scream_duodevicesimus_working_noaudiobooks_7e5_v2

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

• step: 19999
• validation_fleurs_loss: 0.3089
• train_loss: 0.7173
• validation_fleurs_wer: 8.2391
• validation_fleurs_cer: 3.7039
• validation_fleurs_exact_wer: 12.6643
• validation_fleurs_exact_cer: 4.6485
• validation_stortinget_loss: 0.2845
• validation_stortinget_wer: 13.9625
• validation_stortinget_cer: 10.0306
• validation_stortinget_exact_wer: 17.2389
• validation_stortinget_exact_cer: 10.5844
• validation_nrk_tv_loss: 0.7447
• validation_nrk_tv_wer: 40.1880
• validation_nrk_tv_cer: 31.3161
• validation_nrk_tv_exact_wer: 47.6494
• validation_nrk_tv_exact_cer: 32.4497

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: 7e-05
• lr_scheduler_type: linear
• per_device_train_batch_size: 32
• total_train_batch_size_per_node: 128
• total_train_batch_size: 1024
• total_optimization_steps: 20,000
• starting_optimization_step: None
• finishing_optimization_step: 20,000
• num_train_dataset_workers: 32
• num_hosts: 8
• total_num_training_examples: 20,480,000
• steps_per_epoch: 11275
• num_beams: 5
• dropout: True
• bpe_dropout_probability: 0.1
• activation_dropout_probability: 0.1

Training results

step	validation_fleurs_loss	train_loss	validation_fleurs_wer	validation_fleurs_cer	validation_fleurs_exact_wer	validation_fleurs_exact_cer	validation_stortinget_loss	validation_stortinget_wer	validation_stortinget_cer	validation_stortinget_exact_wer	validation_stortinget_exact_cer	validation_nrk_tv_loss	validation_nrk_tv_wer	validation_nrk_tv_cer	validation_nrk_tv_exact_wer	validation_nrk_tv_exact_cer
0	1.3211	3.0189	110.1725	80.3659	196.8041	131.4230	1.5012	76.6096	51.2561	82.1890	54.4126	1.8187	259.8656	217.2117	269.5665	222.7746
1000	0.6977	1.1353	13.4444	4.3105	17.5926	5.2863	0.4717	21.7105	13.9604	25.3783	14.6687	0.9934	86.4845	70.4142	93.7677	73.6462
2000	0.3926	0.8912	10.5889	3.7088	14.7849	4.6968	0.3930	18.7212	12.5960	22.2213	13.2354	0.8926	49.9691	39.8385	57.6635	41.2514
3000	0.3620	0.8106	10.7674	4.3007	15.0836	5.2573	0.3632	17.5019	11.9674	21.0430	12.5977	0.8606	44.9157	34.5510	52.6419	35.8510
4000	0.3363	0.8043	10.3807	3.8518	14.0980	4.7886	0.3443	16.1694	11.2786	19.6917	11.8983	0.8431	44.9487	34.0425	52.5379	35.4061
5000	0.3060	0.7682	9.6074	3.6694	13.8590	4.5808	0.3329	16.0903	11.1667	19.5724	11.7732	0.8154	45.4598	35.0224	52.7292	36.3997
6000	0.3477	0.7510	9.2207	3.5510	13.3214	4.5083	0.3246	15.9711	11.2829	19.4232	11.8775	0.8097	43.0897	33.1321	50.5325	34.3331
7000	0.3152	0.7608	9.6074	4.1034	13.7395	5.0834	0.3217	15.1188	10.6651	18.5510	11.2540	0.7959	42.0139	32.2852	49.4716	33.4915
8000	0.3232	0.7680	9.8453	3.9258	13.7993	4.8128	0.3161	15.1877	10.7202	18.5356	11.2959	0.7938	42.1829	32.6832	49.6256	34.2256
9000	0.3376	0.7386	8.9827	3.4967	12.8734	4.4262	0.3082	14.8650	10.5644	18.2372	11.1377	0.7793	41.7501	32.6252	49.4924	33.8756
10000	0.3152	0.7408	9.0422	3.7335	13.5006	4.6678	0.3068	14.7458	10.4670	18.1324	11.0457	0.7773	41.3750	31.9683	49.1887	33.1957
11000	0.3167	0.7246	9.2802	3.7976	13.2318	4.7354	0.3010	14.4811	10.3391	17.8357	10.9036	0.7745	41.2926	31.8771	48.8018	33.0807
12000	0.3087	0.7240	8.7745	3.6447	12.7838	4.5712	0.2996	14.2912	10.2057	17.6353	10.7722	0.7683	41.1525	32.0549	48.8018	33.2402
13000	0.3330	0.7061	8.3284	3.5362	12.4851	4.4552	0.2981	14.3397	10.2971	17.7257	10.8680	0.7642	41.1401	32.0379	48.7685	33.1949
14000	0.3176	0.6887	8.8935	3.7680	12.8734	4.6726	0.2940	14.1728	10.2323	17.5270	10.8029	0.7618	39.9942	30.7597	47.3873	31.9470
15000	0.3113	0.7296	8.7151	3.8617	13.0526	4.8224	0.2924	14.0962	10.1638	17.4495	10.7350	0.7595	40.2951	31.0757	47.7201	32.2502
16000	0.3058	0.6820	9.2802	4.0688	13.2019	4.9481	0.2922	14.0766	10.1057	17.4222	10.6784	0.7544	40.4641	31.5116	47.9947	32.7092
17000	0.3030	0.7120	8.9233	3.8913	13.0824	4.8321	0.2878	14.1413	10.1954	17.4785	10.7684	0.7559	40.3487	31.4791	48.0113	32.6455
18000	0.3080	0.6951	8.5961	3.7138	12.5747	4.6533	0.2863	13.8595	9.9432	17.1562	10.5007	0.7467	40.0437	31.2512	47.5745	32.4163
19000	0.3104	0.6771	8.5961	3.6743	12.7838	4.6050	0.2854	13.9702	10.0538	17.2858	10.6153	0.7477	40.2003	31.3663	47.6743	32.5098
19999	0.3089	0.7173	8.2391	3.7039	12.6643	4.6485
19999	0.2845	0.7173	13.9625	10.0306	17.2389	10.5844
19999	0.7447	0.7173	40.1880	31.3161	47.6494	32.4497

Framework versions

• Transformers 4.31.0.dev0
• Datasets 2.13.0
• Tokenizers 0.13.3