zionia/whisper-small-isixhosa

Whisper Small IsiXhosa

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

• Loss: 0.9605
• Wer: 57.9545

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: 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: 100
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.5789	1.0	17	3.0669	116.6667
2.5749	2.0	34	2.1387	94.1919
1.6536	3.0	51	1.6168	83.5859
1.2888	4.0	68	1.2976	73.8636
0.747	5.0	85	1.0975	67.6768
0.4241	6.0	102	0.9918	63.1313
0.2659	7.0	119	0.9502	61.2374
0.1003	8.0	136	0.9372	58.7121
0.0531	9.0	153	0.9341	58.0808
0.0245	10.0	170	0.9448	58.3333
0.016	11.0	187	0.9444	58.4596
0.009	12.0	204	0.9559	58.4596
0.0069	13.0	221	0.9572	57.3232
0.0063	14.0	238	0.9594	58.0808
0.0058	15.0	255	0.9605	57.9545

Framework versions

• Transformers 4.57.0
• Pytorch 2.8.0+cu128
• Datasets 4.2.0
• Tokenizers 0.22.1