xls-r-300m-model-1_NEW

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: 0.0024
• Wer: 0.1630
• Cer: 0.0416

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: 8
• 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_ratio: 0.1
• num_epochs: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
29.2334	1.0	47	15.0050	1.0	1.0
9.4438	2.0	94	4.0843	1.0	1.0
3.6591	3.0	141	3.1691	1.0	1.0
3.1767	4.0	188	3.1468	1.0	1.0
3.1669	5.0	235	3.1420	1.0	1.0
3.173	6.0	282	3.0161	1.0	1.0
2.9708	7.0	329	2.8614	1.0	1.0
2.8228	8.0	376	2.8736	1.0	1.0
2.6826	9.0	423	2.4797	1.0	0.9697
2.461	10.0	470	2.1226	1.0	0.9365
1.9205	11.0	517	1.4120	1.0	0.8236
1.3314	12.0	564	1.0121	1.0	0.5688
0.9524	13.0	611	0.5546	0.9969	0.4305
0.6561	14.0	658	0.3299	0.7241	0.1927
0.5809	15.0	705	0.2833	0.5329	0.1454
0.3833	16.0	752	0.1045	0.2288	0.0600
0.3026	17.0	799	0.1345	0.1975	0.0515
0.215	18.0	846	0.0719	0.0909	0.0360
0.2002	19.0	893	0.1058	0.0815	0.0296
0.2244	20.0	940	0.1058	0.1975	0.0910
0.1574	21.0	987	0.1111	0.1661	0.0614
0.1632	22.0	1034	0.0619	0.1348	0.0438
0.1399	23.0	1081	0.0577	0.0909	0.0191
0.1747	24.0	1128	0.0411	0.0846	0.0205
0.129	25.0	1175	0.0320	0.0564	0.0127
0.14	26.0	1222	0.0580	0.0470	0.0141
0.1135	27.0	1269	0.0482	0.0282	0.0085
0.1172	28.0	1316	0.0829	0.0596	0.0198
0.0736	29.0	1363	0.0688	0.0972	0.0318
0.0693	30.0	1410	0.0207	0.0815	0.0191
0.0862	31.0	1457	0.0106	0.0157	0.0035
0.0788	32.0	1504	0.0095	0.0909	0.0205
0.0524	33.0	1551	0.0102	0.1003	0.0240
0.062	34.0	1598	0.0331	0.1097	0.0275
0.0646	35.0	1645	0.0188	0.0909	0.0254
0.0793	36.0	1692	0.0148	0.1066	0.0261
0.0607	37.0	1739	0.0135	0.1379	0.0395
0.0417	38.0	1786	0.0209	0.1379	0.0360
0.0559	39.0	1833	0.0216	0.1599	0.0388
0.0668	40.0	1880	0.0136	0.1066	0.0275
0.0501	41.0	1927	0.0218	0.0940	0.0233
0.0397	42.0	1974	0.0065	0.1066	0.0261
0.0379	43.0	2021	0.0053	0.0784	0.0183
0.041	44.0	2068	0.0018	0.1411	0.0339
0.0265	45.0	2115	0.0065	0.1505	0.0346
0.0316	46.0	2162	0.0067	0.0721	0.0169
0.0401	47.0	2209	0.0169	0.1003	0.0275
0.0435	48.0	2256	0.0119	0.1129	0.0275
0.0287	49.0	2303	0.0108	0.1097	0.0254
0.0416	50.0	2350	0.0060	0.1348	0.0311
0.0295	51.0	2397	0.0078	0.1442	0.0332
0.0269	52.0	2444	0.0050	0.1285	0.0296
0.0285	53.0	2491	0.0044	0.1411	0.0332
0.0364	54.0	2538	0.0016	0.1473	0.0332
0.03	55.0	2585	0.0026	0.1536	0.0388
0.0218	56.0	2632	0.0032	0.1599	0.0423
0.0183	57.0	2679	0.0022	0.1630	0.0416
0.0238	58.0	2726	0.0018	0.1661	0.0416
0.0232	59.0	2773	0.0026	0.1661	0.0423
0.0174	60.0	2820	0.0024	0.1630	0.0416

Framework versions

• Transformers 4.28.0
• Pytorch 2.1.0+cu121
• Datasets 2.16.1
• Tokenizers 0.13.3