whisper-small-multilingual-naija-11-03-2024

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

• Loss: 0.8445

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: 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: 1.5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
2.926	0.0296	100	2.1605
2.1251	0.0592	200	1.6041
1.8711	0.0889	300	1.3987
1.5748	0.1185	400	1.2746
1.4264	0.1481	500	1.2025
1.4298	0.1777	600	1.1450
1.3167	0.2073	700	1.0946
1.2684	0.2370	800	1.0631
1.1892	0.2666	900	1.0364
1.16	0.2962	1000	1.0220
1.1475	0.3258	1100	1.0080
1.0742	0.3555	1200	0.9798
1.066	0.3851	1300	0.9837
1.0404	0.4147	1400	0.9693
1.055	0.4443	1500	0.9536
1.0622	0.4739	1600	0.9489
0.9947	0.5036	1700	0.9403
0.9904	0.5332	1800	0.9300
0.9607	0.5628	1900	0.9207
0.9464	0.5924	2000	0.9250
0.9086	0.6220	2100	0.9131
0.9332	0.6517	2200	0.8991
0.8954	0.6813	2300	0.8989
0.9175	0.7109	2400	0.8899
0.9203	0.7405	2500	0.8886
0.84	0.7701	2600	0.8874
0.8379	0.7998	2700	0.8796
0.8328	0.8294	2800	0.8765
0.8455	0.8590	2900	0.8671
0.8391	0.8886	3000	0.8622
0.7729	0.9182	3100	0.8699
0.803	0.9479	3200	0.8597
0.7808	0.9775	3300	0.8507
0.729	1.0071	3400	0.8573
0.7048	1.0367	3500	0.8581
0.7018	1.0664	3600	0.8551
0.7297	1.0960	3700	0.8545
0.6594	1.1256	3800	0.8504
0.6682	1.1552	3900	0.8566
0.6621	1.1848	4000	0.8548
0.6626	1.2145	4100	0.8510
0.6611	1.2441	4200	0.8493
0.6505	1.2737	4300	0.8464
0.6323	1.3033	4400	0.8483
0.6564	1.3329	4500	0.8421
0.6293	1.3626	4600	0.8454
0.5936	1.3922	4700	0.8447
0.6114	1.4218	4800	0.8458
0.6551	1.4514	4900	0.8454
0.5963	1.4810	5000	0.8445

Framework versions

• Transformers 4.46.1
• Pytorch 2.1.0+cu118
• Datasets 3.1.0
• Tokenizers 0.20.1