learning_rate: 5e-05
train_batch_size: 16
eval_batch_size: 8
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: linear
lr_scheduler_warmup_steps: 1000
training_steps: 15000
mixed_precision_training: Native AMP
label_smoothing_factor: 0.1

Training Loss	Epoch	Step	Validation Loss	Wer	Bleu	Chrf++	Comet
2.2968	0.4264	2500	3.6900	119.2808	3.4280	20.6598	0.4696
2.0666	0.8528	5000	3.5337	119.8048	4.5093	23.0651	0.4885
2.3623	1.2792	7500	3.4509	102.3130	6.5378	26.0156	0.5112
2.1774	1.7056	10000	3.4396	109.8392	6.3905	26.5266	0.5161
2.0625	2.1320	12500	3.4622	98.0033	7.2093	27.2407	0.5218
2.0016	2.5584	15000	3.4918	100.2168	6.8330	27.4191	0.5177

jjnhuang/whisper-small-finetuned-iwslt-irish-welsh-curriculum-synthetic-b

whisper-small-finetuned-iwslt-irish-welsh-curriculum-synthetic-b