okezieowen/whisper-small-naija-9-7-2024

Whisper Small Naija

This model is a fine-tuned version of openai/whisper-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5707
• Wer: 47.7271

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
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.4056	0.1054	250	1.4307	78.3916
0.9509	0.2108	500	1.0383	71.7728
0.7805	0.3162	750	0.8800	65.6676
0.6558	0.4216	1000	0.7990	62.0093
0.6439	0.5270	1250	0.7510	64.0119
0.5898	0.6324	1500	0.7163	58.3060
0.5943	0.7378	1750	0.6829	57.5576
0.5335	0.8432	2000	0.6615	56.5056
0.528	0.9486	2250	0.6344	54.6675
0.4149	1.0540	2500	0.6291	54.5847
0.3842	1.1594	2750	0.6208	53.1334
0.3883	1.2648	3000	0.6095	47.0400
0.362	1.3702	3250	0.6022	53.3288
0.3747	1.4755	3500	0.5925	49.1806
0.3457	1.5809	3750	0.5834	48.9277
0.3529	1.6863	4000	0.5780	49.6644
0.3579	1.7917	4250	0.5735	51.2159
0.3446	1.8971	4500	0.5695	52.3765
0.319	2.0025	4750	0.5670	50.8363
0.256	2.1079	5000	0.5707	47.7271

Framework versions

• Transformers 4.44.2
• Pytorch 2.0.1+cu118
• Datasets 2.21.0
• Tokenizers 0.19.1