wav2vec2-1b-E50_speed_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: 0.4912
• Cer: 16.3886

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
19.2486	0.2580	200	7.2820	92.9041
10.1359	0.5160	400	5.3859	92.7103
5.6439	0.7741	600	4.8377	92.8748
3.771	1.0321	800	2.8214	72.4448
1.7721	1.2901	1000	1.4723	36.9126
1.1644	1.5481	1200	1.1277	28.0839
0.9193	1.8062	1400	0.9123	25.8224
0.7471	2.0642	1600	0.9125	25.1645
0.5896	2.3222	1800	0.8251	23.2025
0.5233	2.5802	2000	0.7436	20.9880
0.479	2.8383	2200	0.6931	20.5945
0.3969	3.0963	2400	0.5986	18.4445
0.3252	3.3543	2600	0.6044	18.6090
0.2941	3.6123	2800	0.5848	18.4915
0.2755	3.8703	3000	0.5531	18.5620
0.2495	4.1284	3200	0.5614	18.2507
0.2098	4.3864	3400	0.5199	17.5282
0.1991	4.6444	3600	0.5103	16.8057
0.1877	4.9024	3800	0.4912	16.3886

Framework versions

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