learning_rate: 3e-05
train_batch_size: 96
eval_batch_size: 48
seed: 42
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: cosine
lr_scheduler_warmup_ratio: 0.05
num_epochs: 2

Training Loss	Epoch	Step	Validation Loss	Wer
0.2949	0.0879	300	0.2728	69.7139
0.1643	0.1758	600	0.1619	48.3076
0.135	0.2637	900	0.1303	40.1983
0.1123	0.3516	1200	0.1142	35.9610
0.1011	0.4395	1500	0.1052	33.3392
0.0965	0.5274	1800	0.0971	30.9336
0.0895	0.6153	2100	0.0921	29.5138
0.0844	0.7032	2400	0.0867	28.1619
0.0803	0.7911	2700	0.0829	26.9852
0.0794	0.8790	3000	0.0800	26.1783
0.0764	0.9669	3300	0.0771	25.3417
0.0671	1.0548	3600	0.0752	24.3019
0.0712	1.1427	3900	0.0736	24.0849
0.0657	1.2306	4200	0.0722	23.6805
0.0646	1.3185	4500	0.0710	23.4355
0.0648	1.4064	4800	0.0699	23.1196
0.0627	1.4943	5100	0.0689	22.7004
0.0609	1.5822	5400	0.0683	22.5094
0.0618	1.6701	5700	0.0678	22.2923
0.0592	1.7580	6000	0.0675	22.2988
0.059	1.8459	6300	0.0673	22.1880
0.059	1.9338	6600	0.0671	22.1786

b1n1yam/shook-tiny-amharic-600hr

tiny Amharic - Biniyam Daniel