learning_rate: 1e-05
train_batch_size: 32
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
total_train_batch_size: 64
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: 500
num_epochs: 25
mixed_precision_training: Native AMP

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
1.5448	1.0	78	1.5003	100.0	137.4657
1.3444	2.0	156	1.3168	100.0	114.8952
1.2049	3.0	234	1.1834	100.0	100.3741
1.0807	4.0	312	1.0514	100.0	100.7715
0.8051	5.0	390	0.6328	100.0	71.2413
0.4002	6.0	468	0.3343	100.0	45.9660
0.233	7.0	546	0.2361	99.2793	39.8592
0.1433	8.0	624	0.1929	98.7387	34.2659
0.0852	9.0	702	0.1712	98.5586	32.4906
0.0542	10.0	780	0.1727	97.4775	31.2846
0.0306	11.0	858	0.1784	97.6577	30.4294
0.0187	12.0	936	0.1918	97.2973	30.3388
0.011	13.0	1014	0.2040	97.1171	30.3527
0.0077	14.0	1092	0.2102	96.3964	30.2784
0.0052	15.0	1170	0.2110	96.3964	29.5534
0.0036	16.0	1248	0.2198	96.0360	29.0770
0.0018	17.0	1326	0.2267	96.3964	28.9841
0.0013	18.0	1404	0.2316	96.9369	28.7842
0.0009	19.0	1482	0.2333	95.6757	28.4775
0.0006	20.0	1560	0.2368	96.5766	28.8981
0.0005	21.0	1638	0.2388	96.2162	28.7494
0.0004	22.0	1716	0.2410	96.0360	28.6680
0.0005	23.0	1794	0.2421	96.3964	28.7447
0.0004	24.0	1872	0.2428	96.3964	28.7819
0.0004	25.0	1950	0.2433	96.2162	28.7772

S-Sethisak/whisper-small-kh-baseline

Whisper small-baseline kh - Sethisak San