Skip to content

Latest commit

 

History

History
10 lines (9 loc) · 768 Bytes

README.md

File metadata and controls

10 lines (9 loc) · 768 Bytes
Raw

finetuned T5 detoxifier

  • we are using T5-base as our base model.
  • finetuned on paradetox dataset from huggingface.
  • This finetuned model is available to download from Huggingface : link1 or link2
  • trained on kaggle environment took 5 hours to train the model and achieved good results.
  • This is instruction based finetuning not applying PEFT here.
  • during data preparation adding instructions as prefix is a good practice.
    • example: input: "Toxic version: i didnt vote for the liar" , output: "Non-toxic version: I didn't vote for him"