xlsr-sl-adap-ru

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: 0.4726
• Wer: 0.4518
• Cer: 0.1009

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: 3e-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: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
12.5552	2.2989	100	13.9758	1.0189	0.8746
3.2145	4.5977	200	3.1893	1.0	1.0
3.0478	6.8966	300	3.0401	1.0	1.0
2.8114	9.1954	400	2.7815	1.0	1.0
1.6159	11.4943	500	1.1846	0.8243	0.2357
0.7215	13.7931	600	0.6544	0.6268	0.1437
0.4679	16.0920	700	0.5544	0.5789	0.1304
0.5392	18.3908	800	0.5222	0.5383	0.1212
0.3497	20.6897	900	0.4828	0.5157	0.1153
0.3216	22.9885	1000	0.4861	0.5118	0.1128
0.3924	25.2874	1100	0.4713	0.4879	0.1088
0.3363	27.5862	1200	0.4800	0.4759	0.1059
0.3672	29.8851	1300	0.4664	0.4725	0.1046
0.2735	32.1839	1400	0.4708	0.4725	0.1046
0.2	34.4828	1500	0.4790	0.4652	0.1040
0.3515	36.7816	1600	0.4688	0.4577	0.1018
0.3233	39.0805	1700	0.4734	0.4593	0.1019
0.2715	41.3793	1800	0.4670	0.4516	0.1003
0.2793	43.6782	1900	0.4727	0.4556	0.1014
0.2367	45.9770	2000	0.4729	0.4552	0.1008
0.2503	48.2759	2100	0.4726	0.4518	0.1009

Framework versions

• Transformers 4.42.0.dev0
• Pytorch 2.3.1+cu121
• Datasets 2.19.2
• Tokenizers 0.19.1