whisper-medium-ach

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: 0.3501
• Wer: 62.2984

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
1.081	0.05	200	0.4844	138.8609
0.592	1.0248	400	0.3859	154.4355
0.5445	1.0748	600	0.3434	146.1694
0.3446	2.0495	800	0.3272	163.9113
0.2614	3.0242	1000	0.3098	86.2903
0.2542	3.0743	1200	0.3414	91.4315
0.1972	4.049	1400	0.3289	89.3145
0.1172	5.0237	1600	0.3224	100.1008
0.1226	5.0738	1800	0.3377	72.3286
0.0721	6.0485	2000	0.3277	105.8972
0.0504	7.0232	2200	0.3483	80.1411
0.0503	7.0732	2400	0.3514	95.0101
0.0375	8.048	2600	0.3378	64.5665
0.0348	9.0228	2800	0.3492	122.5806
0.0338	9.0727	3000	0.3502	88.6089
0.0273	10.0475	3200	0.3554	88.2560
0.0194	11.0222	3400	0.3501	62.2984
0.0165	11.0723	3600	0.3478	73.3871
0.0117	12.047	3800	0.3618	74.2440
0.0125	13.0218	4000	0.3587	97.6815

Framework versions

• Transformers 4.41.0.dev0
• Pytorch 2.2.0
• Datasets 2.19.1.dev0
• Tokenizers 0.19.1