whisper-small-multilingual-naija-10-20-2024

This model is a fine-tuned version of ccibeekeoc42/whisper-small-yoruba-07-17 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.7923

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: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 1
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
1.5348	0.0458	100	1.4792
1.1822	0.0916	200	1.1469
0.9827	0.1374	300	1.0446
0.9354	0.1832	400	0.9927
0.8592	0.2290	500	0.9554
0.7696	0.2749	600	0.9287
0.7795	0.3207	700	0.9012
0.712	0.3665	800	0.8869
0.6793	0.4123	900	0.8753
0.7472	0.4581	1000	0.8578
0.7303	0.5039	1100	0.8460
0.6875	0.5497	1200	0.8380
0.7384	0.5955	1300	0.8287
0.7037	0.6413	1400	0.8211
0.712	0.6871	1500	0.8195
0.679	0.7329	1600	0.8073
0.6814	0.7787	1700	0.8033
0.6344	0.8246	1800	0.7994
0.6103	0.8704	1900	0.7950
0.6994	0.9162	2000	0.7926
0.5639	0.9620	2100	0.7923

Framework versions

• Transformers 4.45.2
• Pytorch 2.0.1+cu118
• Datasets 3.0.1
• Tokenizers 0.20.1