CHUNGNAM_FM_AddressesM_model

This model is a fine-tuned version of openai/whisper-medium on the Marcusxx/CHUNGNAM_Addresses_NO_NUM dataset. It achieves the following results on the evaluation set:

• Loss: 0.2603
• Cer: 6.2263

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: 100
• training_steps: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.1938	0.6906	1000	0.2020	5.9531
0.1554	1.3812	2000	0.1852	5.9452
0.1048	2.0718	3000	0.1793	5.8234
0.1126	2.7624	4000	0.1794	7.6374
0.0695	3.4530	5000	0.1922	6.2990
0.0382	4.1436	6000	0.1999	6.2872
0.0385	4.8343	7000	0.2019	7.5529
0.0203	5.5249	8000	0.2141	7.6944
0.0142	6.2155	9000	0.2211	6.0239
0.0129	6.9061	10000	0.2190	8.6417
0.0109	7.5967	11000	0.2262	8.0187
0.0062	8.2873	12000	0.2286	10.8626
0.0074	8.9779	13000	0.2323	7.1874
0.005	9.6685	14000	0.2370	7.7829
0.0046	10.3591	15000	0.2415	6.2243
0.0021	11.0497	16000	0.2459	6.0946
0.002	11.7403	17000	0.2474	6.1713
0.0009	12.4309	18000	0.2572	6.0887
0.0001	13.1215	19000	0.2582	6.2715
0.0002	13.8122	20000	0.2603	6.2263

Framework versions

• Transformers 4.41.2
• Pytorch 2.2.2+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1