wav2vec2-1b-E50_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: 0.5279
• Cer: 15.9833

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
27.5942	0.2580	200	15.1915	93.3799
4.8195	0.5160	400	5.0466	93.8381
4.6163	0.7741	600	4.6972	93.6149
4.4212	1.0321	800	4.1616	89.2270
3.6842	1.2901	1000	2.4339	49.4596
1.7035	1.5481	1200	1.3759	31.8550
1.1079	1.8062	1400	1.1419	29.9460
0.8743	2.0642	1600	1.0240	27.7256
0.6885	2.3222	1800	0.9708	28.9356
0.6163	2.5802	2000	0.8797	27.3555
0.5719	2.8383	2200	0.7727	24.1835
0.4769	3.0963	2400	0.7156	23.4962
0.384	3.3543	2600	0.6899	20.6180
0.3428	3.6123	2800	0.6663	21.0291
0.3288	3.8703	3000	0.5853	20.8353
0.2779	4.1284	3200	0.5770	18.0980
0.23	4.3864	3400	0.5491	16.7058
0.2244	4.6444	3600	0.5386	16.0538
0.2006	4.9024	3800	0.5279	15.9833

Framework versions

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