wav2vec2-E50_freq_speed_pause

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: 2.4582
• Cer: 51.0397

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
31.096	0.1289	200	4.8982	100.0
4.8903	0.2579	400	4.6629	100.0
4.8136	0.3868	600	4.6742	100.0
4.7638	0.5158	800	4.6152	100.0
4.6775	0.6447	1000	4.5802	97.7385
4.6512	0.7737	1200	4.5786	97.5799
4.6319	0.9026	1400	4.5442	97.3391
4.5669	1.0316	1600	4.5205	97.1041
4.5562	1.1605	1800	4.5660	96.1701
4.5185	1.2895	2000	4.5936	95.9410
4.5125	1.4184	2200	4.4987	95.9821
4.4809	1.5474	2400	4.6603	96.1349
4.4064	1.6763	2600	4.4259	96.4051
4.3209	1.8053	2800	4.2883	95.4828
4.1089	1.9342	3000	3.9987	78.3717
3.6228	2.0632	3200	3.5262	68.7324
3.2898	2.1921	3400	3.1194	63.9391
3.0643	2.3211	3600	2.9216	59.5277
2.9142	2.4500	3800	2.7525	56.1678
2.7786	2.5790	4000	2.6561	54.2528
2.6701	2.7079	4200	2.6118	54.0179
2.6038	2.8369	4400	2.5045	52.1793
2.5875	2.9658	4600	2.4582	51.0397

Framework versions

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