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

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice_16_1 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7163
• Wer: 0.4302
• Cer: 0.1003

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
0.6172	3.23	100	0.6599	0.6999	0.1787
0.4414	6.45	200	0.6030	0.6251	0.1524
0.289	9.68	300	0.5899	0.5670	0.1336
0.1953	12.9	400	0.6095	0.5457	0.1308
0.1388	16.13	500	0.6628	0.5159	0.1224
0.1187	19.35	600	0.7075	0.4932	0.1180
0.0994	22.58	700	0.7131	0.4780	0.1143
0.0816	25.81	800	0.6959	0.4752	0.1101
0.0727	29.03	900	0.7201	0.4644	0.1104
0.0637	32.26	1000	0.7288	0.4630	0.1080
0.0592	35.48	1100	0.7219	0.4524	0.1056
0.0549	38.71	1200	0.7204	0.4480	0.1041
0.0473	41.94	1300	0.7238	0.4470	0.1048
0.0412	45.16	1400	0.7109	0.4278	0.1011
0.0423	48.39	1500	0.7252	0.4407	0.1045
0.0419	51.61	1600	0.7193	0.4393	0.1028
0.0365	54.84	1700	0.7231	0.4318	0.1010
0.0347	58.06	1800	0.7163	0.4302	0.1003

Framework versions

• Transformers 4.38.2
• Pytorch 2.2.1+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2