ihanif/whisper-small-pashto-dropout

Whisper Small Pashto

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

• Loss: 1.2273
• Wer: 56.6510

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: 3e-07
• train_batch_size: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 800
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.1183	3.7	100	1.3170	76.9522
0.8565	7.41	200	0.9367	61.9930
0.2246	11.11	300	0.9642	58.8302
0.054	14.81	400	1.0876	57.9903
0.0159	18.52	500	1.1798	57.8768
0.0045	22.22	600	1.2309	56.6510
0.0026	100.0	700	1.2581	56.8478
0.0023	114.29	800	1.2710	56.7570

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 1.13.1+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2