facebook_large_CV_bn2

This model is a fine-tuned version of Sameen53/facebook_large_CV_bn3 on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 0.3172
• Wer: 0.2524

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: linear
• lr_scheduler_warmup_steps: 200
• num_epochs: 16
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.92	500	0.4366	0.2610
0.0648	1.85	1000	0.3689	0.2563
0.097	2.77	1500	0.3613	0.2677
0.0998	3.7	2000	0.3669	0.2573
0.1003	4.62	2500	0.3472	0.2520
0.1017	5.54	3000	0.3635	0.2559
0.1052	6.47	3500	0.3546	0.2559
0.1128	7.39	4000	0.3439	0.2554
0.1136	8.32	4500	0.3483	0.2515
0.1178	9.24	5000	0.3469	0.2526
0.1217	10.17	5500	0.3408	0.2523
0.1269	11.09	6000	0.3420	0.2556
0.1273	12.01	6500	0.3331	0.2533
0.1273	12.94	7000	0.3344	0.2531
0.136	13.86	7500	0.3302	0.2551
0.1397	14.79	8000	0.3252	0.2522
0.1447	15.71	8500	0.3172	0.2524

Framework versions

• Transformers 4.20.1
• Pytorch 1.11.0
• Datasets 2.1.0
• Tokenizers 0.12.1