xls-r-300m-hbs-pl-unfrozen

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: 0.6301
• Wer: 0.3828
• Cer: 0.0893

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: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
3.2312	6.4516	100	3.2431	1.0	1.0
1.4572	12.9032	200	1.3541	0.9984	0.4356
0.236	19.3548	300	0.6024	0.5225	0.1264
0.1377	25.8065	400	0.5973	0.4714	0.1143
0.1784	32.2581	500	0.5847	0.4695	0.1125
0.0932	38.7097	600	0.6098	0.4241	0.1037
0.0667	45.1613	700	0.6210	0.4243	0.1022
0.0624	51.6129	800	0.5971	0.3988	0.0949
0.0543	58.0645	900	0.6406	0.4126	0.0989
0.0438	64.5161	1000	0.6430	0.3997	0.0940
0.0582	70.9677	1100	0.6553	0.4105	0.0974
0.0504	77.4194	1200	0.6454	0.3896	0.0907
0.0465	83.8710	1300	0.6320	0.3920	0.0920
0.0322	90.3226	1400	0.6312	0.3861	0.0904
0.0375	96.7742	1500	0.6301	0.3828	0.0893

Framework versions

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