whisper-small-bn-09092023

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

• Loss: 0.0498
• Wer: 7.7181

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: 32
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 256
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2500
• training_steps: 8000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4968	0.16	400	0.3922	44.0283
0.2059	0.32	800	0.1848	27.1190
0.1486	0.48	1200	0.1359	20.7105
0.1203	0.64	1600	0.1094	16.8171
0.1032	0.79	2000	0.0938	14.5648
0.0912	0.95	2400	0.0816	12.6067
0.0778	1.11	2800	0.0749	11.5556
0.0717	1.27	3200	0.0686	10.8017
0.0676	1.43	3600	0.0639	9.7786
0.0641	1.59	4000	0.0607	9.4620
0.0622	1.75	4400	0.0588	9.2326
0.0597	1.91	4800	0.0561	8.8085
0.0498	2.07	5200	0.0546	8.3926
0.0493	2.23	5600	0.0539	8.3443
0.0483	2.38	6000	0.0525	8.1459
0.0478	2.54	6400	0.0518	8.0740
0.0472	2.7	6800	0.0510	7.9834
0.0463	2.86	7200	0.0502	7.7784
0.0439	3.02	7600	0.0499	7.7532
0.0407	3.18	8000	0.0498	7.7181

Framework versions

• Transformers 4.32.1
• Pytorch 2.0.1+cu117
• Datasets 2.14.4
• Tokenizers 0.13.3