Quentin Meeus
finetune small model on NER task (slu_weight=1.)
cfee215
metadata
base_model: qmeeus/whisper-small-multilingual-spoken-ner-pipeline-step-1
tags:
  - generated_from_trainer
datasets:
  - facebook/voxpopuli
metrics:
  - wer
model-index:
  - name: WhisperForSpokenNER
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: facebook/voxpopuli de+es+fr+nl
          type: facebook/voxpopuli
          config: de+es+fr+nl
          split: train
        metrics:
          - name: Wer
            type: wer
            value: 0.10856103413576902

WhisperForSpokenNER

This model is a fine-tuned version of /esat/audioslave/qmeeus/exp/whisper_slu/train/whisper-small-spoken-ner on the facebook/voxpopuli de+es+fr+nl dataset. It achieves the following results on the evaluation set:

  • Loss: 0.0444
  • F1 Score: 0.6098
  • Label F1: 0.8369
  • Wer: 0.1086

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: 5e-05
  • train_batch_size: 4
  • eval_batch_size: 4
  • seed: 42
  • gradient_accumulation_steps: 32
  • 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: 100
  • training_steps: 5000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss F1 Score Label F1 Wer
0.0433 0.36 200 0.0523 0.6251 0.8320 0.1043
0.0391 0.71 400 0.0504 0.6207 0.8346 0.1047
0.0381 1.07 600 0.0496 0.6142 0.8322 0.1065
0.0374 1.43 800 0.0484 0.6158 0.8360 0.1071
0.0374 1.79 1000 0.0474 0.6155 0.8370 0.1069
0.0342 2.14 1200 0.0474 0.6118 0.8362 0.1077
0.0362 2.5 1400 0.0468 0.6138 0.8375 0.1079
0.0351 2.86 1600 0.0461 0.6102 0.8361 0.1082
0.0339 3.22 1800 0.0466 0.6111 0.8388 0.1079
0.0323 3.57 2000 0.0467 0.6168 0.8419 0.1088
0.0338 3.93 2200 0.0457 0.6093 0.8426 0.1086
0.032 4.29 2400 0.0452 0.6090 0.8398 0.1085
0.0307 4.65 2600 0.0451 0.6139 0.8422 0.1086
0.0321 5.0 2800 0.0452 0.6116 0.8398 0.1083
0.0313 5.36 3000 0.0448 0.6116 0.8404 0.1092
0.0309 5.72 3200 0.0449 0.6109 0.8402 0.1083
0.0305 6.08 3400 0.0448 0.6086 0.8402 0.1083
0.0301 6.43 3600 0.0447 0.6116 0.8375 0.1081
0.03 6.79 3800 0.0446 0.6103 0.8401 0.1087
0.0302 7.15 4000 0.0445 0.6120 0.8388 0.1084
0.0294 7.51 4200 0.0442 0.6132 0.8396 0.1086
0.03 7.86 4400 0.0444 0.6112 0.8382 0.1088
0.03 8.22 4600 0.0445 0.6109 0.8371 0.1087
0.0296 8.58 4800 0.0444 0.6117 0.8378 0.1084
0.0297 8.94 5000 0.0444 0.6098 0.8369 0.1086

Framework versions

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