Makkoen/whisper-medium.en-cit-do015-wd0-lr1e-06-SF-200

./200

This model is a fine-tuned version of openai/whisper-medium.en on the 200 SF 200 dataset. It achieves the following results on the evaluation set:

• Loss: 0.8389
• Wer Ortho: 35.1676
• Wer: 23.8967

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-06
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 2
• 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: 200
• training_steps: 800
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
1.5629	8.0	100	1.1016	41.3994	29.9964
0.7012	16.0	200	0.8286	36.0787	25.1525
0.4369	24.0	300	0.8091	36.1152	25.3678
0.3073	32.0	400	0.8257	34.7303	23.7890
0.2298	40.0	500	0.8354	34.8397	22.8920
0.1938	48.0	600	0.8306	35.4592	23.4661
0.1658	56.0	700	0.8359	35.5321	23.8967
0.1534	64.0	800	0.8389	35.1676	23.8967

Framework versions

• Transformers 4.44.0
• Pytorch 1.13.1+cu117
• Datasets 2.20.0
• Tokenizers 0.19.1