250	1.1441	0.6302	0.2487
0.4475	8.1085	1500	1.1992	0.6330	0.2537
0.4341	9.4613	1750	1.2065	0.6362	0.2531
0.4306	10.8141	2000	1.2479	0.6373	0.2526
0.4268	12.1628	2250	1.2320	0.6359	0.2523
0.4302	13.5156	2500	1.2432	0.6356	0.2522
0.3962	14.8684	2750	1.2413	0.6364	0.2527
0.422	16.2171	3000	1.2409	0.6363	0.2528

KasuleTrevor/cdli-xlsr-lg-xlsr-transformer-a40-5e-4-cosine-wr01

cdli-xlsr-lg-xlsr-transformer-a40-5e-4-cosine-wr01