Bateesa/whisper-small-multilingual-ug_v1

Whisper-Small-Multilingual-Uganda

This model is a fine-tuned version of openai/whisper-small on the Bateesa/popolivoice, Bateesa/buaiir_voice_jap dataset. It achieves the following results on the evaluation set:

• Loss: 0.7019
• Wer: 20.9620

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: 0.0001
• train_batch_size: 64
• eval_batch_size: 32
• 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
• training_steps: 1500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0917	2.2989	200	0.4674	24.6231
0.4938	4.5977	400	0.5308	21.3927
0.2699	6.8966	600	0.5277	22.6849
0.1000	9.1954	800	0.6029	21.8234
0.0515	11.4943	1000	0.6263	23.9770
0.0196	13.7931	1200	0.6510	20.6748
0.0034	16.0920	1400	0.6879	20.8902
0.0035	17.2414	1500	0.7019	20.9620

Framework versions

• Transformers 5.8.0
• Pytorch 2.11.0+cu130
• Datasets 2.21.0
• Tokenizers 0.22.2