Hugging Face's logo

qmeeus
/
whisper-small-multilingual-spoken-ner-end2end-lora

PEFT
Safetensors
whisper
Generated from Trainer
Eval Results
Model card Files Community
whisper-small-multilingual-spoken-ner-end2end-lora / README.md
Quentin Meeus
Finetune E2E NER model for 5000 steps with LoRA
9aefa35
preview code
|
raw
history blame contribute delete
3.32 kB
metadata
license: apache-2.0
library_name: peft
tags:
  - generated_from_trainer
base_model: openai/whisper-small
datasets:
  - facebook/voxpopuli
metrics:
  - wer
model-index:
  - name: WhisperForSpokenNER-end2end
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: facebook/voxpopuli de+es+fr+nl
          type: facebook/voxpopuli
          split: de+es+fr+nl
        metrics:
          - type: wer
            value: 0.38886263390044107
            name: Wer

WhisperForSpokenNER-end2end

This model is a fine-tuned version of openai/whisper-small on the facebook/voxpopuli de+es+fr+nl dataset. It achieves the following results on the evaluation set:

  • Loss: 0.3381
  • Wer: 0.3889

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: 32
  • eval_batch_size: 16
  • seed: 42
  • gradient_accumulation_steps: 4
  • 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 Wer
2.3436 0.36 200 1.8791 0.8871
1.1682 0.71 400 1.0307 0.5048
0.7321 1.07 600 0.6300 0.3665
0.4564 1.43 800 0.4381 0.3515
0.4095 1.79 1000 0.4027 0.3330
0.3813 2.14 1200 0.3847 0.3360
0.3667 2.5 1400 0.3734 0.3392
0.3583 2.86 1600 0.3649 0.3490
0.3454 3.22 1800 0.3588 0.3572
0.3422 3.57 2000 0.3537 0.3705
0.3371 3.93 2200 0.3503 0.3811
0.3291 4.29 2400 0.3475 0.3678
0.324 4.65 2600 0.3451 0.3670
0.3262 5.0 2800 0.3431 0.3710
0.3168 5.36 3000 0.3419 0.3847
0.3178 5.72 3200 0.3406 0.3833
0.3136 6.08 3400 0.3400 0.3853
0.3092 6.43 3600 0.3393 0.3896
0.3106 6.79 3800 0.3389 0.3900
0.3057 7.15 4000 0.3388 0.3803
0.3087 7.51 4200 0.3383 0.3941
0.308 7.86 4400 0.3382 0.3874
0.3036 8.22 4600 0.3381 0.3896
0.3087 8.58 4800 0.3380 0.3910
0.3079 8.94 5000 0.3381 0.3889

Framework versions

  • PEFT 0.7.1.dev0
  • Transformers 4.37.0.dev0
  • Pytorch 2.1.0
  • Datasets 2.14.6
  • Tokenizers 0.14.1