yesj1234/xlsr_koen_exp3

ko-xlsr2

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: 0.4239
• Cer: 0.1113
• Wer: 0.3038

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: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.7721	0.94	2000	1.1368	0.2903	0.6589
1.3501	1.89	4000	0.8561	0.2240	0.5451
1.2133	2.83	6000	0.7505	0.2003	0.4974
1.0981	3.77	8000	0.6768	0.1842	0.4686
1.0375	4.72	10000	0.6413	0.1707	0.4404
0.9927	5.66	12000	0.6106	0.1634	0.4246
0.9439	6.6	14000	0.5999	0.1613	0.4159
0.9059	7.55	16000	0.5740	0.1535	0.3985
0.8772	8.49	18000	0.5569	0.1478	0.3954
0.8483	9.43	20000	0.5407	0.1427	0.3784
0.81	10.37	22000	0.5283	0.1415	0.3744
0.793	11.32	24000	0.5179	0.1366	0.3663
0.7577	12.26	26000	0.5059	0.1359	0.3595
0.7379	13.2	28000	0.4969	0.1333	0.3532
0.7328	14.15	30000	0.4908	0.1308	0.3475
0.7119	15.09	32000	0.4887	0.1286	0.3478
0.7572	16.03	34000	0.5170	0.1327	0.3577
0.8198	16.98	36000	0.5839	0.1432	0.3825
0.8008	17.92	38000	0.5447	0.1376	0.3661
0.759	18.86	40000	0.4998	0.1337	0.3534
0.6907	19.81	42000	0.4710	0.1288	0.3412
0.659	20.75	44000	0.4578	0.1242	0.3325
0.6345	21.69	46000	0.4531	0.1221	0.3257
0.6242	22.64	48000	0.4498	0.1209	0.3218
0.6163	23.58	50000	0.4552	0.1194	0.3188
0.6121	24.52	52000	0.4633	0.1154	0.3137
0.6054	25.47	54000	0.4623	0.1176	0.3171
0.591	26.41	56000	0.4413	0.1146	0.3116
0.5713	27.35	58000	0.4338	0.1135	0.3093
0.5703	28.3	60000	0.4280	0.1121	0.3061
0.5576	29.24	62000	0.4248	0.1119	0.3047

Framework versions

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