whisper-medium-konnakol-rests

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

• Loss: 0.0952
• Wer: 25.9953

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: 2
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6638	13.3333	50	0.0648	50.5855
0.0132	26.6667	100	0.0678	40.7494
0.0068	40.0	150	0.0794	30.6792
0.0033	53.3333	200	0.0909	28.3372
0.0008	66.6667	250	0.0821	25.9953
0.0021	80.0	300	0.0725	29.5082
0.0002	93.3333	350	0.0933	26.2295
0.0	106.6667	400	0.0932	26.2295
0.0	120.0	450	0.0937	26.2295
0.0	133.3333	500	0.0940	26.2295
0.0	146.6667	550	0.0942	26.2295
0.0	160.0	600	0.0944	25.9953
0.0	173.3333	650	0.0947	25.9953
0.0	186.6667	700	0.0946	25.9953
0.0	200.0	750	0.0947	25.9953
0.0	213.3333	800	0.0948	25.9953
0.0	226.6667	850	0.0950	25.9953
0.0	240.0	900	0.0950	25.9953
0.0	253.3333	950	0.0952	25.9953
0.0	266.6667	1000	0.0952	25.9953

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.19.1