Hugging Face's logo

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

PEFT
Safetensors
whisper_for_slu
Generated from Trainer
Eval Results
Model card Files Community
whisper-small-multilingual-spoken-ner-pipeline-lora / README.md
Quentin Meeus
Finetune NER+ASR module for 5000 steps with LoRA
1e16abf
preview code
|
raw
history blame
4.05 kB
metadata
library_name: peft
tags:
  - generated_from_trainer
base_model: qmeeus/whisper-small-multilingual-spoken-ner-pipeline-step-1
datasets:
  - facebook/voxpopuli
metrics:
  - wer
model-index:
  - name: WhisperForSpokenNER
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: facebook/voxpopuli
          type: facebook/voxpopuli
          split: de+es+fr+nl
        metrics:
          - type: wer
            value: 0.09799520086694016
            name: Wer

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: 0.2264
  • F1 Score: 0.6872
  • Label F1: 0.8325
  • Wer: 0.0980

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: 8
  • 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
0.2746 0.36 200 0.2602 0.6565 0.8343 0.1090
0.2481 0.71 400 0.2465 0.6578 0.8348 0.1022
0.2385 1.07 600 0.2410 0.6685 0.8323 0.1048
0.2316 1.43 800 0.2374 0.6724 0.8317 0.1022
0.2291 1.79 1000 0.2348 0.6698 0.8292 0.0968
0.2205 2.14 1200 0.2334 0.6745 0.8339 0.0964
0.2211 2.5 1400 0.2319 0.6729 0.8327 0.0961
0.2163 2.86 1600 0.2305 0.6732 0.8300 0.0981
0.2108 3.22 1800 0.2299 0.6734 0.8328 0.0954
0.2104 3.57 2000 0.2297 0.6792 0.8369 0.0992
0.2124 3.93 2200 0.2279 0.6782 0.8346 0.0945
0.2027 4.29 2400 0.2279 0.6790 0.8336 0.0944
0.2055 4.65 2600 0.2275 0.6832 0.8347 0.0949
0.209 5.0 2800 0.2269 0.6822 0.8336 0.0983
0.2017 5.36 3000 0.2272 0.6835 0.8346 0.0979
0.2029 5.72 3200 0.2266 0.6819 0.8321 0.0966
0.201 6.08 3400 0.2266 0.6800 0.8327 0.0978
0.1985 6.43 3600 0.2267 0.6856 0.8335 0.0995
0.1996 6.79 3800 0.2265 0.6864 0.8343 0.0988
0.197 7.15 4000 0.2264 0.6843 0.8347 0.0986
0.1985 7.51 4200 0.2264 0.6838 0.8352 0.0985
0.1999 7.86 4400 0.2263 0.6861 0.8346 0.0978
0.1963 8.22 4600 0.2264 0.6864 0.8318 0.0978
0.1977 8.58 4800 0.2264 0.6875 0.8329 0.0979
0.1961 8.94 5000 0.2264 0.6872 0.8325 0.0980

Framework versions

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