wav2vec2-E10_freq

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: 1.2533
• Cer: 25.5169

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
36.7712	0.1289	200	4.9631	100.0
5.0134	0.2579	400	4.6670	100.0
4.8451	0.3868	600	4.6504	100.0
4.8129	0.5158	800	4.6193	100.0
4.769	0.6447	1000	4.6958	100.0
4.7464	0.7737	1200	4.6290	100.0
4.7596	0.9026	1400	4.5727	100.0
4.6833	1.0316	1600	4.5497	100.0
4.5065	1.1605	1800	4.3251	100.0
3.8519	1.2895	2000	3.2409	59.3221
2.9075	1.4184	2200	2.6012	47.5388
2.4945	1.5474	2400	2.2988	43.0804
2.211	1.6763	2600	2.0536	38.1990
1.9965	1.8053	2800	1.8464	34.2575
1.8016	1.9342	3000	1.7530	33.5585
1.6726	2.0632	3200	1.6394	31.5966
1.5525	2.1921	3400	1.4910	28.8828
1.4379	2.3211	3600	1.4146	28.3012
1.3769	2.4500	3800	1.3507	27.0207
1.3162	2.5790	4000	1.3658	27.2498
1.2929	2.7079	4200	1.2914	26.2042
1.2405	2.8369	4400	1.2875	26.1396
1.2097	2.9658	4600	1.2533	25.5169

Framework versions

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