Unfathomable?

Description:

     Stumble upon a fine tuning that is unfathomable?

Purposes:

     I just happened to use a PEFT fine tuned model from my previous "First Fine Tuning LLM" project to infer this question "What is the capital of USA?" Surprisingly, the model returned correct answer. So, I tried it on the full fine turning model from the same project, but it did not give the right answer. The same was true for the base model.

I then put together a small test dataset to run through three models. The results are:

PEFT model got %77

Full fine tuning model got %50 (the Washington or Washington, D. C. )

Base model got %27

File Descriptions

• flant5_full.ipynb: A simple full fine-tuning for the Flan-T5 model lab in Jupyter Notebook

• flant5_peft.ipynb: A simple PEFT for the Flan-T5 model lab in Jupyter Notebook

• testing_full.ipynb: Inferring Q/A test for full fine-tuning model in Jupyter Notebook.

• testing_peft.ipynb: Inferring Q/A test for PEFT model in Jupyter Notebook.

• testing_base.ipynb: Inferring Q/A test for Flan-T5 base model in Jupyter Notebook.

Contributors

     The scripts are based on the online class "Generative AI with Large Language Models" from Coursera.

     ChatGPT-3.5 the coding machine!

Project Attribution

Model from Hugging Face: 
- google/flan-t5-base

Disclaimer

     This project is provided "as is" and without any warranty. Use it at your own risk.