wav2vec2-E50_freq_speed

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.2867
• Cer: 26.4509

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: 50
• num_epochs: 3
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
36.1912	0.1289	200	4.9281	100.0
4.8744	0.2579	400	4.6344	100.0
4.7389	0.3868	600	4.6396	100.0
4.7053	0.5158	800	4.6142	100.0
4.6271	0.6447	1000	4.5886	98.9779
4.5325	0.7737	1200	4.3545	97.4213
4.0987	0.9026	1400	3.3740	62.2768
3.1619	1.0316	1600	2.8281	48.0733
2.8074	1.1605	1800	2.4434	44.3257
2.5099	1.2895	2000	2.2456	40.7542
2.3202	1.4184	2200	2.0216	38.0169
2.2438	1.5474	2400	1.8903	35.2796
2.0245	1.6763	2600	1.8335	34.7098
1.9285	1.8053	2800	1.7468	34.4690
1.83	1.9342	3000	1.5999	31.0503
1.6842	2.0632	3200	1.5379	30.3454
1.5576	2.1921	3400	1.4967	29.8755
1.4787	2.3211	3600	1.3937	28.0721
1.458	2.4500	3800	1.3861	27.7079
1.3927	2.5790	4000	1.2945	26.1631
1.3578	2.7079	4200	1.3155	27.2968
1.3148	2.8369	4400	1.2744	26.1454
1.3245	2.9658	4600	1.2867	26.4509

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.19.1