Makkoen/whisper-medium.en-cit-do015-wd0-lr1e-06-SF-500

./500

This model is a fine-tuned version of openai/whisper-medium.en on the 500 SF 1000 dataset. It achieves the following results on the evaluation set:

• Loss: 0.6792
• Wer Ortho: 31.5962
• Wer: 21.0621

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
1.6525	3.1746	100	1.1367	40.1968	29.4223
0.8573	6.3492	200	0.7964	30.8309	20.3803
0.6108	9.5238	300	0.7344	28.6808	18.9092
0.4957	12.6984	400	0.7017	29.1181	18.7298
0.4164	15.8730	500	0.6860	29.2274	18.8016
0.3577	19.0476	600	0.6802	29.3367	18.6939
0.3168	22.2222	700	0.6787	31.2682	20.7750
0.3023	25.3968	800	0.6792	31.5962	21.0621

Framework versions

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