whisper_medium

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

• Loss: 0.8150
• Cer: 18.5268
• Wer: 29.2412

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 1000
• num_epochs: 16
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
1.2907	0.9999	4710	23.0132	0.7406	36.5517
0.7187	2.0	9421	20.8767	0.7060	33.3982
0.6	2.9997	14130	19.9648	0.6988	31.7264
0.5865	3.9999	18840	20.0014	0.7123	31.6218
0.5186	5.0	23551	19.2677	0.7172	30.4461
0.472	5.9997	28260	18.9852	0.7162	30.0906
0.4797	6.9999	32970	19.3379	0.7248	30.4956
0.4354	8.0	37681	19.2062	0.7400	30.2514
0.4001	8.9999	42391	19.0646	0.7527	29.9263
0.3767	9.9996	47100	18.7777	0.7587	29.5899
0.3792	10.9999	51810	18.6863	0.7727	29.4977
0.3552	12.0	56521	18.5969	0.7830	29.4470
0.3371	12.9997	61230	18.6109	0.7887	29.4437
0.3418	13.9999	65940	0.7929	18.6870	29.5922
0.3225	15.0	70651	0.8090	18.6225	29.4099
0.3084	15.9997	75360	0.8150	18.5268	29.2412

Framework versions

• Transformers 4.41.2
• Pytorch 2.1.2+cu118
• Datasets 2.19.0
• Tokenizers 0.19.1