wav2vec2-xls-r-ewe-50-hours

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

• Loss: 0.8057
• Wer: 0.3383
• Cer: 0.0967

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: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
11.6182	3.3670	500	2.5123	1.0	0.6328
1.3765	6.7340	1000	0.3954	0.3612	0.1030
0.8498	10.1010	1500	0.3698	0.3378	0.0963
0.7262	13.4680	2000	0.3916	0.3283	0.0956
0.6137	16.8350	2500	0.3981	0.3335	0.0971
0.52	20.2020	3000	0.4405	0.3467	0.0999
0.4375	23.5690	3500	0.4442	0.3506	0.1038
0.3707	26.9360	4000	0.5235	0.3471	0.0997
0.2992	30.3030	4500	0.5823	0.3550	0.1009
0.2632	33.6700	5000	0.6557	0.3449	0.0984
0.2181	37.0370	5500	0.6931	0.3412	0.0971
0.1888	40.4040	6000	0.7232	0.3420	0.0980
0.1653	43.7710	6500	0.7812	0.3394	0.0963
0.1463	47.1380	7000	0.8057	0.3383	0.0967

Framework versions

• Transformers 4.46.1
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.20.3