w2v-bert-2.0-yogera-fleurs-cv-lg-173hrs-v2

This model is a fine-tuned version of facebook/w2v-bert-2.0 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3119
• Wer: 0.2412
• Cer: 0.0552

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: 8
• eval_batch_size: 4
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.2605	1.0	11368	0.3181	0.3125	0.0723
0.1703	2.0	22736	0.2794	0.2773	0.0620
0.1428	3.0	34104	0.2543	0.2740	0.0614
0.1256	4.0	45472	0.2769	0.2652	0.0584
0.1135	5.0	56840	0.2611	0.2559	0.0576
0.1025	6.0	68208	0.2578	0.2551	0.0582
0.0936	7.0	79576	0.2717	0.2505	0.0563
0.0855	8.0	90944	0.2509	0.2458	0.0555
0.0785	9.0	102312	0.2566	0.2449	0.0551
0.0708	10.0	113680	0.2623	0.2408	0.0557
0.0637	11.0	125048	0.2957	0.2441	0.0557
0.0563	12.0	136416	0.3498	0.2486	0.0571
0.0493	13.0	147784	0.2862	0.2513	0.0565
0.0424	14.0	159152	0.3150	0.2486	0.0569
0.0367	15.0	170520	0.4020	0.2465	0.0559
0.0324	16.0	181888	0.3398	0.2472	0.0571
0.0279	17.0	193256	0.3318	0.2444	0.0552
0.0246	18.0	204624	0.3321	0.2464	0.0561
0.022	19.0	215992	0.3338	0.2472	0.0578
0.0197	20.0	227360	0.3119	0.2412	0.0552

Framework versions

• Transformers 4.45.2
• Pytorch 2.1.0+cu118
• Datasets 3.0.1
• Tokenizers 0.20.1