UnarineLeo/whisper-small-lwazi-multilingual

Whisper whisper-small lwazi multilingual

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

• Loss: 0.3685
• Wer Ortho: 36.1025
• Wer: 36.1358

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: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 150
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
1.0123	0.4237	250	0.9345	82.7840	82.9039
0.7055	0.8475	500	0.6634	69.5651	69.6134
0.4688	1.2712	750	0.5473	60.8014	60.8181
0.4106	1.6949	1000	0.4685	54.4493	54.4843
0.2469	2.1186	1250	0.4259	48.8100	48.9050
0.2409	2.5424	1500	0.3983	45.8038	45.8638
0.2243	2.9661	1750	0.3706	37.1201	37.1684
0.1224	3.3898	2000	0.3685	36.1025	36.1358

Framework versions

• Transformers 4.52.0
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.4