whisper-tiny-jiaozhu

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

• Loss: 0.0772
• Wer: 390.0

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: 50
• training_steps: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.9419	1.0	5	3.7933	120.0
3.5809	2.0	10	3.3116	115.0000
2.9392	3.0	15	2.5977	115.0000
2.4416	4.0	20	2.1190	100.0
1.9172	5.0	25	1.5509	100.0
1.3762	6.0	30	1.0480	185.0
0.9056	7.0	35	0.7003	6325.0
0.5587	8.0	40	0.3793	3885.0
0.3094	9.0	45	0.2168	3005.0
0.1747	10.0	50	0.1447	2455.0
0.101	11.0	55	0.1192	160.0
0.0621	12.0	60	0.0981	310.0
0.0389	13.0	65	0.0968	300.0
0.0252	14.0	70	0.0831	365.0
0.0163	15.0	75	0.0837	490.0000
0.0118	16.0	80	0.0790	450.0
0.009	17.0	85	0.0770	395.0
0.0076	18.0	90	0.0772	390.0
0.0069	19.0	95	0.0774	390.0
0.0066	20.0	100	0.0772	390.0

Framework versions

• Transformers 4.43.0.dev0
• Pytorch 2.3.1+cpu
• Datasets 2.20.0
• Tokenizers 0.19.1