This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice_13_0 dataset. It achieves the following results on the evaluation set:

Loss: 1.4687
Wer: 0.5932

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: 300

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
20.8922	6.25	400	4.6827	0.9990
4.0513	12.5	800	2.3657	0.9204
1.5386	18.75	1200	1.2355	0.7392
0.7429	25.0	1600	1.1179	0.6636
0.3746	31.25	2000	1.0465	0.6314
0.2407	37.5	2400	1.1492	0.6596
0.1966	43.75	2800	1.1291	0.6344
0.1697	50.0	3200	1.1897	0.6395
0.1533	56.25	3600	1.2202	0.6193
0.129	62.5	4000	1.2106	0.6516
0.1097	68.75	4400	1.1662	0.6254
0.102	75.0	4800	1.2086	0.6133
0.0918	81.25	5200	1.2295	0.6485
0.0806	87.5	5600	1.2861	0.6123
0.0738	93.75	6000	1.2436	0.6093
0.0697	100.0	6400	1.3496	0.6626
0.0667	106.25	6800	1.2364	0.6133
0.0591	112.5	7200	1.2689	0.6062
0.054	118.75	7600	1.2886	0.6183
0.0523	125.0	8000	1.3328	0.6445
0.0542	131.25	8400	1.4019	0.6133
0.045	137.5	8800	1.3426	0.6042
0.0425	143.75	9200	1.3042	0.6032
0.0378	150.0	9600	1.3638	0.6224
0.0354	156.25	10000	1.3397	0.6294
0.0282	162.5	10400	1.3939	0.6173
0.0288	168.75	10800	1.3674	0.6475
0.0278	175.0	11200	1.3636	0.6324
0.0239	181.25	11600	1.4101	0.6405
0.0238	187.5	12000	1.4528	0.6163
0.0214	193.75	12400	1.4458	0.6093
0.0194	200.0	12800	1.3920	0.6304
0.0168	206.25	13200	1.4277	0.6193
0.0168	212.5	13600	1.3959	0.6203
0.0154	218.75	14000	1.4043	0.6133
0.0144	225.0	14400	1.4508	0.6193
0.0134	231.25	14800	1.4309	0.6224
0.0109	237.5	15200	1.4301	0.6123
0.0107	243.75	15600	1.4373	0.6002
0.0098	250.0	16000	1.4147	0.6113
0.0095	256.25	16400	1.4585	0.6193
0.009	262.5	16800	1.4424	0.6203
0.0079	268.75	17200	1.5019	0.6193
0.0066	275.0	17600	1.4835	0.5932
0.0059	281.25	18000	1.4749	0.5992
0.0057	287.5	18400	1.4897	0.6002
0.0053	293.75	18800	1.4667	0.5901
0.0048	300.0	19200	1.4687	0.5932

Framework versions

Transformers 4.35.0.dev0
Pytorch 1.12.1
Datasets 2.14.5
Tokenizers 0.14.0

kresnik/wav2vec2-large-xls-r-300m-korean

wav2vec2-large-xls-r-300m-korean

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

kresnik/
wav2vec2-large-xls-r-300m-korean