whisper-base-zh

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

• Loss: 0.3426
• Wer: 78.6221

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: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 4000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4992	0.2075	100	0.4841	93.0091
0.4325	0.4149	200	0.4223	82.7761
0.4028	0.6224	300	0.3979	81.6616
0.3866	0.8299	400	0.3846	79.8886
0.3322	1.0373	500	0.3731	80.3951
0.3108	1.2448	600	0.3672	79.2300
0.3139	1.4523	700	0.3601	79.1287
0.324	1.6598	800	0.3558	78.7741
0.2629	1.8672	900	0.3525	78.1155
0.2421	2.0747	1000	0.3521	78.5208
0.217	2.2822	1100	0.3495	78.3688
0.2071	2.4896	1200	0.3490	78.5714
0.2183	2.6971	1300	0.3452	78.6727
0.2158	2.9046	1400	0.3426	78.6221

Framework versions

• Transformers 4.46.2
• Pytorch 2.5.1+cu121
• Datasets 3.1.0
• Tokenizers 0.20.3