whisper-tiny-finetuning

This model is a fine-tuned version of openai/whisper-tiny on the PolyAI/minds14 dataset. It achieves the following results on the evaluation set:

• Loss: 0.6195
• Wer: 0.3374

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: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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
12.7444	1.0	29	1.8826	0.4948
4.5635	2.0	58	0.5162	0.4090
1.6167	3.0	87	0.4736	0.3640
1.0818	4.0	116	0.4711	0.3738
0.7325	5.0	145	0.4878	0.3387
0.5243	6.0	174	0.4995	0.3337
0.1949	7.0	203	0.5141	0.3362
0.1124	8.0	232	0.5422	0.3337
0.0617	9.0	261	0.5497	0.3350
0.0507	10.0	290	0.5687	0.3374
0.0314	11.0	319	0.5725	0.3313
0.0136	12.0	348	0.5918	0.3399
0.0085	13.0	377	0.6056	0.3418
0.0062	14.0	406	0.6109	0.3381
0.0049	15.0	435	0.6145	0.3387
0.0053	16.0	464	0.6186	0.3393
0.0042	17.0	493	0.6197	0.3350
0.0041	17.2478	500	0.6195	0.3374

Framework versions

• Transformers 5.0.1.dev0
• Pytorch 2.9.0+cu126
• Datasets 2.18.0
• Tokenizers 0.22.2