whisper-small-transfer

This model is a fine-tuned version of Leonel-Maia/fongbe-whisper-small on the Leonel-Maia/ewe_dataset_splitted dataset. It achieves the following results on the evaluation set:

• Loss: 0.2392
• Wer: 0.2136

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• optimizer: Use 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: 500
• num_epochs: 60.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3978	0.6033	500	0.3847	0.3484
0.249	1.2075	1000	0.2890	0.2585
0.2481	1.8109	1500	0.2585	0.2387
0.1996	2.4151	2000	0.2470	0.2233
0.1669	3.0193	2500	0.2410	0.2157
0.1535	3.6226	3000	0.2392	0.2136
0.1272	4.2268	3500	0.2459	0.2150
0.1226	4.8302	4000	0.2428	0.2119
0.0939	5.4344	4500	0.2541	0.2142
0.0665	6.0386	5000	0.2640	0.2156
0.0717	6.6419	5500	0.2720	0.2185

Framework versions

• Transformers 4.50.3
• Pytorch 2.7.0+cu126
• Datasets 3.5.0
• Tokenizers 0.21.1