miosipof/whisper-medium-ft-balbus-sep28k-v1.4

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

• Loss: 0.1096
• Accuracy: 0.8007

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: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.5
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.1827	0.0313	50	0.1708	0.5653
0.1681	0.0627	100	0.1632	0.6341
0.1603	0.0940	150	0.1498	0.6924
0.1443	0.1254	200	0.1371	0.7404
0.139	0.1567	250	0.1258	0.7640
0.1194	0.1880	300	0.1158	0.7880
0.105	0.2194	350	0.1173	0.7914
0.1114	0.2507	400	0.1143	0.7965
0.1214	0.2820	450	0.1109	0.8001
0.1178	0.3134	500	0.1096	0.8007

Framework versions

• Transformers 4.45.2
• Pytorch 2.2.0
• Datasets 3.2.0
• Tokenizers 0.20.3