arun100/whisper-base-cn-1

Whisper Base Chinese-Mandarin

This model is a fine-tuned version of openai/whisper-base on the mozilla-foundation/common_voice_16_0 zh-CN dataset. It achieves the following results on the evaluation set:

• Loss: 0.5263
• Wer: 89.1344

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-07
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• 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	Wer
0.9769	1.02	500	0.6812	94.6411
0.8022	3.0	1000	0.6262	92.5794
0.9109	4.02	1500	0.6009	92.6229
0.7132	6.0	2000	0.5845	92.3967
0.8416	7.02	2500	0.5725	91.7616
0.6527	9.0	3000	0.5636	91.4659
0.812	10.02	3500	0.5561	90.8917
0.6584	12.0	4000	0.5504	90.7960
0.7825	13.02	4500	0.5455	90.4045
0.6174	15.0	5000	0.5416	90.0565
0.7925	16.02	5500	0.5381	90.0217
0.5983	18.0	6000	0.5355	89.7695
0.741	19.02	6500	0.5331	89.7086
0.5831	21.0	7000	0.5312	89.4998
0.7414	22.02	7500	0.5296	89.5259
0.5902	24.0	8000	0.5284	89.3084
0.7242	25.02	8500	0.5275	89.4041
0.5815	27.0	9000	0.5268	89.1518
0.717	28.02	9500	0.5265	89.2562
0.5887	30.0	10000	0.5263	89.1344

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0