meiiny00/whisper-tuboo-better-total

Whisper tubooo wow

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

• Loss: 0.2246
• Wer: 15.8605

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: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4558	0.3340	1000	0.3721	42.1604
0.3283	0.6680	2000	0.2633	31.2767
0.2531	1.0020	3000	0.2261	27.4282
0.2247	1.3360	4000	0.2015	24.4423
0.2034	1.6700	5000	0.1846	22.3749
0.1436	2.0040	6000	0.1717	20.6016
0.1209	2.3380	7000	0.1676	20.3995
0.1307	2.6720	8000	0.1562	19.1456
0.0944	3.0060	9000	0.1696	18.7013
0.0867	3.3400	10000	0.1676	18.6676
0.0944	3.6740	11000	0.1856	18.4072
0.0627	4.0080	12000	0.1811	17.5885
0.0712	4.3420	13000	0.1684	17.7518
0.0554	4.6760	14000	0.1848	17.8683
0.0376	5.0100	15000	0.1784	17.3593
0.0414	5.3440	16000	0.1887	16.9758
0.0517	5.6780	17000	0.1807	16.8087
0.0415	6.0120	18000	0.1957	16.7129
0.0349	6.3460	19000	0.1932	16.9434
0.0426	6.6800	20000	0.2042	16.5678
0.0173	7.0140	21000	0.2124	16.4823
0.0169	7.3480	22000	0.2044	16.4033
0.023	7.6820	23000	0.2081	16.2206
0.023	8.0160	24000	0.2053	16.1831
0.0163	8.3500	25000	0.2166	16.1727
0.0185	8.6840	26000	0.2226	16.0276
0.0132	9.0180	27000	0.2166	15.9279
0.0139	9.3520	28000	0.2202	15.9590
0.0125	9.6860	29000	0.2246	15.8605

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu124
• Datasets 2.20.0
• Tokenizers 0.19.1