Welcome to the coolest corner of the AI universe! π This project isn't just about fine-tuning a language model; it's about giving Gemma a fabulous makeover using the power of Low-Rank Adaptation (LoRA). Get ready to witness a transformation that would make any AI fashionista proud!
- Clone this cosmic repository:
git clone https://github.com/utkarshpophli/gemma-finetuning-keras-lora.git
cd gemma-finetuning-lora
- Suit up with the right gear:
pip install -r requirements.txt
- Ignite the training sequence:
python src/train.py
- Launch the Streamlit spaceship:
streamlit run app.py
src/
: The engine room where all the AI magic happensapp.py
: Your control panel to interact with the fabulous fine-tuned Gemmaconfig.py
: The blueprint for our AI makeoverlogger.py
: Our cosmic diary keeper
- Gemma: Our talented AI star, ready for her big performance
- LoRA: The magical stylist giving Gemma her glow-up
- You: The brilliant director orchestrating this AI masterpiece
This project is like giving Gemma a starring role in her own blockbuster movie. We start with her raw talent (pre-trained model), send her to LoRA's beauty school (fine-tuning), and then watch her dazzle the audience (generate amazing responses) in her new role!
- π§ Brain Boost: Gemma now understands context like a champ
- π Speed Demon: Responses faster than a caffeinated coder
- π Adaptation Master: Gemma can now switch roles quicker than a chameleon
Feel free to tweak the config.py
file to experiment with different:
- π LoRA ranks (how dramatic should Gemma's makeover be?)
- π Sequence lengths (how long-winded should we allow Gemma to be?)
- ποΈ Batch sizes (how many scripts should Gemma memorize at once?)
- If Gemma were a coffee, she'd be a triple shot espresso after this fine-tuning
- The number of parameters we're training is smaller than the number of stars you can see on a clear night (but the results are just as magical)
This project is licensed under the MIT License - see the LICENSE file for details. (Spoiler: You're free to use it, just don't blame us if Gemma becomes too sassy)
- Shoutout to the Keras team for making AI as easy as pie (a very complex, multi-layered pie)
- High-five to the LoRA inventors for showing us that sometimes, less really is more
- Virtual hugs to the open-source community for being awesome
Now go forth and make Gemma shine brighter than a supernova! πβ¨