wav2vec2-xlsr-ln-50hr-v1

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5500
• Model Preparation Time: 0.0092
• Wer: 0.2237
• Cer: 0.0739

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: 0.0003
• train_batch_size: 16
• eval_batch_size: 8
• 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: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 120
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Wer	Cer
4.1139	0.9986	362	0.5868	0.0092	0.4335	0.1216
0.3001	2.0	725	0.3474	0.0092	0.2614	0.0792
0.2033	2.9986	1087	0.3256	0.0092	0.2023	0.0670
0.1629	4.0	1450	0.3155	0.0092	0.2089	0.0641
0.1366	4.9986	1812	0.2904	0.0092	0.1899	0.0577
0.1182	6.0	2175	0.2895	0.0092	0.1864	0.0572
0.1064	6.9986	2537	0.2815	0.0092	0.1671	0.0535
0.0945	8.0	2900	0.3037	0.0092	0.1706	0.0559
0.0845	8.9986	3262	0.3142	0.0092	0.1743	0.0581
0.0779	10.0	3625	0.3031	0.0092	0.1758	0.0572
0.0754	10.9986	3987	0.3111	0.0092	0.1704	0.0568
0.0687	12.0	4350	0.3130	0.0092	0.1664	0.0539
0.0582	12.9986	4712	0.3364	0.0092	0.1619	0.0526
0.0552	14.0	5075	0.3039	0.0092	0.1568	0.0527
0.054	14.9986	5437	0.3176	0.0092	0.1561	0.0507
0.0453	16.0	5800	0.3283	0.0092	0.1550	0.0519
0.046	16.9986	6162	0.3320	0.0092	0.1556	0.0504
0.0443	18.0	6525	0.3443	0.0092	0.1560	0.0510
0.0441	18.9986	6887	0.3392	0.0092	0.1549	0.0518
0.0375	20.0	7250	0.3526	0.0092	0.1565	0.0529
0.0371	20.9986	7612	0.3552	0.0092	0.1574	0.0541
0.0412	22.0	7975	0.3313	0.0092	0.1762	0.0565
0.041	22.9986	8337	0.3649	0.0092	0.1695	0.0572
0.0377	24.0	8700	0.3603	0.0092	0.1578	0.0532
0.0332	24.9986	9062	0.3496	0.0092	0.1513	0.0509
0.032	26.0	9425	0.3436	0.0092	0.1504	0.0517
0.0314	26.9986	9787	0.3573	0.0092	0.1545	0.0523
0.0281	28.0	10150	0.3644	0.0092	0.1504	0.0504
0.0268	28.9986	10512	0.3628	0.0092	0.1521	0.0506
0.0304	30.0	10875	0.3692	0.0092	0.1512	0.0510
0.0296	30.9986	11237	0.3573	0.0092	0.1493	0.0505
0.023	32.0	11600	0.3767	0.0092	0.1562	0.0516
0.0292	32.9986	11962	0.3462	0.0092	0.1496	0.0492
0.0261	34.0	12325	0.3927	0.0092	0.1500	0.0490
0.0248	34.9986	12687	0.3771	0.0092	0.1438	0.0492
0.0238	36.0	13050	0.3763	0.0092	0.1457	0.0474
0.0223	36.9986	13412	0.3627	0.0092	0.1523	0.0510
0.0225	38.0	13775	0.3825	0.0092	0.1468	0.0494
0.022	38.9986	14137	0.3830	0.0092	0.1614	0.0547
0.0226	40.0	14500	0.3851	0.0092	0.1488	0.0509
0.0225	40.9986	14862	0.4072	0.0092	0.1592	0.0530
0.0197	42.0	15225	0.4024	0.0092	0.1460	0.0502
0.0205	42.9986	15587	0.4099	0.0092	0.1491	0.0510
0.0195	44.0	15950	0.3746	0.0092	0.1449	0.0501
0.0187	44.9986	16312	0.3902	0.0092	0.1417	0.0487
0.0196	46.0	16675	0.3923	0.0092	0.1453	0.0497
0.0177	46.9986	17037	0.4107	0.0092	0.1458	0.0490
0.0175	48.0	17400	0.4043	0.0092	0.1478	0.0503
0.0178	48.9986	17762	0.4009	0.0092	0.1450	0.0514
0.0161	50.0	18125	0.4172	0.0092	0.1374	0.0472
0.015	50.9986	18487	0.4006	0.0092	0.1342	0.0463
0.015	52.0	18850	0.3975	0.0092	0.1399	0.0492
0.0173	52.9986	19212	0.3690	0.0092	0.1399	0.0493
0.0156	54.0	19575	0.4321	0.0092	0.1439	0.0504
0.0151	54.9986	19937	0.4353	0.0092	0.1443	0.0508
0.0151	56.0	20300	0.3784	0.0092	0.1394	0.0488
0.015	56.9986	20662	0.4225	0.0092	0.1415	0.0499
0.0128	58.0	21025	0.4172	0.0092	0.1421	0.0486
0.0124	58.9986	21387	0.3899	0.0092	0.1400	0.0479
0.0109	60.0	21750	0.4265	0.0092	0.1364	0.0468
0.0109	60.9986	22112	0.4143	0.0092	0.1400	0.0486
0.0118	62.0	22475	0.4204	0.0092	0.1446	0.0495
0.0125	62.9986	22837	0.4020	0.0092	0.1367	0.0472

Framework versions

• Transformers 4.43.3
• Pytorch 2.1.0+cu118
• Datasets 2.20.0
• Tokenizers 0.19.1