wav2vec2-1b-E10_freq_pause

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: 0.7622
• Cer: 19.5371

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
8.7376	0.2580	200	4.2865	79.6581
1.9384	0.5160	400	2.1712	51.7211
1.2999	0.7741	600	2.0590	47.0688
1.086	1.0321	800	1.3574	34.8214
0.8407	1.2901	1000	1.3538	32.7244
0.7729	1.5481	1200	1.2432	30.6450
0.693	1.8062	1400	1.0388	27.3026
0.5904	2.0642	1600	1.1171	28.5597
0.4983	2.3222	1800	1.0724	27.1323
0.4467	2.5802	2000	0.9789	27.1910
0.4194	2.8383	2200	0.9319	24.7533
0.37	3.0963	2400	0.8807	22.3802
0.3058	3.3543	2600	0.8707	22.7737
0.2883	3.6123	2800	0.8241	21.1349
0.2757	3.8703	3000	0.7938	20.4535
0.2276	4.1284	3200	0.8549	21.4344
0.2027	4.3864	3400	0.7780	19.9542
0.18	4.6444	3600	0.7515	19.2258
0.1783	4.9024	3800	0.7622	19.5371

Framework versions

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