fpulaar_v3

This model is a fine-tuned version of openai/whisper-small on the abdouaziz/pulaar_speech_solited dataset. It achieves the following results on the evaluation set:

• Loss: 1.2722
• Wer: 0.3846

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: 4e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• 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: 50
• training_steps: 48000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3089	11.3636	500	0.9118	0.6109
0.0144	22.7273	1000	0.9989	0.4202
0.0084	34.0909	1500	1.0294	0.4198
0.0084	45.4545	2000	1.0798	0.4304
0.0063	56.8182	2500	1.0848	0.4178
0.0066	68.1818	3000	1.1000	0.4083
0.0063	79.5455	3500	1.1406	0.4116
0.0041	90.9091	4000	1.1605	0.4092
0.0003	102.2727	4500	1.1999	0.4010
0.0002	113.6364	5000	1.2183	0.3965
0.0002	125.0	5500	1.2352	0.3944
0.0002	136.3636	6000	1.2441	0.3903
0.0001	147.7273	6500	1.2607	0.3912
0.0001	159.0909	7000	1.2722	0.3846
0.0162	170.4545	7500	1.2051	0.4497
0.0107	181.8182	8000	1.1968	0.4296
0.002	193.1818	8500	1.1994	0.4173

Framework versions

• Transformers 4.46.0
• Pytorch 2.7.0+cu126
• Datasets 3.0.0
• Tokenizers 0.20.3