wav2vec2-1b-E50_freq_pause_speed

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.9857
• Cer: 25.4112

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: 2
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
19.722	0.2580	200	10.6272	92.3226
9.1071	0.5160	400	4.8280	97.6034
4.6864	0.7741	600	5.1023	93.8381
4.894	1.0321	800	4.4428	92.8983
4.5268	1.2901	1000	4.4166	92.9864
4.4975	1.5481	1200	4.3971	92.8160
4.3905	1.8062	1400	4.2999	92.3931
4.0944	2.0642	1600	4.0658	89.2505
3.4243	2.3222	1800	3.5627	74.0073
2.5159	2.5802	2000	2.6364	54.6346
1.9341	2.8383	2200	2.1328	45.0070
1.4936	3.0963	2400	1.7956	41.7939
1.1786	3.3543	2600	1.4282	34.3163
1.0121	3.6123	2800	1.2753	30.2162
0.9051	3.8703	3000	1.2181	30.0164
0.7404	4.1284	3200	1.1032	29.0061
0.6383	4.3864	3400	1.0560	24.8884
0.609	4.6444	3600	0.9713	24.7063
0.5477	4.9024	3800	0.9857	25.4112

Framework versions

• Transformers 4.45.2
• Pytorch 2.3.1.post100
• Datasets 2.19.1
• Tokenizers 0.20.1