learning_rate: 0.0001
train_batch_size: 4
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 16
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: linear
lr_scheduler_warmup_steps: 300
training_steps: 5000

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.8951	0.5302	500	1.0302	0.5759	0.3943
0.5489	1.0604	1000	0.9787	0.4641	0.2641
0.5598	1.5907	1500	0.8964	0.4461	0.2572
0.2869	2.1209	2000	0.9103	0.4183	0.2357
0.2853	2.6511	2500	0.8916	0.4204	0.2374
0.1514	3.1813	3000	0.9429	0.4035	0.2172
0.1272	3.7116	3500	0.9458	0.3993	0.2172
0.0615	4.2418	4000	1.0039	0.3987	0.2159
0.0599	4.7720	4500	1.0157	0.3972	0.2153
0.0527	5.3022	5000	1.0315	0.3961	0.2144

KasuleTrevor/cdli-whisper-ml-ugeng-lug-swa-sunbird-full-a40-lr1e4-5k

cdli-whisper-ml-ugeng-lug-swa-sunbird-full-a40-lr1e4-5k