whisper-large-no-fo-100h-30k-steps

This model is a fine-tuned version of NbAiLab/nb-whisper-large on the ravnursson_asr dataset. It achieves the following results on the evaluation set:

• Loss: 0.0743
• Wer: 4.0668

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: 30000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2496	0.2320	1000	0.2656	19.1061
0.162	0.4640	2000	0.1777	13.0914
0.1308	0.6961	3000	0.1410	10.9573
0.1272	0.9281	4000	0.1217	9.6889
0.0629	1.1601	5000	0.1131	8.5263
0.0636	1.3921	6000	0.1089	8.4004
0.053	1.6241	7000	0.1026	7.3384
0.0528	1.8561	8000	0.0911	6.9609
0.0232	2.0882	9000	0.0963	6.9307
0.0218	2.3202	10000	0.0936	6.6841
0.0241	2.5522	11000	0.0868	6.4526
0.0326	2.7842	12000	0.0866	6.5331
0.0136	3.0162	13000	0.0820	5.6020
0.0098	3.2483	14000	0.0822	5.5969
0.0116	3.4803	15000	0.0806	5.4812
0.0117	3.7123	16000	0.0811	5.4963
0.013	3.9443	17000	0.0792	5.3704
0.0042	4.1763	18000	0.0778	4.8722
0.0078	4.4084	19000	0.0795	4.8822
0.0078	4.6404	20000	0.0785	4.7010
0.0076	4.8724	21000	0.0763	4.7061
0.0042	5.1044	22000	0.0753	4.5098
0.0037	5.3364	23000	0.0765	4.5400
0.0027	5.5684	24000	0.0757	4.4041
0.0037	5.8005	25000	0.0745	4.4242
0.0012	6.0325	26000	0.0742	4.3336
0.0006	6.2645	27000	0.0744	4.1675
0.0009	6.4965	28000	0.0760	4.1574
0.0006	6.7285	29000	0.0743	4.1222
0.0017	6.9606	30000	0.0743	4.0668

Framework versions

• Transformers 4.42.3
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1