képzési információ
A modell, egy újragondolt adatbázissal került kiképzésre.
Az adatbázisból ki lettek véve:
- a numerikus számok, ezért a modell az elhangzott számokat szövegesen fogja leírni
- speciális karakterek, ezért ezeket is fonetikusan fogja leírni
- mozaikszavak
- nagybetűk
Ezek miatt a változtatások miatt a WER elszállt kicsit, viszont a normalizált WER, tovább javult. A hipernormalizált WER vélhetően mégjobb lenne (ahhol a tesztataok is át lennének javítva a fentiek szerint).
A képzés ezesetben a transformer könyvtár mintascriptjével történt: https://github.com/huggingface/transformers/tree/main/examples/pytorch/speech-recognition#whisper-model egyedi 2000 órás adatkészleten, ami most a CV17 train+validate spliteket is tartalmazta.
whisper-base-hu-V2
This model is a fine-tuned version of openai/whisper-base on an unknown dataset. It achieves the following results on the evaluation set:
- Loss: 0.0880
- Wer: 0.0960
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: 7e-05
- train_batch_size: 32
- eval_batch_size: 32
- seed: 42
- distributed_type: multi-GPU
- num_devices: 2
- gradient_accumulation_steps: 2
- total_train_batch_size: 128
- total_eval_batch_size: 64
- optimizer: Use OptimizerNames.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: 1000
- num_epochs: 3.0
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.551 | 0.0904 | 1000 | 0.2710 | 0.2694 |
| 0.4016 | 0.1807 | 2000 | 0.2009 | 0.2061 |
| 0.3449 | 0.2711 | 3000 | 0.1707 | 0.1770 |
| 0.3147 | 0.3614 | 4000 | 0.1588 | 0.1650 |
| 0.2936 | 0.4518 | 5000 | 0.1472 | 0.1551 |
| 0.2758 | 0.5421 | 6000 | 0.1406 | 0.1479 |
| 0.2663 | 0.6325 | 7000 | 0.1322 | 0.1393 |
| 0.2613 | 0.7228 | 8000 | 0.1283 | 0.1402 |
| 0.2491 | 0.8132 | 9000 | 0.1216 | 0.1319 |
| 0.238 | 0.9035 | 10000 | 0.1192 | 0.1291 |
| 0.2287 | 0.9939 | 11000 | 0.1151 | 0.1276 |
| 0.1798 | 1.0842 | 12000 | 0.1131 | 0.1234 |
| 0.1791 | 1.1746 | 13000 | 0.1113 | 0.1186 |
| 0.1787 | 1.2649 | 14000 | 0.1085 | 0.1186 |
| 0.1771 | 1.3553 | 15000 | 0.1068 | 0.1154 |
| 0.1728 | 1.4456 | 16000 | 0.1046 | 0.1135 |
| 0.1714 | 1.5360 | 17000 | 0.1029 | 0.1152 |
| 0.1706 | 1.6263 | 18000 | 0.1007 | 0.1117 |
| 0.163 | 1.7167 | 19000 | 0.0998 | 0.1074 |
| 0.1613 | 1.8070 | 20000 | 0.0982 | 0.1075 |
| 0.1568 | 1.8974 | 21000 | 0.0967 | 0.1087 |
| 0.1525 | 1.9878 | 22000 | 0.0945 | 0.1045 |
| 0.1063 | 2.0781 | 23000 | 0.0967 | 0.1046 |
| 0.1075 | 2.1684 | 24000 | 0.0951 | 0.1030 |
| 0.1035 | 2.2588 | 25000 | 0.0936 | 0.1015 |
| 0.1056 | 2.3491 | 26000 | 0.0928 | 0.1013 |
| 0.1019 | 2.4395 | 27000 | 0.0921 | 0.1000 |
| 0.1004 | 2.5298 | 28000 | 0.0911 | 0.0986 |
| 0.0992 | 2.6202 | 29000 | 0.0904 | 0.0980 |
| 0.1011 | 2.7105 | 30000 | 0.0898 | 0.0978 |
| 0.095 | 2.8009 | 31000 | 0.0892 | 0.0975 |
| 0.0975 | 2.8913 | 32000 | 0.0885 | 0.0960 |
| 0.0963 | 2.9816 | 33000 | 0.0880 | 0.0962 |
Framework versions
- Transformers 4.48.0.dev0
- Pytorch 2.5.1+cu124
- Datasets 3.2.0
- Tokenizers 0.21.0