yesj1234/xlsr_mid1_zh-ko

zh-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: 1.8449
• Cer: 0.4954

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: 2
• total_train_batch_size: 32
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 150
• num_epochs: 15

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
6.0153	0.5	330	5.3438	0.9522
5.3776	1.0	660	5.1534	0.9409
5.2604	1.5	990	5.0832	0.9108
5.2393	2.01	1320	5.0655	0.9073
5.1721	2.51	1650	5.0464	0.9000
5.1619	3.01	1980	5.0244	0.9045
5.1308	3.51	2310	5.0216	0.9020
5.0971	4.01	2640	4.9341	0.9040
5.0137	4.51	2970	4.8795	0.9144
4.9341	5.02	3300	4.7250	0.9039
4.6832	5.52	3630	4.2140	0.8367
4.1627	6.02	3960	3.4010	0.7318
3.5448	6.52	4290	2.8830	0.6480
3.2576	7.02	4620	2.6253	0.6266
2.8561	7.52	4950	2.4300	0.5866
2.7894	8.02	5280	2.2998	0.5750
2.6018	8.53	5610	2.1878	0.5549
2.546	9.03	5940	2.1450	0.5351
2.3787	9.53	6270	2.1027	0.5340
2.335	10.03	6600	2.0304	0.5166
2.2138	10.53	6930	2.0100	0.5165
2.2381	11.03	7260	1.9651	0.5031
2.1108	11.53	7590	1.9666	0.5035
2.0916	12.04	7920	1.9136	0.4998
2.0229	12.54	8250	1.8988	0.5028
2.0056	13.04	8580	1.8769	0.4996
1.9245	13.54	8910	1.8716	0.4955
1.9378	14.04	9240	1.8561	0.4946
1.9003	14.54	9570	1.8485	0.4936

Framework versions

• Transformers 4.34.0
• Pytorch 2.1.0+cu121
• Datasets 2.14.5
• Tokenizers 0.14.1