wav2vec2-xls-r-300m-lg-CV-Fleurs_filtered-100hrs-v11

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

• Loss: 0.4508
• Wer: 0.4631
• Cer: 0.0931

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: 4
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 70
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.8467	1.0	7125	0.4811	0.5677	0.1299
0.4252	2.0	14250	0.4469	0.5447	0.1216
0.3597	3.0	21375	0.4205	0.4940	0.1069
0.3176	4.0	28500	0.3913	0.4938	0.1051
0.2867	5.0	35625	0.3860	0.4845	0.1012
0.2644	6.0	42750	0.3783	0.4805	0.1006
0.246	7.0	49875	0.3639	0.4769	0.0988
0.2311	8.0	57000	0.3620	0.4667	0.0968
0.2174	9.0	64125	0.3594	0.4591	0.0950
0.204	10.0	71250	0.3591	0.4629	0.0973
0.1922	11.0	78375	0.3671	0.4510	0.0928
0.1816	12.0	85500	0.3671	0.4541	0.0939
0.1719	13.0	92625	0.3620	0.4738	0.0977
0.1631	14.0	99750	0.3830	0.4690	0.0961
0.1539	15.0	106875	0.3865	0.4512	0.0926
0.1459	16.0	114000	0.3894	0.4723	0.0956
0.1387	17.0	121125	0.3923	0.4528	0.0923
0.131	18.0	128250	0.3757	0.4736	0.0961
0.1245	19.0	135375	0.4248	0.4611	0.0941
0.1181	20.0	142500	0.4436	0.4629	0.0939
0.1119	21.0	149625	0.4508	0.4631	0.0931

Framework versions

• Transformers 4.46.2
• Pytorch 2.1.0+cu118
• Datasets 3.1.0
• Tokenizers 0.20.3