xuliu15/FT-frisian-1h

Whisper Small Frisian 1h

This model is a fine-tuned version of openai/whisper-small on the Common Voice 6.1 dataset. It achieves the following results on the evaluation set:

• Loss: 0.9900
• Wer: 47.7918

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-06
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.4073	1.1236	100	2.2555	82.9549
1.5143	2.2472	200	1.6651	73.4557
1.1865	3.3708	300	1.4237	65.1256
0.9368	4.4944	400	1.2874	59.4832
0.8009	5.6180	500	1.1957	56.5461
0.6722	6.7416	600	1.1345	54.6890
0.5726	7.8652	700	1.0894	53.1919
0.5068	8.9888	800	1.0575	51.7769
0.4239	10.1124	900	1.0351	50.8002
0.3799	11.2360	1000	1.0197	49.9198
0.295	12.3596	1100	1.0110	49.3673
0.2852	13.4831	1200	1.0022	48.7507
0.2478	14.6067	1300	0.9965	48.3800
0.2267	15.7303	1400	0.9931	48.1911
0.1986	16.8539	1500	0.9916	48.1412
0.1922	17.9775	1600	0.9907	47.9558
0.1724	19.1011	1700	0.9905	47.8703
0.1709	20.2247	1800	0.9900	47.9059
0.1749	21.3483	1900	0.9900	47.7598
0.145	22.4719	2000	0.9900	47.7918

Framework versions

• Transformers 4.40.1
• Pytorch 2.3.0+cu121
• Datasets 2.19.0
• Tokenizers 0.19.1