wav2vec2-1b-E30_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: 0.5370
• Cer: 14.0038

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
10.2692	0.2580	200	3.0964	66.8527
1.6691	0.5160	400	1.5270	38.1461
1.1317	0.7741	600	1.3893	32.4894
0.9089	1.0321	800	1.0586	26.2394
0.7259	1.2901	1000	0.9065	21.8633
0.6488	1.5481	1200	0.8797	22.4977
0.5697	1.8062	1400	0.8176	20.8118
0.5042	2.0642	1600	0.8276	21.7693
0.4017	2.3222	1800	0.7741	20.5945
0.3848	2.5802	2000	0.7376	19.0848
0.366	2.8383	2200	0.7161	18.7441
0.3035	3.0963	2400	0.6149	16.0538
0.2463	3.3543	2600	0.6609	17.2991
0.2197	3.6123	2800	0.6462	16.7000
0.2193	3.8703	3000	0.6023	15.5251
0.1859	4.1284	3200	0.5970	15.5369
0.1645	4.3864	3400	0.5764	14.9319
0.1441	4.6444	3600	0.5557	14.1976
0.1459	4.9024	3800	0.5370	14.0038

Framework versions

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