Scripts for the Lean formal2formal (tactic prediction) experiments. Adapted from llemma_formal2formal.
Install Python packages:
bash scripts/prepare_env.sh
Install Lean:
# from https://leanprover-community.github.io/install/linux.html
# After running this command, select (2), then `nightly`, then `y`:
curl https://raw.githubusercontent.com/leanprover/elan/master/elan-init.sh -sSf | sh
source $HOME/.elan/env
lake
Configure LeanDojo:
export CONTAINER="native"
cd gpt-fast
bash scripts_intern/prepare_intern_math_7b.sh
bash scripts_intern/finetune_7b_intern.sh
bash scripts_intern/finetune_7b_cot.sh
bash scripts_intern/finetune_7b_star.sh
cd gpt-fast
bash scripts_intern/inverse_intern_math_7b.sh
bash scripts_intern/sample_cot_7b.sh
The code is partially adopted from gpt-accelera.