XLS-R_Synthesis_ALL_v1

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1635
• Wer: 0.1696

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: 18
• eval_batch_size: 9
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 2
• total_train_batch_size: 72
• total_eval_batch_size: 18
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: polynomial
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 100.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.5765	1.0	2611	1.4891	0.9957
0.7865	2.0	5223	0.3830	0.4573
0.4459	3.0	7834	0.2696	0.3355
0.3452	4.0	10446	0.2259	0.2741
0.2932	5.0	13057	0.2103	0.2403
0.2617	6.0	15669	0.1945	0.2206
0.2421	7.0	18280	0.1853	0.2170
0.2285	8.0	20892	0.1780	0.2050
0.2184	9.0	23503	0.1864	0.1997
0.2115	10.0	26115	0.1737	0.1980
0.2048	11.0	28726	0.1703	0.1926
0.195	12.0	31338	0.1830	0.1892
0.1864	13.0	33949	0.1676	0.1844
0.1785	14.0	36561	0.1609	0.1770
0.1727	15.0	39172	0.1690	0.1751
0.1676	16.0	41784	0.1639	0.1750
0.1638	17.0	44395	0.1634	0.1751
0.1594	18.0	47007	0.1659	0.1769
0.1553	19.0	49618	0.1635	0.1696

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.1+cu121
• Datasets 2.15.0
• Tokenizers 0.15.0