small_finetune_CM01

This model is a fine-tuned version of facebook/wav2vec2-base on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.9764
• Wer: 1.0

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: 20
• 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: 700
• num_epochs: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
40.8946	100.0	700	3.6118	1.0
3.1203	200.0	1400	3.4805	1.0
2.2986	300.0	2100	2.6437	1.0
1.9839	400.0	2800	2.3215	1.0
1.7708	500.0	3500	2.2025	1.0
1.6458	600.0	4200	2.1303	1.0
1.5592	700.0	4900	2.0806	1.0
1.4861	800.0	5600	2.0559	1.0
1.4415	900.0	6300	2.0361	1.0
1.3975	1000.0	7000	2.0248	1.0
1.3595	1100.0	7700	2.0246	1.0
1.3413	1200.0	8400	1.9947	1.0
1.3142	1300.0	9100	2.0000	1.0
1.2986	1400.0	9800	1.9909	1.0
1.2823	1500.0	10500	1.9862	1.0
1.2724	1600.0	11200	1.9851	1.0
1.2588	1700.0	11900	1.9862	1.0
1.2599	1800.0	12600	1.9800	1.0
1.2536	1900.0	13300	1.9788	1.0
1.2551	2000.0	14000	1.9764	1.0

Framework versions

• Transformers 4.18.0
• Pytorch 1.10.2+cu102
• Datasets 2.3.2
• Tokenizers 0.12.1