whisper-finetuned-local

This model is a fine-tuned version of openai/whisper-tiny on the None dataset. It achieves the following results on the evaluation set:

• Loss: 2.7654
• Wer Dho: 0.847
• Wer Mean: 0.847
• Cer Dho: 0.361
• Cer Mean: 0.361

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• 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: 2000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Dho	Wer Mean	Cer Dho	Cer Mean
0.3363	11.1143	200	2.1328	0.861	0.861	0.346	0.346
0.0063	22.2286	400	2.4564	0.823	0.823	0.392	0.392
0.0019	33.3429	600	2.5752	0.831	0.831	0.343	0.343
0.0012	44.4571	800	2.6316	0.823	0.823	0.345	0.345
0.0008	55.5714	1000	2.6798	0.829	0.829	0.353	0.353
0.0007	66.6857	1200	2.7150	0.831	0.831	0.347	0.347
0.0006	77.8	1400	2.7348	0.861	0.861	0.366	0.366
0.0005	88.9143	1600	2.7526	0.841	0.841	0.357	0.357
0.0004	100.0	1800	2.7644	0.839	0.839	0.351	0.351
0.0004	111.1143	2000	2.7654	0.847	0.847	0.361	0.361

Framework versions

• Transformers 4.57.3
• Pytorch 2.9.0+cu126
• Datasets 4.0.0
• Tokenizers 0.22.1