phatjmo/whisper-medium-hat

Whisper Medium Ht - TranslateLive

This model is a fine-tuned version of openai/whisper-medium on the Carnegie Mellon University - Haitian Creole dataset. It achieves the following results on the evaluation set:

• Loss: 0.7836
• Wer: 19.0404

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: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0571	3.4483	1000	0.5976	21.8190
0.0037	6.8966	2000	0.7251	19.5136
0.0003	10.3448	3000	0.7623	19.4807
0.0005	13.7931	4000	0.7836	19.0404

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.0+cu121
• Datasets 3.0.0
• Tokenizers 0.19.1