CasperMuz/whisper-base-sna

Whisper Base Shona - Uncleaned Data Extended

This model is a fine-tuned version of openai/whisper-base on the Google WAXAL Shona dataset. It achieves the following results on the evaluation set:

• Loss: 0.4706
• Wer: 41.4760

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: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use adamw_torch_fused with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• training_steps: 4500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6454	0.5269	450	0.6835	54.7544
0.5069	1.0539	900	0.5653	48.1483
0.4578	1.5808	1350	0.5204	45.6777
0.3822	2.1077	1800	0.4981	43.4642
0.4036	2.6347	2250	0.4861	43.4907
0.3300	3.1616	2700	0.4768	41.8365
0.3245	3.6885	3150	0.4729	40.2274
0.2943	4.2155	3600	0.4721	40.2327
0.2856	4.7424	4050	0.4709	40.6569
0.2750	5.2693	4500	0.4706	41.4760

Framework versions

• Transformers 5.14.1
• Pytorch 2.13.0+cu130
• Datasets 5.0.1
• Tokenizers 0.22.2