ehzawad/whisper-small-bn

Whisper small by ehzawad

This model is a fine-tuned version of openai/whisper-small on the Common Voice 13.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.1104
• Wer: 31.3274

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: 2
• 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
0.2424	0.27	500	0.2407	63.1783
0.1559	0.53	1000	0.1633	48.0380
0.1255	0.8	1500	0.1394	42.6625
0.0899	1.07	2000	0.1231	38.6982
0.0872	1.34	2500	0.1172	37.3415
0.0755	1.6	3000	0.1091	35.4971
0.0786	1.87	3500	0.1042	34.6567
0.0499	2.14	4000	0.1047	33.2752
0.0468	2.4	4500	0.1027	32.7874
0.0436	2.67	5000	0.1019	32.2877
0.0379	2.94	5500	0.1000	31.7168
0.025	3.2	6000	0.1062	31.6455
0.0282	3.47	6500	0.1050	31.4699
0.0249	3.74	7000	0.1060	31.3737
0.0231	4.01	7500	0.1049	31.1969
0.0183	4.27	8000	0.1104	31.3274

Framework versions

• Transformers 4.30.0.dev0
• Pytorch 2.0.1+cu117
• Datasets 2.12.0
• Tokenizers 0.13.3