rjac/whisper-medium-BTCv4

Whisper-medium-BTC

This model is a fine-tuned version of openai/whisper-medium.en on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.3882
• Wer: 6.6247

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: 4e-07
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 300
• training_steps: 350
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.8163	1.01	25	0.8356	10.0635
0.7916	2.02	50	0.8102	9.7351
0.7563	3.03	75	0.7621	9.5088
0.7154	4.05	100	0.7107	9.2337
0.6548	5.06	125	0.6589	9.3801
0.6017	7.01	150	0.6062	9.0074
0.5333	8.02	175	0.5347	8.6214
0.4493	9.03	200	0.4738	8.2842
0.4016	10.04	225	0.4333	7.1172
0.3738	11.05	250	0.4057	6.7001
0.3544	13.01	275	0.3882	6.6247
0.3294	14.02	300	0.3764	6.6957
0.313	15.03	325	0.3692	6.6602
0.3023	16.04	350	0.3668	6.6468

Framework versions

• Transformers 4.27.0.dev0
• Pytorch 1.13.1+cu117
• Datasets 2.9.1.dev0
• Tokenizers 0.13.2