mike272/whisper-large-v3-polish-impaired-v4

Whisper Large v3 - impaired polish speech v4

This model is a fine-tuned version of openai/whisper-large-v3 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.6501
• Wer: 56.9170

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: 16
• 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: 5
• training_steps: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.0749	0.1	2	2.6171	189.7233
1.9538	0.2	4	2.5450	189.7233
2.3076	0.3	6	2.2305	190.5138
1.7689	0.4	8	1.5214	100.0
1.1168	0.5	10	1.0848	100.0
1.5201	0.6	12	0.9314	98.4190
0.9803	0.7	14	0.8602	88.1423
0.4297	0.8	16	0.8011	71.9368
0.8011	0.9	18	0.7641	88.1423
0.7116	1.0	20	0.7268	71.1462
0.5078	1.1	22	0.6961	70.3557
0.3434	1.2	24	0.6913	102.7668
0.2949	1.3	26	0.6912	64.4269
0.3083	1.4	28	0.6876	70.3557
0.3996	1.5	30	0.6735	99.2095
0.4961	1.6	32	0.6827	100.0
0.3809	1.7	34	0.7010	100.0
0.3569	1.8	36	0.7126	100.0
0.2856	1.9	38	0.7077	100.0
0.7014	2.0	40	0.7168	100.0
0.0922	2.1	42	0.7038	100.0
0.2666	2.2	44	0.6838	100.0
0.1529	2.3	46	0.6524	97.2332
0.107	2.4	48	0.6326	95.6522
0.2065	2.5	50	0.6132	94.8617
0.1471	2.6	52	0.6077	87.7470
0.2814	2.7	54	0.6123	74.7036
0.1103	2.8	56	0.6161	66.0079
0.1729	2.9	58	0.6163	55.7312
0.0296	3.0	60	0.6138	49.8024
0.1011	3.1	62	0.6180	49.4071
0.1036	3.2	64	0.6289	53.3597
0.1189	3.3	66	0.6299	47.8261
0.0629	3.4	68	0.6301	49.8024
0.0678	3.5	70	0.6332	76.6798
0.0677	3.6	72	0.6340	73.9130
0.0486	3.7	74	0.6345	48.6166
0.082	3.8	76	0.6340	55.7312
0.0342	3.9	78	0.6369	55.3360
0.0443	4.0	80	0.6357	58.4980
0.0198	4.1	82	0.6331	52.9644
0.0407	4.2	84	0.6325	54.1502
0.0333	4.3	86	0.6360	58.1028
0.0144	4.4	88	0.6405	57.7075
0.0159	4.5	90	0.6443	56.9170
0.0315	4.6	92	0.6472	57.3123
0.0211	4.7	94	0.6489	56.5217
0.0311	4.8	96	0.6504	56.5217
0.0173	4.9	98	0.6500	56.1265
0.0074	5.0	100	0.6501	56.9170

Framework versions

• Transformers 4.37.2
• Pytorch 2.2.1+cu121
• Datasets 3.2.0
• Tokenizers 0.15.2