yesj1234/xlsr_koen_exp2

ko-xlsr

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

• Loss: 0.4269
• Cer: 0.1119
• Wer: 0.3072

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: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 1000
• num_epochs: 15

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.54	0.94	2000	1.0057	0.2617	0.6135
1.1895	1.89	4000	0.7782	0.2040	0.5035
1.0582	2.83	6000	0.6767	0.1826	0.4655
0.9586	3.77	8000	0.6273	0.1690	0.4380
0.8831	4.72	10000	0.5884	0.1552	0.4071
0.8318	5.66	12000	0.5510	0.1469	0.3897
0.7725	6.6	14000	0.5327	0.1407	0.3726
0.7254	7.55	16000	0.5081	0.1416	0.3676
0.6802	8.49	18000	0.4846	0.1313	0.3502
0.6386	9.43	20000	0.4676	0.1241	0.3344
0.5949	10.37	22000	0.4510	0.1185	0.3250
0.5736	11.32	24000	0.4416	0.1161	0.3189
0.5451	12.26	26000	0.4338	0.1143	0.3144
0.5375	13.2	28000	0.4287	0.1126	0.3095
0.5335	14.15	30000	0.4273	0.1122	0.3079

Framework versions

• Transformers 4.34.1
• Pytorch 2.1.0+cu121
• Datasets 2.14.5
• Tokenizers 0.14.1