wav2vec2-base-cv-10000

This model is a fine-tuned version of jiobiala24/wav2vec2-base-cv on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 1.3393
• Wer: 0.3684

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: 0.0001
• 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_steps: 1000
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4243	1.6	1000	0.7742	0.4210
0.3636	3.2	2000	0.8621	0.4229
0.2638	4.8	3000	0.9328	0.4094
0.2273	6.4	4000	0.9556	0.4087
0.187	8.0	5000	0.9093	0.4019
0.1593	9.6	6000	0.9842	0.4029
0.1362	11.2	7000	1.0651	0.4077
0.1125	12.8	8000	1.0550	0.3959
0.103	14.4	9000	1.1919	0.4002
0.0948	16.0	10000	1.1901	0.3983
0.0791	17.6	11000	1.1091	0.3860
0.0703	19.2	12000	1.2823	0.3904
0.0641	20.8	13000	1.2625	0.3817
0.057	22.4	14000	1.2821	0.3776
0.0546	24.0	15000	1.2975	0.3770
0.0457	25.6	16000	1.2998	0.3714
0.0433	27.2	17000	1.3574	0.3721
0.0423	28.8	18000	1.3393	0.3684

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.0+cu111
• Datasets 1.13.3
• Tokenizers 0.10.3