arun100/whisper-small-derived-hi-2

Whisper Base Bengali

This model is a fine-tuned version of arun100/whisper-small-derived-hi-1 on the mozilla-foundation/common_voice_16_0 hi dataset. It achieves the following results on the evaluation set:

• Loss: 0.1458
• Wer: 9.2881

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: 5e-07
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 20
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1368	1.02	100	0.1665	9.6951
0.1196	2.05	200	0.1586	9.6011
0.107	3.07	300	0.1538	9.4008
0.1051	5.02	400	0.1504	9.3235
0.0988	6.05	500	0.1486	9.4467
0.0939	7.07	600	0.1474	9.4425
0.0901	9.02	700	0.1464	9.3006
0.0859	10.04	800	0.1459	9.4362
0.0859	11.07	900	0.1458	9.2881
0.0839	13.02	1000	0.1456	9.2901

Framework versions

• Transformers 4.38.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0