WhisperForSpokenNER
This model is a fine-tuned version of openai/whisper-large-v2 on the facebook/voxpopuli de+es+fr+nl dataset. It achieves the following results on the evaluation set:
- Loss: 0.4253
- F1 Score: 0.7984
- Label F1: 0.8971
- Wer: 0.0599
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: 0.0001
- train_batch_size: 64
- eval_batch_size: 32
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 128
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 500
- training_steps: 5000
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | F1 Score | Label F1 | Wer |
---|---|---|---|---|---|---|
0.4435 | 0.36 | 200 | 0.4357 | 0.4513 | 0.7168 | 0.0599 |
0.4309 | 0.71 | 400 | 0.4306 | 0.6751 | 0.8354 | 0.0599 |
0.4235 | 1.07 | 600 | 0.4282 | 0.6722 | 0.8548 | 0.0599 |
0.4267 | 1.43 | 800 | 0.4269 | 0.7073 | 0.8455 | 0.0599 |
0.4254 | 1.79 | 1000 | 0.4264 | 0.7273 | 0.8678 | 0.0599 |
0.4264 | 2.14 | 1200 | 0.4264 | 0.7398 | 0.8780 | 0.0599 |
0.4206 | 2.5 | 1400 | 0.4262 | 0.7206 | 0.8583 | 0.0599 |
0.4232 | 2.86 | 1600 | 0.4260 | 0.7410 | 0.8685 | 0.0599 |
0.4249 | 3.22 | 1800 | 0.4255 | 0.7603 | 0.8926 | 0.0599 |
0.4239 | 3.57 | 2000 | 0.4256 | 0.7631 | 0.8835 | 0.0599 |
0.4213 | 3.93 | 2200 | 0.4255 | 0.7692 | 0.8988 | 0.0599 |
0.4213 | 4.29 | 2400 | 0.4256 | 0.7769 | 0.8926 | 0.0599 |
0.4244 | 4.65 | 2600 | 0.4253 | 0.7711 | 0.8996 | 0.0599 |
0.4234 | 5.0 | 2800 | 0.4254 | 0.7386 | 0.8797 | 0.0599 |
0.4222 | 5.36 | 3000 | 0.4252 | 0.7917 | 0.9 | 0.0599 |
0.4239 | 5.72 | 3200 | 0.4254 | 0.7801 | 0.8963 | 0.0599 |
0.4201 | 6.08 | 3400 | 0.4254 | 0.7950 | 0.8954 | 0.0599 |
0.4194 | 6.43 | 3600 | 0.4253 | 0.7851 | 0.9008 | 0.0599 |
0.4203 | 6.79 | 3800 | 0.4252 | 0.7934 | 0.9091 | 0.0599 |
0.4214 | 7.15 | 4000 | 0.4253 | 0.8050 | 0.9046 | 0.0599 |
0.4206 | 7.51 | 4200 | 0.4253 | 0.8 | 0.9 | 0.0599 |
0.4205 | 7.86 | 4400 | 0.4253 | 0.8050 | 0.9129 | 0.0599 |
0.4207 | 8.22 | 4600 | 0.4253 | 0.7951 | 0.9016 | 0.0599 |
0.4218 | 8.58 | 4800 | 0.4253 | 0.7984 | 0.8971 | 0.0599 |
0.4201 | 8.94 | 5000 | 0.4253 | 0.7984 | 0.8971 | 0.0599 |
Framework versions
- Transformers 4.37.0.dev0
- Pytorch 2.1.0
- Datasets 2.14.6
- Tokenizers 0.14.1
- Downloads last month
- 14
Model tree for qmeeus/whisper-large-multilingual-spoken-ner-pipeline-step-1
Base model
openai/whisper-large-v2Dataset used to train qmeeus/whisper-large-multilingual-spoken-ner-pipeline-step-1
Evaluation results
- Wer on facebook/voxpopuli de+es+fr+nlself-reported0.060