lorenzoncina/whisper-medium-zh

Whisper Small Chinese

This model is a fine-tuned version of openai/whisper-medium on the mozilla-foundation/common_voice_11_0 zh-CN dataset. It achieves the following results on the evaluation set:

• Loss: 0.3226
• Cer: 10.9782

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: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.3998	0.1	1000	0.2898	19.1261
0.2414	1.07	2000	0.2826	12.7761
0.1197	2.04	3000	0.2952	12.4320
0.2034	3.0	4000	0.2962	13.1970
0.0344	3.1	5000	0.3039	11.5122
0.0226	4.07	6000	0.3083	11.3549
0.0097	5.04	7000	0.3187	11.4440
0.0121	6.01	8000	0.3173	11.2258
0.0015	6.11	9000	0.3219	11.1410
0.0019	7.07	10000	0.3226	10.9782

Framework versions

• Transformers 4.28.0.dev0
• Pytorch 2.0.0+cu117
• Datasets 2.11.1.dev0
• Tokenizers 0.13.2