Kathernie/whisper-small-all

Whisper Small Chinese Filtered

This model is a fine-tuned version of openai/whisper-small on the Learn Chinese dataset. It achieves the following results on the evaluation set:

• Loss: 0.2198
• Cer: 9.3505

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 250
• training_steps: 2000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.8566	0.22	100	0.6834	16.0219
0.4275	0.45	200	0.3398	17.3939
0.2534	0.67	300	0.2417	15.4507
0.2654	0.9	400	0.2306	19.4430
0.1627	1.12	500	0.2253	12.5478
0.1587	1.34	600	0.2183	9.8039
0.1807	1.57	700	0.2107	11.3349
0.1394	1.79	800	0.2109	10.8638
0.118	2.02	900	0.2084	9.8510
0.0886	2.24	1000	0.2134	9.9865
0.0663	2.46	1100	0.2157	9.6862
0.0997	2.69	1200	0.2125	9.4153
0.0707	2.91	1300	0.2129	9.2563
0.0476	3.14	1400	0.2156	9.0443
0.0435	3.36	1500	0.2174	9.0031
0.0543	3.58	1600	0.2168	8.9855
0.044	3.81	1700	0.2175	9.5154
0.0446	4.03	1800	0.2186	9.2387
0.0364	4.26	1900	0.2199	9.1739
0.037	4.48	2000	0.2198	9.3505

Framework versions

• Transformers 4.37.2
• Pytorch 2.1.1
• Datasets 2.15.0
• Tokenizers 0.15.1