burkimbia/BIA-WHISPER-LARGE-SACHI_V1

SaCHi_ASR

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the sawadogosalif/MooreFRCollectionsAudios dataset. It achieves the following results on the evaluation set:

• Loss: 0.1507
• Wer: 10.9659

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: 5e-05
• train_batch_size: 16
• eval_batch_size: 8
• 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: linear
• lr_scheduler_warmup_steps: 20
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3863	0.4657	400	0.3211	23.1126
0.2748	0.9313	800	0.2645	19.4439
0.2308	1.3970	1200	0.2361	20.9174
0.212	1.8626	1600	0.2167	17.2891
0.1232	2.3283	2000	0.2039	13.1813
0.137	2.7939	2400	0.1852	17.8341
0.0894	3.2596	2800	0.1734	11.8642
0.0883	3.7253	3200	0.1615	10.6732
0.0525	4.1909	3600	0.1556	11.1930
0.0602	4.6566	4000	0.1507	10.9659

Framework versions

• Transformers 4.50.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1