README.md

metadata

base_model: openai/whisper-medium
datasets:
  - Marcusxx/CHUNGNAM_FM_Addresses
language:
  - ko
license: apache-2.0
tags:
  - hf-asr-leaderboard
  - generated_from_trainer
model-index:
  - name: CHUNGNAM_FM_AddressesM_model
    results: []

CHUNGNAM_FM_AddressesM_model

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

• Loss: 0.0038
• Cer: 0.0909

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
0.0184	0.6906	1000	0.0177	37.1099
0.0102	1.3812	2000	0.0118	12.6786
0.0019	2.0718	3000	0.0097	0.3414
0.0037	2.7624	4000	0.0072	0.4861
0.0015	3.4530	5000	0.0069	0.1818
0.0012	4.1436	6000	0.0043	0.1410
0.0018	4.8343	7000	0.0049	0.1967
0.0004	5.5249	8000	0.0050	0.1299
0.0055	6.2155	9000	0.0063	0.1800
0.0014	6.9061	10000	0.0049	0.1447
0.0012	7.5967	11000	0.0045	0.1021
0.0002	8.2873	12000	0.0044	0.1150
0.0002	8.9779	13000	0.0036	0.1076
0.0001	9.6685	14000	0.0041	0.1076
0.0	10.3591	15000	0.0036	0.1021
0.0001	11.0497	16000	0.0040	0.0965
0.0	11.7403	17000	0.0038	0.1039
0.0	12.4309	18000	0.0038	0.0928
0.0	13.1215	19000	0.0038	0.0909
0.0	13.8122	20000	0.0038	0.0909

Framework versions

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