whisper-medium-lora

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

• Loss: 0.3493

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: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
1.636	0.2051	250	1.3245
1.0308	0.4102	500	0.8528
0.5665	0.6153	750	0.4152
0.419	0.8203	1000	0.3934
0.3843	1.0254	1250	0.3830
0.3811	1.2305	1500	0.3760
0.3602	1.4356	1750	0.3708
0.3609	1.6407	2000	0.3665
0.3541	1.8458	2250	0.3631
0.3682	2.0509	2500	0.3610
0.3527	2.2559	2750	0.3587
0.3468	2.4610	3000	0.3574
0.354	2.6661	3250	0.3559
0.3392	2.8712	3500	0.3546
0.3507	3.0763	3750	0.3533
0.3343	3.2814	4000	0.3526
0.3439	3.4865	4250	0.3520
0.3491	3.6916	4500	0.3512
0.3486	3.8966	4750	0.3505
0.3544	4.1017	5000	0.3502
0.3352	4.3068	5250	0.3498
0.3429	4.5119	5500	0.3492
0.3451	4.7170	5750	0.3494
0.324	4.9221	6000	0.3493

Framework versions

• PEFT 0.10.0
• Transformers 4.40.1
• Pytorch 2.0.1
• Datasets 2.19.0
• Tokenizers 0.19.1