whisper-medium-ach-only

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

• Loss: 0.3942
• Wer: 21.1520

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
1.1662	0.05	200	0.7062	47.0255
0.6014	1.0248	400	0.4606	32.9556
0.5638	1.0748	600	0.4021	27.2899
0.3677	2.0495	800	0.3736	24.3626
0.2711	3.0242	1000	0.3648	23.5127
0.2862	3.0743	1200	0.3402	23.7016
0.2023	4.049	1400	0.3665	22.4740
0.1166	5.0237	1600	0.4023	23.6072
0.1089	5.0738	1800	0.3871	22.5685
0.0859	6.0485	2000	0.3837	25.6846
0.0557	7.0232	2200	0.3942	21.1520
0.0572	7.0732	2400	0.3805	22.0963
0.0469	8.048	2600	0.3995	23.6072
0.0308	9.0228	2800	0.4057	21.5297
0.0288	9.0727	3000	0.3999	21.1520
0.0222	10.0475	3200	0.4121	21.7186
0.0239	11.0222	3400	0.4162	21.9075
0.024	11.0723	3600	0.4154	21.9075
0.0219	12.047	3800	0.4186	21.3409
0.0133	13.0218	4000	0.4173	21.1520

Framework versions

• Transformers 4.41.0.dev0
• Pytorch 2.2.0
• Datasets 2.16.1
• Tokenizers 0.19.1