xls-r-300m-model-C

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: 0.3299
• Wer: 0.4263
• Cer: 0.2096

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
26.9708	1.0	47	11.3572	1.0	1.0
8.8174	2.0	94	4.0787	1.0	1.0
3.6382	3.0	141	3.1642	1.0	1.0
3.1662	4.0	188	3.1463	1.0	1.0
3.1612	5.0	235	3.1479	1.0	1.0
3.1256	6.0	282	3.0059	1.0	1.0
2.9892	7.0	329	3.1620	1.0	1.0
2.8792	8.0	376	2.7184	1.0	1.0
2.7882	9.0	423	2.6526	1.0	0.8751
2.6556	10.0	470	2.5504	1.0	0.8299
2.5528	11.0	517	2.4123	1.0	0.8659
2.416	12.0	564	2.2641	1.0	0.7833
2.2796	13.0	611	2.1311	1.0	0.7459
2.1592	14.0	658	1.9811	1.0	0.7234
2.1128	15.0	705	2.0185	1.0	0.7622
2.0422	16.0	752	1.9119	1.0	0.7565
2.0077	17.0	799	1.8053	1.0	0.7530
1.9033	18.0	846	1.7593	1.0031	0.7198
1.8717	19.0	893	1.6809	0.9937	0.6443
1.7898	20.0	940	1.6353	0.9624	0.6683
1.7377	21.0	987	1.5580	1.0094	0.7114
1.6328	22.0	1034	1.3971	0.9718	0.6069
1.596	23.0	1081	1.3844	1.0784	0.6013
1.5422	24.0	1128	1.2812	0.9028	0.4989
1.3451	25.0	1175	1.2919	0.9279	0.5695
1.246	26.0	1222	1.1572	0.8871	0.5039
1.237	27.0	1269	1.2349	0.9028	0.5349
1.1147	28.0	1316	0.9799	0.8401	0.4771
0.9404	29.0	1363	0.9032	0.7743	0.4390
0.9353	30.0	1410	0.7554	0.8088	0.4700
0.8467	31.0	1457	0.8404	0.7586	0.4107
0.8398	32.0	1504	0.7486	0.7618	0.4234
0.7437	33.0	1551	0.6802	0.6897	0.3529
0.606	34.0	1598	0.7355	0.6959	0.3677
0.6243	35.0	1645	0.7037	0.6708	0.3331
0.5461	36.0	1692	0.7073	0.6207	0.2886
0.5483	37.0	1739	0.6538	0.6301	0.3140
0.5257	38.0	1786	0.5984	0.5674	0.3296
0.4714	39.0	1833	0.5388	0.5956	0.3162
0.4328	40.0	1880	0.4635	0.5737	0.3183
0.4192	41.0	1927	0.6038	0.5768	0.3126
0.3659	42.0	1974	0.5909	0.5705	0.2908
0.3587	43.0	2021	0.3662	0.4639	0.2385
0.3755	44.0	2068	0.4218	0.4859	0.2590
0.2926	45.0	2115	0.3366	0.4890	0.2449
0.3428	46.0	2162	0.4311	0.5016	0.2745
0.301	47.0	2209	0.3943	0.4828	0.2604
0.262	48.0	2256	0.4424	0.4796	0.2414
0.308	49.0	2303	0.3152	0.4639	0.2329
0.2434	50.0	2350	0.4463	0.4671	0.2392
0.1922	51.0	2397	0.4145	0.4483	0.2308
0.1913	52.0	2444	0.3989	0.4295	0.2082
0.2115	53.0	2491	0.3491	0.4545	0.2124
0.215	54.0	2538	0.4084	0.4451	0.2329
0.1854	55.0	2585	0.4447	0.4451	0.2399
0.1556	56.0	2632	0.4437	0.4702	0.2336
0.1564	57.0	2679	0.4627	0.4389	0.2188
0.1502	58.0	2726	0.4350	0.4295	0.2145
0.1571	59.0	2773	0.3444	0.4295	0.2096
0.1319	60.0	2820	0.3299	0.4263	0.2096

Framework versions

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