wav2vec2-ft-datastf5-lre5-adm-ga2b16-st30k-pat3

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

• Loss: 1.4015
• Wer: 0.6044
• Cer: 0.3458

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: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1500
• training_steps: 30000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.2953	1.2393	1000	3.1551	0.9996	0.9865
3.0702	2.4786	2000	3.0619	0.9996	0.9865
2.9863	3.7179	3000	2.9576	0.9996	0.9865
2.0805	4.9572	4000	1.8728	0.7278	0.4246
1.9027	6.1959	5000	1.6639	0.6841	0.4019
1.8786	7.4352	6000	1.6347	0.6739	0.3915
1.7726	8.6745	7000	1.6081	0.6614	0.3830
1.7444	9.9138	8000	1.5278	0.6506	0.3755
1.553	11.1525	9000	1.5127	0.6455	0.3731
1.6228	12.3918	10000	1.4911	0.6353	0.3677
1.6086	13.6311	11000	1.4971	0.6305	0.3630
1.5643	14.8704	12000	1.4366	0.6278	0.3593
1.4588	16.1091	13000	1.4859	0.6273	0.3586
1.5213	17.3484	14000	1.4326	0.6207	0.3564
1.4723	18.5877	15000	1.4080	0.6160	0.3546
1.4055	19.8270	16000	1.4105	0.6143	0.3525
1.4283	21.0657	17000	1.3965	0.6109	0.3507
1.3166	22.3050	18000	1.4271	0.6106	0.3474
1.3827	23.5443	19000	1.4208	0.6068	0.3467
1.3347	24.7836	20000	1.4015	0.6044	0.3458

Framework versions

• Transformers 4.52.3
• Pytorch 2.7.0+cu118
• Datasets 3.6.0
• Tokenizers 0.21.1