Ramoslee/whisper-small-th_10000

Whisper Small Thai

This model is a fine-tuned version of Ramoslee/Whishper-small-th on the mozilla-foundation/common_voice_11_0 th dataset. It achieves the following results on the evaluation set:

• Loss: 0.1836
• Wer: 18.8761

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: 1e-05
• train_batch_size: 6
• eval_batch_size: 10
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1262	0.27	1000	0.2267	25.1536
0.1174	0.55	2000	0.2190	24.6093
0.1363	0.82	3000	0.2059	24.5492
0.0618	1.1	4000	0.1970	22.1944
0.0686	1.37	5000	0.1916	21.2372
0.0722	1.65	6000	0.1854	20.3488
0.0771	1.92	7000	0.1801	19.8033
0.0191	2.2	8000	0.1859	19.5656
0.0237	2.47	9000	0.1862	19.1376
0.0205	2.74	10000	0.1836	18.8761

Framework versions

• Transformers 4.33.0.dev0
• Pytorch 1.12.1+cu116
• Datasets 2.14.4.dev0
• Tokenizers 0.12.1