wav2vec2-1b-E30_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: 0.7420
• Cer: 19.3492

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
14.0949	0.2580	200	4.8294	96.0056
4.5868	0.5160	400	4.6642	92.5634
2.6944	0.7741	600	2.0854	49.9001
1.4258	1.0321	800	1.4597	37.8877
1.1378	1.2901	1000	1.3179	36.4603
1.0007	1.5481	1200	1.3876	33.3059
0.9293	1.8062	1400	1.1012	28.2895
0.7962	2.0642	1600	1.2496	30.9445
0.7125	2.3222	1800	1.1189	27.5670
0.6505	2.5802	2000	1.2186	30.6861
0.6209	2.8383	2200	0.9885	24.7885
0.5445	3.0963	2400	0.9220	22.9793
0.4863	3.3543	2600	0.8563	21.7105
0.4351	3.6123	2800	0.8386	21.6283
0.4201	3.8703	3000	0.8432	21.3875
0.3752	4.1284	3200	0.8012	20.9293
0.3281	4.3864	3400	0.7879	20.5063
0.3045	4.6444	3600	0.8009	20.2361
0.2893	4.9024	3800	0.7420	19.3492

Framework versions

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