/NanoPeft

The simplest repository & Neat implementation of different Lora methods for training/fine-tuning Transformer-based models (i.e., BERT, GPTs). [ Research purpose ]

Primary LanguageJupyter Notebook

NanoPeft

The simplest repository & Neat implementation of different Lora methods for training/fine-tuning Transformer-based models (i.e., BERT, GPTs).

Why NanoPeft?

  • Hacking the Hugging Face PEFT (Parameter-Efficient Fine-Tuning) or LitGit packages seems like a lot of work to integrate a new LoRA method quickly and benchmark it.
  • By keeping the code so simple, it is very easy to hack to your needs, add new LoRA methods from papers in the layers/ directory, and fine-tune easily as per your needs.
  • This is mostly for experimental/research purposes, not for scalable solutions.