wav2vec2-1b-E10_freq_speed

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: 0.7340
• Cer: 19.0026

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
11.3207	0.2580	200	3.4053	86.8891
1.8939	0.5160	400	1.8047	41.0303
1.1854	0.7741	600	1.4651	34.3691
0.9917	1.0321	800	1.0588	25.9516
0.744	1.2901	1000	1.1790	27.6962
0.6919	1.5481	1200	1.0604	25.9927
0.6347	1.8062	1400	0.9300	22.9323
0.5428	2.0642	1600	0.9996	24.9648
0.4724	2.3222	1800	0.9695	23.8252
0.4267	2.5802	2000	0.9463	23.7606
0.4096	2.8383	2200	0.8589	22.4448
0.3507	3.0963	2400	0.8145	20.7707
0.2874	3.3543	2600	0.8739	22.6856
0.2767	3.6123	2800	0.8657	21.8280
0.2663	3.8703	3000	0.8732	22.0630
0.2196	4.1284	3200	0.7671	19.4843
0.1867	4.3864	3400	0.7652	19.7016
0.1685	4.6444	3600	0.7288	18.7559
0.1677	4.9024	3800	0.7340	19.0026

Framework versions

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