Makkoen/whisper-medium.en-cit-do015-wd0-lr5e-06-SF-1000

./1000

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

• Loss: 0.6318
• Wer Ortho: 32.5802
• Wer: 21.4926

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-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
1.1316	1.7699	100	0.6968	29.0816	18.8733
0.4669	3.5398	200	0.5156	27.4417	17.5816
0.2075	5.3097	300	0.5303	27.6968	16.7205
0.1163	7.0796	400	0.5391	28.6443	17.8687
0.0712	8.8496	500	0.5811	28.9723	17.5816
0.0518	10.6195	600	0.6104	31.8513	21.2415
0.0388	12.3894	700	0.6245	32.4344	21.4926
0.034	14.1593	800	0.6318	32.5802	21.4926

Framework versions

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