pashto-asr-cws-v1

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

• Loss: 0.7501
• Wer: 49.3795
• Cer: 20.8766

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: 32
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 30
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.7899	0.8065	100	0.8629	75.8323	47.9045
0.5051	1.6129	200	0.6860	52.9888	23.1242
0.3555	2.4194	300	0.6612	51.6798	22.9404
0.257	3.2258	400	0.6565	49.2660	21.0588
0.2174	4.0323	500	0.6560	47.1777	19.8523
0.1626	4.8387	600	0.6787	48.7591	20.4305
0.1268	5.6452	700	0.7062	49.4401	21.0137
0.0871	6.4516	800	0.7279	49.2433	20.7062
0.0685	7.2581	900	0.7443	49.3795	21.2276
0.0662	8.0645	1000	0.7501	49.3795	20.8766

Framework versions

• Transformers 4.42.4
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1