SGzK/whisper-small-inbrowser-proctor

Whisper-Small-Proctor-lora

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

• Loss: 0.3882
• Wer: 17.3304

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: 5e-06
• 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: 50
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.7627	0.8929	25	0.7457	31.6183
0.3173	1.7857	50	0.4644	29.4680
0.2013	2.6786	75	0.3984	20.4205
0.1569	3.5714	100	0.3754	14.9570
0.1148	4.4643	125	0.3716	18.4932
0.0794	5.3571	150	0.3643	14.9888
0.0658	6.25	175	0.3637	20.1338
0.0272	7.1429	200	0.3527	18.5091
0.0169	8.0357	225	0.3457	22.0452
0.0085	8.9286	250	0.3644	24.3071
0.0052	9.8214	275	0.3716	20.5639
0.0042	10.7143	300	0.3802	18.7958
0.0019	11.6071	325	0.3738	16.6773
0.0017	12.5	350	0.3811	18.5409
0.0015	13.3929	375	0.3824	19.1940
0.0015	14.2857	400	0.3863	19.0347
0.0014	15.1786	425	0.3866	19.1144
0.0012	16.0714	450	0.3869	17.3304
0.0012	16.9643	475	0.3878	17.3304
0.001	17.8571	500	0.3882	17.3304

Framework versions

• PEFT 0.14.1.dev0
• Transformers 4.49.0.dev0
• Pytorch 2.2.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0