JunWorks's picture
Update README.md
6158930 verified
metadata
library_name: transformers
tags: []

Model Card for Model ID

Trained with commonVoice 16.1 Taigi (nan-tw), about 4 ish validated hours
Raw train logs in screenlog.0
CER before fine tuning: 100.45 (practically gibberish)
CER 75.41 :( (understandable, it is trained only on 4 hours for a very niche language)

DEMO: https://79ed50f9b823872678.gradio.live

Model Details

Model Description

This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.

  • Model type: peft lora whisper-base
  • Language(s) (NLP): Taigi (nan-tw)
  • Finetuned from model [optional]: openai/whisper-base

Training results

Training Loss Epoch Validation Loss
2.0245 1.0 2.0332
1.7279 2.0 1.8450
1.6811 3.0 1.7509
1.6153 4.0 1.6890
1.5804 5.0 1.6443
1.5382 6.0 1.6113
1.5547 7.0 1.5843
1.5229 8.0 1.5626
1.4683 9.0 1.5455
1.4458 10.0 1.5330
1.4628 11.0 1.5210
1.4278 12.0 1.5136
1.4231 13.0 1.5046
1.3957 14.0 1.4997
1.4187 15.0 1.4950
1.4219 16.0 1.4936
1.5940 17.0 1.49281