yacht/whisper-large-v3-cv17-th-ft

Whisper Large v3 Thai Finetuned

This model is a fine-tuned version of openai/whisper-large-v3 on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2345
• Cer: 10.6496
• Wer: 37.1412

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: 2e-05
• train_batch_size: 32
• eval_batch_size: 32
• 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
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
0.2027	0.4873	500	0.1805	107.2858	75.0935
0.1674	0.9747	1000	0.1508	8.7078	41.0794
0.1073	1.4620	1500	0.1506	38.7265	45.4534
0.1035	1.9493	2000	0.1372	10.7331	38.5129
0.0587	2.4366	2500	0.1438	16.8383	50.0563
0.0627	2.9240	3000	0.1397	10.6251	31.3447
0.0356	3.4113	3500	0.1497	7.8515	33.7998
0.0367	3.8986	4000	0.1456	18.7090	37.0359
0.0184	4.3860	4500	0.1606	39.3584	93.1345
0.0204	4.8733	5000	0.1596	8.4796	31.7272
0.0112	5.3606	5500	0.1730	4.8027	25.0106
0.0119	5.8480	6000	0.1697	36.5628	82.3949
0.0057	6.3353	6500	0.1800	17.5990	50.1931
0.0052	6.8226	7000	0.1789	48.1183	98.1247
0.003	7.3099	7500	0.1960	15.7676	41.7634
0.0028	7.7973	8000	0.1980	15.2090	54.8407
0.001	8.2846	8500	0.2091	21.4387	68.7365
0.001	8.7719	9000	0.2175	11.7533	40.0988
0.0001	9.2593	9500	0.2327	13.1280	40.6133
0.0001	9.7466	10000	0.2345	10.6496	37.1412

Framework versions

• Transformers 4.45.2
• Pytorch 2.5.0+cu124
• Datasets 3.0.1
• Tokenizers 0.20.1