metadata
language:
- ga
- en
license: apache-2.0
base_model: openai/whisper-small
tags:
- generated_from_trainer
datasets:
- ymoslem/IWSLT2023-GA-EN
- ymoslem/FLEURS-GA-EN
- ymoslem/BitesizeIrish-GA-EN
- ymoslem/SpokenWords-GA-EN-MTed
metrics:
- bleu
- wer
- chrf
model-index:
- name: Whisper Small GA-EN Speech Translation
results:
- task:
name: Automatic Speech Recognition
type: automatic-speech-recognition
dataset:
name: IWSLT-2023, FLEURS, BiteSize, and SpokenWords
type: ymoslem/IWSLT2023-GA-EN
metrics:
- name: Bleu
type: bleu
value: 27.66
- name: Wer
type: wer
value: 72.0396217919856
library_name: transformers
Whisper Small GA-EN Speech Translation
This model is a fine-tuned version of openai/whisper-small on the IWSLT-2023, FLEURS, BiteSize, and SpokenWords datasets. The best model checkpoint (this version) based on ChrF is at step 2100, epoch 4.5259, and it achieves the following results on the evaluation set:
- Loss: 1.7200
- Bleu: 29.83
- Chrf: 44.87
- Wer: 64.8807
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
Training: IWSLT-2023 (train+dev), FLEURS, BiteSize, and SpokenWords Evaluation: IWSLT-2023 (test)
Training procedure
Hardware:
1 NVIDIA A100-SXM4-80GB
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0001
- train_batch_size: 64
- eval_batch_size: 64
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 0
- training_steps: 3000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Bleu | Chrf | Wer |
|---|---|---|---|---|---|---|
| 1.9416 | 0.2155 | 100 | 1.7899 | 13.09 | 26.48 | 104.4575 |
| 1.5186 | 0.4310 | 200 | 1.5696 | 18.6 | 35.75 | 87.5732 |
| 1.2884 | 0.6466 | 300 | 1.4751 | 17.57 | 37.19 | 87.2580 |
| 1.0729 | 0.8621 | 400 | 1.4345 | 17.92 | 38.23 | 99.2346 |
| 0.4574 | 1.0776 | 500 | 1.5585 | 22.48 | 39.17 | 83.1607 |
| 0.4517 | 1.2931 | 600 | 1.5763 | 22.53 | 38.38 | 81.7650 |
| 0.4385 | 1.5086 | 700 | 1.5852 | 20.05 | 39.46 | 96.8483 |
| 0.3934 | 1.7241 | 800 | 1.5332 | 26.89 | 42.67 | 70.6889 |
| 0.3587 | 1.9397 | 900 | 1.5025 | 28.95 | 44.16 | 64.9707 |
| 0.1528 | 2.1552 | 1000 | 1.5882 | 28.32 | 42.36 | 65.8712 |
| 0.1425 | 2.3707 | 1100 | 1.6056 | 25.5 | 42.42 | 75.0113 |
| 0.1389 | 2.5862 | 1200 | 1.6236 | 26.52 | 42.11 | 70.6439 |
| 0.1532 | 2.8017 | 1300 | 1.6196 | 25.78 | 41.61 | 75.9118 |
| 0.1138 | 3.0172 | 1400 | 1.7185 | 26.01 | 40.88 | 69.6983 |
| 0.0661 | 3.2328 | 1500 | 1.6626 | 28.74 | 43.16 | 71.2292 |
| 0.0625 | 3.4483 | 1600 | 1.6835 | 29.16 | 43.6 | 66.3215 |
| 0.0615 | 3.6638 | 1700 | 1.6756 | 28.93 | 44.08 | 68.3476 |
| 0.0611 | 3.8793 | 1800 | 1.6648 | 27.77 | 43.67 | 72.1747 |
| 0.0344 | 4.0948 | 1900 | 1.7351 | 28.33 | 44.18 | 68.1225 |
| 0.0339 | 4.3103 | 2000 | 1.7715 | 28.9 | 42.98 | 67.0869 |
| 0.0369 | 4.5259 | 2100 | 1.7200 | 29.83 | 44.87 | 64.8807 |
| 0.0326 | 4.7414 | 2200 | 1.7232 | 28.23 | 43.75 | 69.3832 |
| 0.0346 | 4.9569 | 2300 | 1.7688 | 27.72 | 43.1 | 72.8050 |
| 0.0167 | 5.1724 | 2400 | 1.8072 | 28.73 | 43.26 | 67.4471 |
| 0.0146 | 5.3879 | 2500 | 1.7801 | 29.91 | 44.24 | 66.4566 |
| 0.0165 | 5.6034 | 2600 | 1.7782 | 29.34 | 44.33 | 68.2125 |
| 0.0143 | 5.8190 | 2700 | 1.7675 | 27.78 | 43.07 | 72.5799 |
| 0.0106 | 6.0345 | 2800 | 1.7660 | 29.45 | 43.31 | 67.5371 |
| 0.0098 | 6.25 | 2900 | 1.7803 | 27.89 | 42.67 | 71.6344 |
| 0.0087 | 6.4655 | 3000 | 1.7786 | 27.66 | 43.04 | 72.0396 |
Framework versions
- Transformers 4.40.2
- Pytorch 2.2.0+cu121
- Datasets 2.19.1
- Tokenizers 0.19.1