cankaracelebi/whisper-medium-tr-v2

Whisper Medium Tr - Can K V2

This model is a fine-tuned version of openai/whisper-medium on the Common Voice 17.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1892
• Wer: 67.0255

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: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 4
• 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: 500
• training_steps: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.146	0.34	1000	0.2164	105.0754
0.1562	0.69	2000	0.2115	43.5238
0.0803	1.03	3000	0.1979	42.8919
0.0668	1.38	4000	0.1944	37.3397
0.0693	1.72	5000	0.1869	36.3910
0.0305	2.07	6000	0.1898	49.8254
0.0272	2.41	7000	0.1908	60.8005
0.0274	2.76	8000	0.1892	67.0255

Framework versions

• Transformers 4.28.1
• Pytorch 2.4.0+cu121
• Datasets 2.20.0
• Tokenizers 0.13.3