yesj1234/xlsr_exp1_ko-en

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: 0.4228
• Cer: 0.1091
• Wer: 0.3025

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: 15

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.5566	0.94	2000	1.0226	0.2632	0.6184
1.179	1.89	4000	0.7682	0.2001	0.4990
1.0432	2.83	6000	0.6633	0.1749	0.4516
0.9413	3.77	8000	0.6159	0.1624	0.4259
0.8765	4.72	10000	0.5792	0.1538	0.4061
0.8248	5.66	12000	0.5456	0.1446	0.3877
0.7714	6.6	14000	0.5316	0.1397	0.3710
0.7388	7.55	16000	0.5172	0.1356	0.3657
0.6912	8.49	18000	0.4892	0.1291	0.3508
0.6549	9.43	20000	0.4694	0.1241	0.3397
0.614	10.37	22000	0.4615	0.1205	0.3309
0.5901	11.32	24000	0.4489	0.1177	0.3215
0.555	12.26	26000	0.4419	0.1148	0.3163
0.5377	13.2	28000	0.4320	0.1122	0.3103
0.5253	14.15	30000	0.4251	0.1102	0.3052

Framework versions

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