wav2vec2-large-xls-r-300m-cs-colab

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1447
• Wer: 0.1342

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: 5e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.3706	0.5485	500	3.4060	1.0
3.2801	1.0971	1000	3.2557	1.0
0.7687	1.6456	1500	0.4183	0.3893
0.515	2.1942	2000	0.2665	0.2576
0.4229	2.7427	2500	0.2231	0.2245
0.345	3.2913	3000	0.2010	0.2019
0.3364	3.8398	3500	0.1865	0.1874
0.2881	4.3884	4000	0.1759	0.1741
0.304	4.9369	4500	0.1684	0.1693
0.2849	5.4855	5000	0.1695	0.1683
0.2492	6.0340	5500	0.1634	0.1596
0.2419	6.5826	6000	0.1606	0.1569
0.2282	7.1311	6500	0.1566	0.1520
0.2233	7.6796	7000	0.1549	0.1506
0.2063	8.2282	7500	0.1537	0.1489
0.2045	8.7767	8000	0.1541	0.1499
0.2127	9.3253	8500	0.1545	0.1446
0.1954	9.8738	9000	0.1509	0.1422
0.1931	10.4224	9500	0.1532	0.1419
0.2022	10.9709	10000	0.1472	0.1404
0.1868	11.5195	10500	0.1472	0.1373
0.1852	12.0680	11000	0.1469	0.1377
0.1817	12.6166	11500	0.1473	0.1374
0.1941	13.1651	12000	0.1443	0.1360
0.1632	13.7137	12500	0.1464	0.1342
0.1705	14.2622	13000	0.1451	0.1348
0.1725	14.8108	13500	0.1447	0.1342

Framework versions

• Transformers 4.45.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.2
• Tokenizers 0.20.1