wav2vec2-large-xls-r-300m-Mezge

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3245
• Wer: 0.1777

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: 0.0003
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
29.8978	1.9242	400	3.4282	1.0
8.319	3.8472	800	0.5659	0.5170
1.6729	5.7702	1200	0.3004	0.2911
0.8499	7.6931	1600	0.2747	0.2542
0.5621	9.6161	2000	0.2893	0.2320
0.4271	11.5391	2400	0.2720	0.2185
0.3494	13.4621	2800	0.2883	0.2143
0.2881	15.3851	3200	0.3053	0.2050
0.2588	17.3081	3600	0.3074	0.1977
0.2287	19.2310	4000	0.3137	0.1924
0.1936	21.1540	4400	0.3121	0.1907
0.1728	23.0770	4800	0.3246	0.1847
0.1567	25.0	5200	0.3268	0.1855
0.1359	26.9242	5600	0.3212	0.1798
0.1199	28.8472	6000	0.3245	0.1777

Framework versions

• Transformers 4.47.1
• Pytorch 2.2.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0