File size: 4,374 Bytes
d8887ee 274f03f d8887ee |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 |
---
library_name: transformers
language:
- ko
license: mit
base_model: openai/whisper-large-v3-turbo
tags:
- generated_from_trainer
datasets:
- Bingsu/zeroth-korean
metrics:
- wer
model-index:
- name: openai/whisper-large-v3-turbo Korean - Fine-tuned
results:
- task:
name: Automatic Speech Recognition
type: automatic-speech-recognition
dataset:
name: Bingsu/zeroth-korean
type: Bingsu/zeroth-korean
args: 'transcription column: text'
metrics:
- name: Wer
type: wer
value: 4.321638307483813
---
> ⚠️ **Author's Note**: This model was fine-tuned for sanity-checking purposes using only a single Korean dataset.
> As a result, it may be overfitted and may not generalize well to other datasets.
> You can find the training code and related resources in my GitHub repository: [2025-korean-asr-benchmark](https://github.com/baeseongsu/2025-korean-asr-benchmark)
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->
# openai/whisper-large-v3-turbo Korean - Fine-tuned
This model is a fine-tuned version of [openai/whisper-large-v3-turbo](https://huggingface.co/openai/whisper-large-v3-turbo) on the Bingsu/zeroth-korean dataset.
It achieves the following results on the evaluation set:
- Loss: 0.0733
- Wer: 4.3216
## 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: 128
- eval_batch_size: 8
- seed: 42
- optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- num_epochs: 5
- mixed_precision_training: Native AMP
### Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|:-------------:|:------:|:----:|:---------------:|:-------:|
| 0.4622 | 0.1437 | 25 | 0.3690 | 19.2290 |
| 0.2 | 0.2874 | 50 | 0.1578 | 15.3290 |
| 0.1213 | 0.4310 | 75 | 0.1396 | 13.0703 |
| 0.1068 | 0.5747 | 100 | 0.1314 | 12.2572 |
| 0.1 | 0.7184 | 125 | 0.1242 | 11.0676 |
| 0.0922 | 0.8621 | 150 | 0.1181 | 10.6460 |
| 0.0895 | 1.0057 | 175 | 0.1122 | 9.6371 |
| 0.0667 | 1.1494 | 200 | 0.1098 | 9.2155 |
| 0.0608 | 1.2931 | 225 | 0.1049 | 8.4023 |
| 0.0608 | 1.4368 | 250 | 0.1007 | 7.6946 |
| 0.0577 | 1.5805 | 275 | 0.0992 | 7.4386 |
| 0.0591 | 1.7241 | 300 | 0.0953 | 6.5502 |
| 0.0547 | 1.8678 | 325 | 0.0920 | 5.9630 |
| 0.0518 | 2.0115 | 350 | 0.0885 | 5.5112 |
| 0.0299 | 2.1552 | 375 | 0.0878 | 5.8877 |
| 0.0311 | 2.2989 | 400 | 0.0872 | 4.8637 |
| 0.0319 | 2.4425 | 425 | 0.0895 | 5.2552 |
| 0.0363 | 2.5862 | 450 | 0.0869 | 5.1197 |
| 0.0325 | 2.7299 | 475 | 0.0851 | 4.9390 |
| 0.0331 | 2.8736 | 500 | 0.0849 | 4.7282 |
| 0.0314 | 3.0172 | 525 | 0.0805 | 4.9240 |
| 0.0196 | 3.1609 | 550 | 0.0805 | 4.5174 |
| 0.0164 | 3.3046 | 575 | 0.0820 | 5.4209 |
| 0.0166 | 3.4483 | 600 | 0.0807 | 6.1135 |
| 0.0153 | 3.5920 | 625 | 0.0775 | 3.9753 |
| 0.0127 | 3.7356 | 650 | 0.0741 | 4.8035 |
| 0.014 | 3.8793 | 675 | 0.0731 | 7.1827 |
| 0.012 | 4.0230 | 700 | 0.0719 | 4.9992 |
| 0.0067 | 4.1667 | 725 | 0.0744 | 4.5475 |
| 0.0061 | 4.3103 | 750 | 0.0732 | 5.2101 |
| 0.0053 | 4.4540 | 775 | 0.0736 | 4.4270 |
| 0.0061 | 4.5977 | 800 | 0.0743 | 4.8938 |
| 0.0048 | 4.7414 | 825 | 0.0740 | 5.3305 |
| 0.0045 | 4.8851 | 850 | 0.0733 | 4.3216 |
### Framework versions
- Transformers 4.50.0.dev0
- Pytorch 2.4.0+cu121
- Datasets 2.21.0
- Tokenizers 0.21.1
|