XLS-R_Synthesis_ALL_v2

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: 0.1577
• Wer: 0.1671

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
4.5648	1.0	2611	1.3992	0.9921
0.7741	2.0	5223	0.3761	0.4506
0.431	3.0	7834	0.2574	0.3203
0.3323	4.0	10446	0.2205	0.2609
0.2818	5.0	13057	0.2050	0.2260
0.2514	6.0	15669	0.1896	0.2105
0.2318	7.0	18280	0.1766	0.2024
0.2189	8.0	20892	0.1736	0.1968
0.2098	9.0	23503	0.1755	0.1917
0.2024	10.0	26115	0.1707	0.1931
0.1966	11.0	28726	0.1636	0.1871
0.1893	12.0	31338	0.1719	0.1839
0.1808	13.0	33949	0.1684	0.1815
0.1756	14.0	36561	0.1631	0.1768
0.1702	15.0	39172	0.1670	0.1757
0.1651	16.0	41784	0.1627	0.1718
0.1596	17.0	44395	0.1572	0.1683
0.1553	18.0	47007	0.1614	0.1675
0.1528	19.0	49618	0.1701	0.1723
0.1502	20.0	52230	0.1598	0.1654
0.1839	21.0	54841	0.2059	0.2144
0.2445	22.0	57453	0.2215	0.2463
0.2002	23.0	60064	0.1703	0.1788
0.1534	24.0	62676	0.1634	0.1698
0.1449	25.0	65287	0.1577	0.1671

Framework versions

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