lnxdx/22_2000_1e-5_hp-masoud

wav2vec2-large-xlsr-persian-asr-shemo_partial

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: 0.8069
• Wer: 0.3490

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
2.3226	0.62	100	1.4548	0.4851
1.7166	1.25	200	1.2460	0.4279
1.4987	1.88	300	1.0671	0.4194
1.3771	2.5	400	0.9784	0.4054
1.3217	3.12	500	0.9450	0.3905
1.3272	3.75	600	0.8851	0.3841
1.3025	4.38	700	0.8748	0.3779
1.2719	5.0	800	0.8674	0.3724
1.2563	5.62	900	0.8467	0.3692
1.2451	6.25	1000	0.8440	0.3645
1.2585	6.88	1100	0.8292	0.3610
1.2633	7.5	1200	0.8137	0.3601
1.1923	8.12	1300	0.8263	0.3575
1.2349	8.75	1400	0.8184	0.3551
1.2511	9.38	1500	0.8078	0.3516
1.1779	10.0	1600	0.8102	0.3505
1.2161	10.62	1700	0.8123	0.3499
1.1967	11.25	1800	0.8086	0.3502
1.2454	11.88	1900	0.8066	0.3490
1.1928	12.5	2000	0.8069	0.3490

Framework versions

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