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

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

• Loss: 0.1873
• Wer: 0.1752

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: 2
• 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: 500
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.5827	0.76	600	0.3985	0.5778
0.2963	1.51	1200	0.2492	0.4142
0.2122	2.27	1800	0.2132	0.3133
0.1686	3.03	2400	0.2078	0.2851
0.1373	3.79	3000	0.1910	0.2750
0.1254	4.54	3600	0.1850	0.2619
0.1137	5.3	4200	0.1874	0.2503
0.1008	6.06	4800	0.1857	0.2554
0.0934	6.81	5400	0.1844	0.2404
0.0876	7.57	6000	0.2001	0.2375
0.0801	8.33	6600	0.2036	0.2512
0.0732	9.09	7200	0.1921	0.2301
0.069	9.84	7800	0.1821	0.2330
0.0628	10.6	8400	0.1915	0.2249
0.0619	11.36	9000	0.1881	0.2113
0.0549	12.11	9600	0.1920	0.2076
0.0524	12.87	10200	0.1901	0.2079
0.0492	13.63	10800	0.1767	0.2020
0.0445	14.38	11400	0.1852	0.1933
0.0427	15.14	12000	0.1995	0.1994
0.0398	15.9	12600	0.1922	0.1932
0.037	16.66	13200	0.1956	0.1920
0.0353	17.41	13800	0.1990	0.1909
0.0327	18.17	14400	0.1906	0.1784
0.0311	18.93	15000	0.1847	0.1765
0.0295	19.68	15600	0.1873	0.1752

Framework versions

• Transformers 4.32.1
• Pytorch 2.0.1+cu118
• Datasets 2.14.4
• Tokenizers 0.13.3