CheeLi03/whisper-tiny-tr-puct-combine-8k

Whisper Tiny Turkish Combine 8k - Chee Li

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

• Loss: 0.6564
• Wer: 173.7311

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: 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: 500
• training_steps: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0132	25.6410	1000	1.3487	149.9555
0.0215	51.2821	2000	0.5751	126.8923
0.0026	76.9231	3000	0.6045	176.7587
0.0011	102.5641	4000	0.6247	148.6198
0.0006	128.2051	5000	0.6369	174.5325
0.0004	153.8462	6000	0.6470	171.5049
0.0003	179.4872	7000	0.6530	184.2386
0.0003	205.1282	8000	0.6564	173.7311

Framework versions

• Transformers 4.46.2
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.20.1