lnxdx/19_2000_1e-5_hp-base

wav2vec2-large-xlsr-persian-asr-shemo_lnxdx

This model is a fine-tuned version of masoudmzb/wav2vec2-xlsr-multilingual-53-fa on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.7064
• Wer: 0.3344

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: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.9455	0.62	100	1.4247	0.4831
1.4824	1.25	200	1.1107	0.4331
1.2516	1.88	300	0.9141	0.4136
1.0859	2.5	400	0.8360	0.3975
1.0357	3.12	500	0.8097	0.3814
1.0472	3.75	600	0.7550	0.3753
0.9963	4.38	700	0.7533	0.3636
0.9767	5.0	800	0.7424	0.3589
0.9667	5.62	900	0.7360	0.3516
0.9385	6.25	1000	0.7355	0.3487
0.9805	6.88	1100	0.7237	0.3464
0.976	7.5	1200	0.7078	0.3455
0.88	8.12	1300	0.7229	0.3438
0.9421	8.75	1400	0.7180	0.3432
0.9584	9.38	1500	0.7059	0.3364
0.88	10.0	1600	0.7106	0.3364
0.9113	10.62	1700	0.7125	0.3344
0.912	11.25	1800	0.7091	0.3353
0.9607	11.88	1900	0.7066	0.3344
0.8974	12.5	2000	0.7064	0.3344

Framework versions

• Transformers 4.35.2
• Pytorch 2.1.0+cu118
• Datasets 2.15.0
• Tokenizers 0.15.0