wav2vec2-xls-r-300m-lg-CV-Fleurs_filtered-100hrs-v13

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: 0.4759
• Wer: 0.4454
• Cer: 0.0854

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: 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
0.7277	1.0	7125	0.4018	0.4833	0.1016
0.3225	2.0	14250	0.3800	0.4945	0.1067
0.2726	3.0	21375	0.3745	0.4588	0.0919
0.2416	4.0	28500	0.3439	0.4419	0.0885
0.2188	5.0	35625	0.3353	0.4657	0.0906
0.2024	6.0	42750	0.3289	0.4563	0.0881
0.1888	7.0	49875	0.3272	0.4451	0.0863
0.1767	8.0	57000	0.3267	0.4226	0.0830
0.1668	9.0	64125	0.3354	0.4305	0.0837
0.1568	10.0	71250	0.3277	0.4297	0.0857
0.1483	11.0	78375	0.3310	0.4425	0.0857
0.1398	12.0	85500	0.3433	0.4299	0.0836
0.1323	13.0	92625	0.3448	0.4472	0.0870
0.125	14.0	99750	0.3585	0.4388	0.0849
0.1174	15.0	106875	0.3623	0.4250	0.0828
0.1121	16.0	114000	0.3813	0.4333	0.0843
0.1059	17.0	121125	0.3788	0.4251	0.0825
0.0996	18.0	128250	0.3882	0.4434	0.0863
0.0944	19.0	135375	0.4082	0.4444	0.0860
0.0889	20.0	142500	0.4227	0.4446	0.0848
0.0846	21.0	149625	0.4323	0.4422	0.0852
0.081	22.0	156750	0.4540	0.4506	0.0881
0.0767	23.0	163875	0.4759	0.4454	0.0854

Framework versions

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