sudoping01/bambara-whisper-large-v5-1

bambara-whisper-large-v5

This model is a fine-tuned version of openai/whisper-large-v2 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.0897

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• 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: cosine_with_restarts
• lr_scheduler_warmup_steps: 100
• num_epochs: 6
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
2.5664	0.2674	250	2.3071
1.9725	0.5348	500	2.0162
1.579	0.8021	750	1.4457
1.4458	1.0695	1000	1.3399
1.3508	1.3369	1250	1.2798
1.3225	1.6043	1500	1.2372
1.2974	1.8717	1750	1.2088
1.2617	2.1390	2000	1.1850
1.2305	2.4064	2250	1.1643
1.2392	2.6738	2500	1.1466
1.2395	2.9412	2750	1.1342
1.1663	3.2086	3000	1.1249
1.1967	3.4759	3250	1.1139
1.1624	3.7433	3500	1.1087
1.1527	4.0107	3750	1.1031
1.1512	4.2781	4000	1.0985
1.1589	4.5455	4250	1.0945
1.1547	4.8128	4500	1.0926
1.1334	5.0802	4750	1.0919
1.1688	5.3476	5000	1.0903
1.1324	5.6150	5250	1.0898
1.1735	5.8824	5500	1.0897

Framework versions

• PEFT 0.14.1.dev0
• Transformers 4.49.0.dev0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0