Makkoen/whisper-large-v3-cit-do015-wd0-lr1e-06-NYC-1000

./949

This model is a fine-tuned version of openai/whisper-large-v3 on the 949 FULL dataset. It achieves the following results on the evaluation set:

• Loss: 0.5601
• Wer Ortho: 29.5461
• Wer: 21.9669

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: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• 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: 300
• training_steps: 500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
1.0667	1.8692	100	0.7607	37.1700	28.4674
0.7153	3.7383	200	0.6157	32.8982	24.5167
0.5672	5.6075	300	0.5747	30.5251	22.3872
0.4809	7.4766	400	0.5630	29.4275	21.7428
0.428	9.3458	500	0.5601	29.5461	21.9669

Framework versions

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