arun100/whisper-base-th-1

Whisper Base Thai

This model is a fine-tuned version of openai/whisper-base on the mozilla-foundation/common_voice_16_0 th dataset. It achieves the following results on the evaluation set:

• Loss: 0.4390
• Wer: 44.5601

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-07
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.8395	1.02	500	0.7280	60.6811
0.7819	2.03	1000	0.6414	56.1244
0.6456	4.01	1500	0.5940	53.4778
0.6091	5.03	2000	0.5633	52.0691
0.5465	7.01	2500	0.5383	50.4822
0.5406	8.02	3000	0.5200	49.5537
0.4863	10.01	3500	0.5047	48.9992
0.4691	11.02	4000	0.4919	47.9767
0.5183	13.0	4500	0.4823	47.6833
0.5025	14.02	5000	0.4730	46.7202
0.5426	15.03	5500	0.4661	46.3501
0.4713	17.01	6000	0.4594	45.9985
0.4274	18.03	6500	0.4546	45.6061
0.4248	20.01	7000	0.4500	45.3598
0.4404	21.03	7500	0.4467	45.1097
0.4144	23.01	8000	0.4438	44.8411
0.4004	24.02	8500	0.4416	44.6938
0.4165	26.0	9000	0.4403	44.6443
0.4218	27.02	9500	0.4393	44.5750
0.453	28.03	10000	0.4390	44.5601

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0