trainpad repro kit: rebuild any model's step-0 weights from its published seed + config.

pip install torch==2.14.1 --index-url https://download.pytorch.org/whl/cpu && pip install safetensors numpy
python tools/repro.py <mint>                      # coins
python tools/repro.py "lab-aesop#<generation>"   # lab runs
python tools/repro.py mother config-mother.json  # the mother model

The printed step0_sha256 equals the coin's Proofs tab. Weight files are bf16 safetensors; lm_head.weight is tied to transformer.wte.weight.
nanoGPT model.py is vendored at the commit in trainer/vendor/nanogpt/COMMIT (MIT, see its LICENSE).
