w2v-bert-2.0-ln-afrivoice-10hr-v3

This model is a fine-tuned version of facebook/w2v-bert-2.0 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5394
• Model Preparation Time: 0.0157
• Wer: 0.2805
• Cer: 0.0704

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: 5e-05
• 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_ratio: 0.033
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Wer	Cer
4.8449	0.9919	61	2.7369	0.0157	0.9995	0.9870
1.9412	2.0	123	0.9095	0.0157	0.4841	0.1700
0.7831	2.9919	184	0.8914	0.0157	0.4018	0.1424
0.67	4.0	246	0.7633	0.0157	0.3919	0.1391
0.5982	4.9919	307	0.8712	0.0157	0.3514	0.1349
0.5577	6.0	369	0.6596	0.0157	0.4425	0.1566
0.4945	6.9919	430	0.7157	0.0157	0.3838	0.1419
0.4363	8.0	492	0.7981	0.0157	0.3582	0.1324
0.395	8.9919	553	0.7956	0.0157	0.3483	0.1310
0.3416	10.0	615	0.7110	0.0157	0.4082	0.1571
0.3181	10.9919	676	0.8728	0.0157	0.3680	0.1334
0.2837	12.0	738	0.8389	0.0157	0.3656	0.1361
0.2482	12.9919	799	0.9984	0.0157	0.3582	0.1296
0.224	14.0	861	0.8696	0.0157	0.3971	0.1515
0.204	14.9919	922	1.0671	0.0157	0.3563	0.1312
0.1665	16.0	984	1.0956	0.0157	0.3622	0.1329
0.1507	16.9919	1045	1.4699	0.0157	0.3481	0.1297
0.1144	18.0	1107	1.4821	0.0157	0.3566	0.1299
0.1327	18.9919	1168	1.2253	0.0157	0.3699	0.1352
0.1085	20.0	1230	1.2042	0.0157	0.3929	0.1452
0.0694	20.9919	1291	1.4515	0.0157	0.3681	0.1317
0.0476	22.0	1353	1.5795	0.0157	0.3551	0.1301
0.0357	22.9919	1414	1.5949	0.0157	0.3527	0.1300
0.0241	24.0	1476	1.7094	0.0157	0.3555	0.1304
0.017	24.9919	1537	1.7941	0.0157	0.3577	0.1311
0.0128	26.0	1599	1.8157	0.0157	0.3555	0.1300
0.0132	26.9919	1660	1.8541	0.0157	0.3621	0.1324

Framework versions

• Transformers 4.44.0
• Pytorch 2.1.0+cu118
• Datasets 2.20.0
• Tokenizers 0.19.1