whisper-medium-lug-only

This model is a fine-tuned version of openai/whisper-medium on the generator dataset. It achieves the following results on the evaluation set:

• Loss: 0.1551
• Wer: 9.7662

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: 500
• training_steps: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.1474	0.025	200	0.7380	71.3893
0.7879	0.05	400	0.4461	44.7043
0.6541	0.075	600	0.3394	32.3246
0.5203	0.1	800	0.2949	26.5475
0.509	0.125	1000	0.2774	24.2091
0.4753	0.15	1200	0.2505	20.4952
0.4726	0.175	1400	0.2375	20.7703
0.4145	0.2	1600	0.2313	18.2944
0.418	0.225	1800	0.2265	18.8446
0.4032	0.25	2000	0.2267	18.7070
0.3797	0.275	2200	0.2184	16.2311
0.3773	0.3	2400	0.2084	14.4429
0.3497	0.325	2600	0.1993	15.2682
0.3657	0.35	2800	0.1951	15.4058
0.3686	0.375	3000	0.1882	13.2050
0.3363	0.4	3200	0.1848	14.3054
0.3286	0.425	3400	0.1769	13.8927
0.3193	0.45	3600	0.1786	12.5172
0.3352	0.475	3800	0.1758	11.9670
0.3182	0.5	4000	0.1737	13.3425
0.2967	0.525	4200	0.1699	12.9298
0.3078	0.55	4400	0.1719	12.3796
0.2788	0.575	4600	0.1663	12.2421
0.2302	1.0075	4800	0.1678	11.4168
0.2109	1.0325	5000	0.1696	11.1417
0.1932	1.0575	5200	0.1713	11.2792
0.2128	1.0825	5400	0.1663	12.6547
0.2269	1.1075	5600	0.1621	12.2421
0.2324	1.1325	5800	0.1581	11.2792
0.2083	1.1575	6000	0.1579	11.1417
0.2156	1.1825	6200	0.1543	10.4539
0.2113	1.2075	6400	0.1551	9.7662
0.2235	1.2325	6600	0.1550	10.5915
0.2137	1.2575	6800	0.1537	10.4539
0.1989	1.2825	7000	0.1536	9.9037
0.2014	1.3075	7200	0.1515	10.1788
0.2109	1.3325	7400	0.1488	10.3164
0.1975	1.3575	7600	0.1500	10.5915
0.1754	1.3825	7800	0.1494	10.0413
0.182	1.4075	8000	0.1487	10.0413

Framework versions

• Transformers 4.41.0.dev0
• Pytorch 2.2.0
• Datasets 2.16.1
• Tokenizers 0.19.1