arun100/whisper-base-fa-1

Whisper Base Persian Iranian

This model is a fine-tuned version of openai/whisper-base on the mozilla-foundation/common_voice_16_0 fa dataset. It achieves the following results on the evaluation set:

• Loss: 0.7142
• Wer: 58.5965

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: 5e-07
• 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: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.1086	1.02	500	1.2735	85.9444
0.8782	3.0	1000	1.0477	76.5527
0.6726	4.02	1500	0.9506	71.8807
0.7501	6.0	2000	0.8943	69.3890
0.6079	7.02	2500	0.8550	67.1322
0.6592	9.0	3000	0.8239	66.2762
0.5703	10.02	3500	0.8007	63.9907
0.5767	12.0	4000	0.7815	63.2562
0.5098	13.02	4500	0.7671	62.1094
0.5373	15.01	5000	0.7555	61.5551
0.4592	16.02	5500	0.7460	61.1086
0.5032	18.01	6000	0.7376	60.5652
0.4262	19.02	6500	0.7329	60.0792
0.4726	21.01	7000	0.7257	59.6696
0.4043	22.02	7500	0.7237	59.3570
0.4758	24.01	8000	0.7187	59.1098
0.412	25.02	8500	0.7173	58.8518
0.5119	27.01	9000	0.7146	58.7276
0.4089	28.03	9500	0.7145	58.6347
0.5186	30.01	10000	0.7142	58.5965

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0