arun100/whisper-small-fa-3

Whisper Small Persian Iranian

This model is a fine-tuned version of arun100/whisper-small-fa-2 on the google/fleurs fa_ir dataset. It achieves the following results on the evaluation set:

• Loss: 0.3192
• Wer: 26.4129

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-06
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0865	43.0	500	0.3192	26.4129
0.008	86.0	1000	0.3816	27.0149
0.0033	130.0	1500	0.4108	27.2289
0.0019	173.0	2000	0.4313	27.4030
0.0013	217.0	2500	0.4479	27.5323
0.001	260.0	3000	0.4612	27.5423
0.0008	304.0	3500	0.4719	27.7861
0.0006	347.0	4000	0.4802	27.9900
0.0006	391.0	4500	0.4859	27.9502
0.0005	434.0	5000	0.4882	27.9154

Framework versions

• Transformers 4.38.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0