wav2vec2-large-xls-r-300m-upper-sorbian-pl-frozen-2-colab

This model is a fine-tuned version of on the common_voice_16_1 dataset. It achieves the following results on the evaluation set:

• Loss: 0.6971
• Wer: 0.3974
• Cer: 0.0976

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: 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: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.6605	3.23	100	0.6446	0.7160	0.1772
0.499	6.45	200	0.5900	0.6256	0.1482
0.3831	9.68	300	0.5681	0.5825	0.1385
0.2866	12.9	400	0.5691	0.5239	0.1235
0.2269	16.13	500	0.6304	0.5061	0.1202
0.1725	19.35	600	0.6649	0.4864	0.1137
0.1365	22.58	700	0.6574	0.4597	0.1107
0.1081	25.81	800	0.6525	0.4452	0.1057
0.0873	29.03	900	0.6799	0.4236	0.1031
0.0782	32.26	1000	0.7263	0.4468	0.1093
0.0715	35.48	1100	0.7404	0.4173	0.1002
0.061	38.71	1200	0.6985	0.4208	0.1009
0.0551	41.94	1300	0.7094	0.4082	0.0988
0.049	45.16	1400	0.7069	0.4096	0.1011
0.0492	48.39	1500	0.6971	0.4117	0.1008
0.0441	51.61	1600	0.6906	0.4025	0.0972
0.0407	54.84	1700	0.7154	0.4035	0.0985
0.0389	58.06	1800	0.6971	0.3974	0.0976

Framework versions

• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.19.0
• Tokenizers 0.15.2