wav2vec2-ft-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: 2.4389
• Wer: 0.8090
• Cer: 0.5184

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
3.399	0.4199	1000	3.4616	1.0000	1.0000
3.2884	0.8398	2000	3.3397	1.0000	1.0000
3.301	1.2595	3000	3.3084	1.0000	1.0000
3.2206	1.6794	4000	3.2740	1.0000	1.0000
3.1569	2.0991	5000	3.2544	1.0000	1.0000
2.8877	2.5190	6000	2.9117	0.9468	0.6794
2.6796	2.9389	7000	2.7232	0.8832	0.5624
2.562	3.3586	8000	2.6058	0.8554	0.5545
2.5128	3.7785	9000	2.6086	0.8538	0.5401
2.4716	4.1982	10000	2.5722	0.8469	0.5392
2.4751	4.6181	11000	2.5502	0.8339	0.5310
2.3622	5.0378	12000	2.5346	0.8398	0.5284
2.3376	5.4577	13000	2.5098	0.8332	0.5259
2.2925	5.8776	14000	2.5177	0.8262	0.5221
2.3228	6.2973	15000	2.4651	0.8229	0.5325
2.3446	6.7172	16000	2.4293	0.8059	0.5261
2.2545	7.1369	17000	2.4779	0.8159	0.5194
2.2923	7.5568	18000	2.5485	0.8336	0.5153
2.2556	7.9767	19000	2.4389	0.8090	0.5184

Framework versions

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