This will be my playground to explore how to recover the LLM prompt that was used to transform a given text.
mamba activate kllmpr
Today I entered the Kaggle LLM Prompt Recovery competition, and I want to have a local playground for this. I have created a new LLM-Prompt-Recovery repo for this.
- mamba create -n kllmpr python=3.11
- mamba activate kllmpr
- mamba install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia
- mamba install conda-forge::transformers
- mamba install conda-forge::jupyterlab
- mamba install conda-forge::ipywidgets
- mamba install conda-forge::accelerate
- mamba install conda-forge::bitsandbytes
mamba activate kllmpr
Continue digging into the workings of gemma.
Keep at it ...
Attempting to run the Kaggle notebook Prompt Recovery with Gemma - KerasNLP Starter locally ... This notebook use Keras and Jax, so yeah, gonna I need to install those libraries.
Hmmm just to be safe, I am going to clone the kllmpr environment to kllmpr-2 and run the installs in that environment.
So I didn't clone the environment yesterday, but will do it now and continue ... working through the notebook 'Local_prompt-recovery-with-gemma-kerasnlp-starter.ipynb'
- mamba create --name kllmpr-2 --clone kllmpr
- mamba activate kllmpr-2
- mamba install conda-forge::keras
- pip install --upgrade "jax[cuda11_pip]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
- pip install --upgrade keras-cv
- pip install --upgrade keras-nlp
- pip install --upgrade keras
- mamba remove conda-forge::keras ... cuz keras was broken ...
- pip install --upgrade keras
- mamba install conda-forge::plotly