haseong8012/wav2vec2-large-xlsr-53_ko

wav2vec2-large-xlsr-53-fine-tune_korean_byAILAB

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

• Loss: 1.4406
• Wer: 0.8577

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: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
No log	0.99	38	54.3133	1.0
No log	2.0	77	33.5397	1.0
No log	2.99	115	19.6459	1.0
No log	4.0	154	11.1346	1.0
No log	4.99	192	5.8854	1.0
No log	6.0	231	4.8784	1.0
No log	6.99	269	4.8369	1.0
No log	8.0	308	4.8535	1.0
No log	8.99	346	4.8388	1.0
No log	10.0	385	4.8360	1.0
15.1801	10.99	423	4.7653	1.0
15.1801	12.0	462	4.7385	1.0
15.1801	12.99	500	4.6927	0.9989
15.1801	14.0	539	4.6673	0.9991
15.1801	14.99	577	4.6948	0.9991
15.1801	16.0	616	4.6713	0.9991
15.1801	16.99	654	4.6603	1.0
15.1801	18.0	693	4.6428	0.9995
15.1801	18.99	731	4.6520	0.9994
15.1801	20.0	770	4.6554	0.9967
4.3888	20.99	808	4.6054	0.9998
4.3888	22.0	847	4.5723	0.9976
4.3888	22.99	885	4.4586	0.9967
4.3888	24.0	924	4.2547	0.9934
4.3888	24.99	962	3.6554	0.9931
4.3888	26.0	1001	2.8387	1.0084
4.3888	26.99	1039	2.4191	1.0551
4.3888	28.0	1078	2.0997	1.0197
4.3888	28.99	1116	2.0103	1.0176
4.3888	30.0	1155	1.8189	0.9461
4.3888	30.99	1193	1.7623	0.9726
2.7217	32.0	1232	1.7383	0.9976
2.7217	32.99	1270	1.6522	0.9584
2.7217	34.0	1309	1.5558	0.9193
2.7217	34.99	1347	1.5811	0.9440
2.7217	36.0	1386	1.5208	0.9158
2.7217	36.99	1424	1.5088	0.9038
2.7217	38.0	1463	1.5039	0.9086
2.7217	38.99	1501	1.4853	0.8987
2.7217	40.0	1540	1.4799	0.8847
2.7217	40.99	1578	1.4259	0.8694
0.7635	42.0	1617	1.4878	0.8883
0.7635	42.99	1655	1.4394	0.8693
0.7635	44.0	1694	1.4623	0.8743
0.7635	44.99	1732	1.4495	0.8710
0.7635	46.0	1771	1.4463	0.8655
0.7635	46.99	1809	1.4553	0.8704
0.7635	48.0	1848	1.4500	0.8646
0.7635	48.99	1886	1.4387	0.8566
0.7635	49.35	1900	1.4406	0.8577

Framework versions

• Transformers 4.33.2
• Pytorch 1.12.1
• Datasets 2.14.5
• Tokenizers 0.13.3