diff-llm 0.0.0a0

LLM that predicts text diffs.

DiffLLM

Extending the next token prediction setting into "diff prediction".

Setup

Create virtual environment:

conda create -n diff-llm python=3.9
conda activate diff-llm
pip install -r requirements.txt

Export secrets:

export $(grep -v '^#' secrets.txt | xargs)

Export env vars:

export PYTHONPATH=.

Usage

Create dataset

python src/create_dataset.py \
    --output-dir ./dataset \
    --page-names '["Deep learning", "Ancient Greece", "Ted Chiang"]' \
    --n-revisions 10