lyimo/whisper-medium-sw-v13

Whisper Small Swahili - Badili

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 13.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4329
• Wer: 98.4012

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: 12000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3563	0.35	1000	0.4938	100.5715
0.2853	0.69	2000	0.4143	100.7007
0.1612	1.04	3000	0.3910	100.9748
0.1399	1.38	4000	0.3762	98.4989
0.1657	1.73	5000	0.3700	90.3357
0.0818	2.08	6000	0.3775	98.0493
0.0749	2.42	7000	0.3768	97.9936
0.0637	2.77	8000	0.3822	92.9440
0.0355	3.11	9000	0.4036	93.8979
0.0299	3.46	10000	0.4141	97.9695
0.0277	3.8	11000	0.4175	98.2961
0.0147	4.15	12000	0.4329	98.4012

Framework versions

• Transformers 4.29.0.dev0
• Pytorch 2.0.0+cu117
• Datasets 2.12.0
• Tokenizers 0.13.3